Quality, Speed and Consumption

commit 581e7e9built 2026-11-16 21:13 UTC

Overview

Master score

85 / 100

Quality, Speed and Consumption weigh a third each: each is the mean of the tabs scored in it, and the master is the mean of the 3 categories, over 13 measured tabs of 19. A tab is the mean of its categories. 85 of 235 metrics measured; one not measured is shown, never scored as 0.

Quality

73

11 of 19 tabs measured

Speed

94

9 of 19 tabs measured

Consumption

87

7 of 18 tabs measured

235 metrics

65 meet8 near12 miss150 not measured

Work queue

Every metric near or missing its goal, across tabs and Quality, Speed and Consumption, ranked by the master points it adds when it meets its goal.

  1. CodeCode areas drawn in a README0%goal ≥ 100 · +3 points
  2. DocsTokens past the reading goal3,779 tokensgoal ≤ 0 · +2.4 points
  3. DocsDocs over 2 minutes to read6 docsgoal ≤ 0 · +1.9 points
  4. SecurityMerged PRs Claude security-reviewed0%goal ≥ 100 · +1 points
  5. ActionsRuns that find nothing to do66.2%goal ≤ 25 · +1 points
  6. SecurityThreat model current with its code0 on/offgoal ≥ 1 · +1 points
  7. PagesRaw hex outside tokens31 valuesgoal ≤ 0 · +0.6 points
  8. PagesProduction security headers1 of 4goal ≥ 4 · +0.6 points
  9. ActionsTriggers that skip work when nothing changed71.4%goal ≥ 80 · +0.5 points
  10. ActionsRunner list price per billed minute0.45 centsgoal ≤ 0.4 · +0.5 points
  11. DocsHeading and image faults23 faultsgoal ≤ 0 · +0.3 points
  12. DocsNamed repo paths that are missing14 pathsgoal ≤ 0 · +0.3 points
  13. PRsPRs linking an issue0%goal ≥ 100 · +0.3 points
  14. DocsDocs with an interactive element18.8%goal ≥ 80 · +0.3 points
  15. PRsShare over 400 lines30%goal ≤ 10 · +0.3 points
  16. DocsText-only docs6 docsgoal ≤ 0 · +0.3 points
  17. DependabotActions pinned to the latest allowed87.5%goal ≥ 100 · +0.3 points
  18. PagesRaw px outside tokens31 valuesgoal ≤ 0 · +0.3 points
  19. IssuesUnused labels1 labelsgoal ≤ 0 · +0.2 points
  20. IssuesProse words, the latest card81 wordsgoal ≤ 75 · +0.2 points

Scoreboard

Review
Code01 of 20–0 of 41–
Documentation6010 of 15501 of 14–
Issues867 of 91001 of 20–
Pull requests8211 of 111006 of 62–
Discussions–0 of 0–0 of 00–
Actions1006 of 131004 of 61–
Projects–0 of 6–0 of 20–
Security and quality333 of 3–0 of 22–
Dependabot905 of 121004 of 60–
Scanning–0 of 4–0 of 00–
Insights1002 of 31001 of 10–
Settings–0 of 7–0 of 20–
Pages505 of 171003 of 82–
Claude–0 of 41001 of 20–
Claude settings1001 of 1–0 of 10–
Claude desktop app–0 of 6–0 of 20–
Blockers1001 of 41001 of 40–
Tokens–0 of 1–0 of 20–
Instructions–0 of 6–0 of 10–

Not measured

150

Code

Quality

0

1 of 20 measured

Speed

–

0 of 4 measured

Consumption

–

0 of 6 measured

Ideal state

The least code that does the job: every part small, used, named, explained and drawn, fast to run and cheap for an agent to read.

12 sources, newest 2026-11-10
  1. Claude Code Docs: Tools reference 2026-11-10
  2. Claude Docs: Context windows 2026-11-10
  3. Anthropic: Effective context engineering for AI agents 2025-09-29
  4. GitHub Docs: Code Quality metrics and scores reference 2026-11-10
  5. Microsoft Learn: Code metrics, cyclomatic complexity 2026-10-20
  6. Microsoft Learn: Architecture strategies for optimizing code and infrastructure 2025-10-30
  7. PEP 257: Docstring Conventions 2024-04-17
  8. Google Engineering Practices: What to look for in a code review 2026-11-04
  9. Cloudflare Workers limits 2026-10-29
  10. Alex Russell: The Performance Inequality Gap, 2026 2025-11-28
  11. Node.js v26 docs: Modules, CommonJS modules, Cycles 2026-11-10
  12. GitHub Docs: About the repository README file 2026-11-10

Quality

No function past complexity 10 or 50 lines, no file past 300, nothing unused, copied or in an import cycle, every area drawn in a README.

Size–

Code lines change, last 20 PRs–goal ≤ 0

Whether the code grew or shrank over the latest merged pull requests.

area not found: src; update params.areas in quality/targets/code.json

files under src

Goal: Git Crafter default goal: the code holds or shrinks over the last merged PRs, so it stays as small and simple as it can. No outside standard

Files over 300 lines–goal ≤ 0

Files so long that they are hard to read in one sitting, in every language, stylesheets included.

area not found: src; update params.areas in quality/targets/code.json

files under src

Goal: Git Crafter default goal: no file over the limit. The limit, params.limits.file_lines (300), is the default of ESLint's max-lines rule (eslint.org/docs/latest/rules/max-lines)

Functions over 50 lines–goal ≤ 0

Pieces of code too long to understand at a glance.

area not found: src; update params.areas in quality/targets/code.json

files under src

Goal: Git Crafter default goal: no function over the limit. The limit, params.limits.function_lines (50), is the default of ESLint's max-lines-per-function rule (eslint.org/docs/latest/rules/max-lines-per-function)

Functions over complexity 10–goal ≤ 0

Pieces of code with so many branches that they are hard to test and change.

area not found: src; update params.areas in quality/targets/code.json

files under src

Goal: Git Crafter default goal: no function over the limit. The limit, params.limits.complexity (10), is McCabe's (A Complexity Measure, IEEE Transactions on Software Engineering, 1976)

Duplication–

Copied code–goal ≤ 3

How much code is copied and pasted instead of shared. Copies drift apart.

area not found: src; update params.areas in quality/targets/code.json

files under src

Goal: goal: duplication in new code of 3.0% or less, a condition of SonarQube Server 2026.5 docs, Introduction to quality gates, the built-in Sonar way gate (docs.sonarsource.com, read 2026-11-04); the near band is a Git Crafter default

Dead code–

Unused files (knip)–goal ≤ 0

Files nothing uses. They can be deleted.

no package.json at the repo root

knip --reporter json (entries from package scripts, plugins and files run directly)

Goal: Git Crafter default goal: an unused file is code to read and keep that does nothing. Unused or dead code is a maintainability finding (GitHub Docs, Code Quality metrics and scores reference, docs.github.com/en/code-security/reference/code-quality/metrics-and-ratings, read 2026-11-10); zero is a Git Crafter default

Unused exports (knip)–goal ≤ 0

Code offered to other files that no file uses.

no package.json at the repo root

knip --reporter json (entries from package scripts, plugins and files run directly)

Goal: Git Crafter default goal: an export no caller needs is dead code, a maintainability finding (GitHub Docs, Code Quality metrics and scores reference, docs.github.com/en/code-security/reference/code-quality/metrics-and-ratings, read 2026-11-10); zero is a Git Crafter default

Unused dependencies (knip)–goal ≤ 0

Outside packages installed but never used.

no package.json at the repo root

knip --reporter json (entries from package scripts, plugins and files run directly)

Goal: Git Crafter default goal: an unused package costs install time and supply-chain risk for nothing; unused code is a maintainability finding (GitHub Docs, Code Quality metrics and scores reference, docs.github.com/en/code-security/reference/code-quality/metrics-and-ratings, read 2026-11-10); zero is a Git Crafter default

Unused Python functions–goal ≤ 0

Python functions that no code or test in the code folders calls or names. They can be deleted.

area not found: src; update params.areas in quality/targets/code.json

files under src

Goal: Git Crafter default goal: an unused function is code to read and keep that does nothing. Unused or dead code is a maintainability finding (GitHub Docs, Code Quality metrics and scores reference, read 2026-11-10); knip reads no Python, so this metric does

Drift–

TODO, FIXME and XXX markers–goal ≤ 0

Notes about unfinished work left in the code instead of the issue tracker.

area not found: src; update params.areas in quality/targets/code.json

files under src

Goal: Git Crafter default goal: work to do belongs in an issue, where it is tracked, not in a comment

Commented-out code blocks–goal ≤ 0

Old code switched off but left behind. Git already keeps the history.

area not found: src; update params.areas in quality/targets/code.json

files under src

Goal: Git Crafter default goal: git keeps old code, so a commented-out copy only misleads

References to deleted files–goal ≤ 0

Comments and strings that point at files which were deleted or moved.

area not found: src; update params.areas in quality/targets/code.json

files under src

Goal: Git Crafter default goal, our own inference: a reference to a deleted file sends a reader, or Claude, to a dead end. An agent keeps file paths as references and loads each file when it needs it (Anthropic, Effective context engineering for AI agents, anthropic.com/research/effective-context-engineering-for-ai-agents, 2025-09-29, read 2026-11-10)

Comments that help Claude–

Shared functions with a header comment–goal ≥ 90

Functions other files can call that say what they do in a comment or docstring right at their start.

area not found: src; update params.areas in quality/targets/code.json

files under src

Goal: Git Crafter default goal: nearly every shared function says what it does, so Claude can use it without reading its body. Python's own rule is that all functions a module exports have docstrings (PEP 257, peps.python.org/pep-0257, last modified 2024-04-17, read 2026-11-10); the 90% is a Git Crafter default

Files that open with a purpose comment–goal ≥ 100

Code files that start by saying why they exist.

area not found: src; update params.areas in quality/targets/code.json

files under src

Goal: Git Crafter default goal: every file says why it exists, so Claude reads the purpose before the code. No outside standard

Structure0

Code areas drawn in a README0%goal ≥ 100

Code folders that a README, or a doc it links, shows in a table or a diagram, so a reader sees the layout.

Gap: 100 pts under goal

src0%

areas named in a table row or a mermaid block of a README.md at their folder or above, or of a doc it links beside it; 0 of 1 read 2026-11-16 21:12 UTC

Goal: Git Crafter default goal: every code area is drawn where a reader lands. A README tells people about a project, and GitHub renders Mermaid diagrams in Markdown (GitHub Docs, About the repository README file, and Creating diagrams, read 2026-11-10); the share is a Git Crafter default

Files and folders named by convention–goal ≥ 100

Files and folders named the way their neighbors are, so they are easy to find.

git ls-files listed no code file in the areas that a naming rule covers

git ls-files

Goal: Git Crafter default goal: every code file and folder follows its folder's convention. No outside standard for the share; Python's own convention is short, all-lowercase module names (PEP 8, peps.python.org/pep-0008, read 2026-11-10)

Import cycles–goal ≤ 0

Groups of files that import each other, so none of them can be read, tested or changed alone.

area not found: src; update params.areas in quality/targets/code.json

files under src

Goal: Git Crafter default goal: no import cycle. In a cycle a module might not have finished executing when it is returned (Node.js v26 docs, Modules: CommonJS modules, Cycles, read 2026-11-10), and low coupling is what Microsoft's code metrics ask for (Microsoft Learn, Code metrics values, updated 2026-10-20, read 2026-11-10)

Coverage–

Line and branch coverage–goal ≥ 80

The share of code lines and branches the automated tests run, together, close to how SonarQube counts coverage.

params.test_suites names no test suite

params.test_suites

Goal: goal: new code test coverage of 80.0% or more, a condition of SonarQube Server 2026.5 docs, Introduction to quality gates, the built-in Sonar way gate (docs.sonarsource.com, read 2026-11-04). SonarQube coverage mixes lines and conditions, coverage = (CT + LC)/(B + EL) (SonarQube Server 2026.5 docs, Metric definitions, read 2026-11-04), so branches count beside lines. v8 counts true and false branch outcomes apart, where that formula counts true once, so the share approximates SonarQube coverage rather than matching it; the near band is a Git Crafter default

GitHub features–

Code quality–goal ≥ 1

Whether GitHub checks the repo's code for maintainability and reliability problems; on is better, even where it costs.

needs an admin token; refreshed by a --local collect run

gh api repos/Colorado-Startups/brand/code-quality/setup (state)

Goal: GitHub docs, GitHub Code Quality

Open findings–

Open code quality findings–goal ≤ 0

Maintainability and reliability problems GitHub found in the code; not counted while code quality is off.

needs an admin token; refreshed by a --local collect run

gh api repos/Colorado-Startups/brand/code-quality/findings?state=open --paginate (by rule.severity)

Goal: template goal: none open (GitHub docs, GitHub Code Quality)

Speed

Requests served well inside a good time to first byte, and a bundle that builds in seconds.

Live site–

Slowest recent Worker request–goal ≤ 800

How long the Workers took on their slowest recent request since the latest deploy, from start to finish.

params.workers names no Worker folder; this repo deploys no Worker

cf workers deployments list (the first is the one serving traffic), then cf sql query on logs.workersLogs: the last params.last_requests fetch invocations since that deploy, newest first; the slowest and heaviest of them

Goal: goal: web.dev's good time to first byte, 0.8 seconds or less, and poor above 1.8 seconds (web.dev/articles/ttfb, updated 2025-11-28, read 2026-11-04), held by even the slowest recent request; the Worker's own time is one part of it

Worker startup time, latest deploy–goal ≤ 200

How long the deployed Worker takes to load its code before it can answer, as Cloudflare measured it.

params.workers names no Worker folder; this repo deploys no Worker

cf workers get (the Worker's tag), cf builds list (the newest successful Workers Build), cf builds logs get: the "Worker Startup Time" wrangler deploy printed

Goal: Git Crafter default goal, 2026-11-04: a fifth of Cloudflare's startup limit; near is the limit itself, 1 second to parse and run the global scope, past which a deploy is rejected (developers.cloudflare.com/workers/platform/limits, updated 2026-10-29, read 2026-11-04)

Build and test time–

Worker build time–goal ≤ 10

How long the server code takes to bundle.

params.workers names no Worker folder; this repo deploys no Worker

wrangler deploy --dry-run --outdir <tmp> in each folder of params.workers (dry run, never deploys)

Goal: Git Crafter default goal: a quick bundle keeps the push gate quick. No outside standard

Check job wall time–goal ≤ 600

How long the check every change waits on runs.

params.check names no { workflow, job }

latest: wall time of params.check's job in its workflow's newest successful run on the default branch

Goal: Git Crafter default goal: the ten-minute build, "For most projects, however, the XP guideline of a ten minute build is perfectly within reason" (martinfowler.com/articles/continuousIntegration.html, updated 2024-01-18); near is twice it

Consumption

Server CPU inside the free plan limit, page scripts within 307 KiB, each file within one 25,000-token Read, each area within a 200K-token window.

Cost–

Runner minutes, the latest Code Quality scan–goal ≤ 5

Billed runner minutes the latest passed run of GitHub's Code Quality scan took; lower is better

no passed run of GitHub's dynamic CodeQL - Code Quality workflow; Code Quality is off or has not run

latest: gh api actions/runs/{id}/jobs: GitHub's dynamic CodeQL - Code Quality workflow, its latest passed run, read with the security scans; billed runner minutes, summed

Goal: template goal (2026-11-13): the Code Quality scan on one change bills at most 5 runner minutes, as the security scans do, half the ten-minute build (Kent Beck, Extreme Programming Explained, 2nd edition, 2004)

Live site–

Heaviest recent Worker request, CPU–goal ≤ 10

Server work time of the heaviest recent request since the latest deploy.

params.workers names no Worker folder; this repo deploys no Worker

cf workers deployments list (the first is the one serving traffic), then cf sql query on logs.workersLogs: the last params.last_requests fetch invocations since that deploy, newest first; the slowest and heaviest of them

Goal: goal: Workers Free's CPU limit of 10 ms per request, which every request must stay under, the heaviest included (developers.cloudflare.com/workers/platform/limits, updated 2026-10-29, read 2026-11-04); the near band is a Git Crafter default

Download size–

Largest Worker bundle, gzipped–goal ≤ 1,000

Size of the server code Cloudflare loads. Smaller starts faster.

params.workers names no Worker folder; this repo deploys no Worker

wrangler deploy --dry-run --outdir <tmp> in each folder of params.workers (dry run, never deploys)

Goal: Git Crafter default goal. No outside standard: Cloudflare sets no compressed size limit and caps a Worker at 64 MiB uncompressed (developers.cloudflare.com/workers/platform/limits, read 2026-11-03)

Client JS per page, gzipped–goal ≤ 307

Script a visitor's browser downloads on the heaviest page.

no site build in dist; run the build first

dist

Goal: goal: 0.3 MiB (307 KiB) of JavaScript, the JS share of a JS-light page's critical-path budget for a 3 s load on the 75th-percentile phone and network; near is 0.62 MiB (635 KiB), the JS-heavy page's share (Alex Russell, The Performance Inequality Gap, 2026 (infrequently.org, published 2025-11-24, updated 2025-11-28, read 2026-11-04))

Size–

Code a session reads, largest area, in tokens–goal ≤ 200K

The largest area's code, tests included, as tokens a session loads to read that area whole.

area not found: src; update params.areas in quality/targets/code.json

state: bytes of the code files in the largest area of params.areas, tests included, over params.bytes_per_token

Goal: Git Crafter default goal: one area fits a 200k-token context window, the smallest Claude still serves; near is the 1M largest, and recall degrades as context grows (platform.claude.com/docs/en/build-with-claude/context-windows, read 2026-11-03). An agent keeps lightweight references and loads files just in time with its tools, so a session reads the area it works in (Anthropic, Effective context engineering for AI agents, anthropic.com/research/effective-context-engineering-for-ai-agents, 2025-09-29, read 2026-11-04)

Largest code file, in tokens–goal ≤ 25,000

The biggest code file, tests included, as tokens a session loads to read it whole.

area not found: src; update params.areas in quality/targets/code.json

state: bytes of the code files in the largest area of params.areas, tests included, over params.bytes_per_token

Goal: goal: one Read call returns at most 25,000 tokens by default, so a larger file is read a page at a time; near is 111,500 tokens, the 256 KB past which Read refuses a whole file, at params.bytes_per_token (2.35) bytes a token (Claude Code Docs, Tools reference, code.claude.com/docs/en/tools-reference, read 2026-11-10)

Documentation

Quality

60

10 of 15 measured

Speed

50

1 of 1 measured

Consumption

50

1 of 1 measured

Ideal state

Every fact a reader needs is present, correct and current, in short, visual, accessible docs a person finishes in two minutes and Claude loads cheaply.

15 sources, newest 2026-11-10
  1. GitHub Docs: About the repository README file 2026-11-10
  2. GitHub Docs: About community profiles for public repositories 2026-11-10
  3. GitHub Docs: Adding support resources to your project 2026-11-10
  4. GitHub Docs: Creating a default community health file 2026-11-10
  5. GitHub Docs: Style guide 2026-11-10
  6. GitHub Docs: Creating diagrams 2026-11-10
  7. GitHub Docs: Organizing information with collapsed sections 2026-11-10
  8. Microsoft Style Guide: Top 10 tips for Microsoft style and voice 2026-07-02
  9. Microsoft Style Guide: Scannable content 2023-06-20
  10. Microsoft Style Guide: Tables 2026-11-10
  11. Microsoft Style Guide: Alternative text 2026-04-03
  12. Microsoft Engineering Fundamentals Playbook: Documentation 2026-11-10
  13. Claude Code Docs: How Claude remembers your project 2026-11-10
  14. Claude Docs: Skill authoring best practices 2026-11-10
  15. Brysbaert, 2019: How many words do we read per minute? (Journal of Memory and Language 109) 2026-11-03

Quality

Paths, scripts and links resolve, nothing is stale, folders and community files are covered, headings and images are accessible, and docs are visual first.

Correct33

Named repo paths that are missing14 pathsgoal ≤ 0

File and folder paths a doc writes in code that no longer exist in the repo

Gap: 14 paths over goal

repo paths checked359 paths
CHANGELOG.md:59 dist/hall/1 paths
CHANGELOG.md:85 dist/hall/guides.jpg1 paths
CHANGELOG.md:127 dist/hall/1 paths
CHANGELOG.md:137 guide/public/tokens.css1 paths
CHANGELOG.md:179 dist/social/brand.png1 paths
CHANGELOG.md:179 dist/social/coloradostartups.org.png1 paths
CHANGELOG.md:186 dist/social/brand.png1 paths
CHANGELOG.md:186 dist/social/coloradostartups.org.png1 paths
README.md:20 tools/brand_pull.py1 paths
art/direction/09-the-keepers.md:19 art/concepts/batch-1.md1 paths
art/direction/11-renders-and-runs.md:46 art/concepts/batch-1.md1 paths

12 of 15 shown

each word of inline code and code blocks under a top-level repo entry, git-ignored paths and records skipped (a heuristic), quality/collectors/docs/docs.mjs over 49 docs (params.doc_set) read 2026-11-16 21:12 UTC

Goal: Git Crafter default goal: every path a doc names exists. Source: Microsoft Engineering Fundamentals Playbook: Documentation (an inaccurate doc names the wrong folders and settings)

Heading and image faults23 faultsgoal ≤ 0

Headings that skip a level or repeat, and images a screen reader cannot describe

Gap: 23 faults over goal

headings and images checked156 faults
CHANGELOG.md:26 repeated heading "Added"1 faults
CHANGELOG.md:35 repeated heading "Added"1 faults
CHANGELOG.md:44 repeated heading "Added"1 faults
CHANGELOG.md:49 repeated heading "Changed"1 faults
CHANGELOG.md:58 repeated heading "Changed"1 faults
CHANGELOG.md:72 repeated heading "Added"1 faults
CHANGELOG.md:76 repeated heading "Changed"1 faults
CHANGELOG.md:84 repeated heading "Added"1 faults
CHANGELOG.md:95 repeated heading "Added"1 faults
CHANGELOG.md:99 repeated heading "Changed"1 faults
CHANGELOG.md:107 repeated heading "Added"1 faults

12 of 24 shown

156 headings (a skipped level, a second H1, a heading repeated at its level) and 0 images (no alt text, or over 150 characters) outside code, quality/collectors/docs/docs.mjs over 49 docs (params.doc_set) read 2026-11-16 21:12 UTC

Goal: Git Crafter default goal: headings run in order with no repeat, and every image has alt text within the limit. Sources: GitHub Docs: Style guide (no header level is skipped, and each header at a level is unique); Microsoft Style Guide: Alternative text (limit the length to 150 characters)

Named npm scripts that are missing–goal ≤ 0

npm commands in a doc that run a script no package.json in the repo defines

package.json at the repo root is missing or unreadable (ENOENT: no such file or directory, open '/home/runner/work/brand/brand/package.json')

package.json

Goal: Git Crafter default goal: every npm script a doc names runs. Source: Microsoft Engineering Fundamentals Playbook: Documentation (inaccurate docs)

Content-lint WARN in docs–goal ≤ 0

Likely breaks of the repo's own writing rules, in the docs those rules cover

quality/targets/docs.json has no params.lint { interpreter, script, args, scope }; this repo names no content lint

quality/targets/docs.json

Goal: Git Crafter default goal: no content-lint warning in the docs the lint covers. Source: Microsoft Style Guide: Top 10 tips for Microsoft style and voice (prune every excess word)

Current100

Stale markers in docs0 markersgoal ≤ 0

Words that flag unfinished text, such as TODO or not yet, outside code

TODO, TBD, FIXME, not yet, coming soon outside code and comments, quality/collectors/docs/docs.mjs over 49 docs (params.doc_set) read 2026-11-16 21:12 UTC

Goal: Git Crafter default goal: no doc says its own text is unfinished. Source: Microsoft Engineering Fundamentals Playbook: Documentation (incomplete docs)

Dated claims older than their area0 datesgoal ≤ 0

Current as of and last verified dates set before the files the doc covers last changed

0 "current as of", "last verified" dates outside code, each against the last commit to its doc's area (params.areas), quality/collectors/docs/docs.mjs over 49 docs (params.doc_set) read 2026-11-16 21:12 UTC

Goal: Git Crafter default goal: a dated claim is at least as new as what it describes. Source: Microsoft Engineering Fundamentals Playbook: Documentation (obsolete docs)

Oldest doc lag behind its area2 daysgoal ≤ 7

Days the doc that trails its subject most has gone without catching up to it

SECURITY.md2 days
THREAT-MODEL.md2 days
AGENTS.md1.9 days
owners.md‎governance/‎1.9 days
README.md‎governance/‎1.9 days
releases.md‎governance/‎1.9 days
02-release-steps.md‎governance/releases/‎1.9 days
03-the-look.md‎art/direction/‎1.4 days
04-characters-and-cards.md‎art/direction/‎1.4 days
05-palette-by-token-role.md‎art/direction/‎1.4 days
09-the-keepers.md‎art/direction/‎1.4 days
10-prompts-fibo-2-0-in-vgl.md‎art/direction/‎1.4 days

12 of 49 shown

git log -1 --format=%ct -- <doc>, and -- <area> (params.areas); the doc that trails its area most, SECURITY.md read 2026-11-16 21:12 UTC

Goal: Git Crafter default goal, 2026-11-04: no doc trails the files it covers by more than a week; a point-in-time value, the one doc to fix, never a mean over docs. Source: Microsoft Engineering Fundamentals Playbook: Documentation (docs not updated along with the code)

Unmapped docs and dead area paths0 docs and pathsgoal ≤ 0

Docs with no usable params.areas entry, and area paths that match no other tracked file: the lag checks skip both

each doc against params.areas, and each area path through git ls-files, quality/collectors/docs/docs.mjs over 49 docs (params.doc_set) read 2026-11-16 21:12 UTC

Goal: Git Crafter default goal, 2026-11-04: none, since the doc lag and dated-claim checks skip an unmapped doc or a dead area path without a word; above the goal, the pages action fails a pull request. Source: Microsoft Engineering Fundamentals Playbook: Documentation (docs not updated along with the code)

Figures with no recent read date–goal ≤ 0

Numbers in living docs whose sentence has no read date, or one older than a month.

rules and memory live on the machine that ran the sessions; refreshed by a --local collect there

quality/collectors/simplicity/simplicity.mjs

Goal: template goal, no public benchmark: a figure is attributed or labeled a goal (Git Crafter rules/writing-and-gates.md); one with no read date cannot be checked; near band: template default

Complete100

Community health files present100%goal ≥ 100

The files GitHub lists for a healthy project, such as a code of conduct and a support guide

Code of conduct (CODE_OF_CONDUCT.md)1
Contributing guide (CONTRIBUTING.md)1
License (LICENSE)1
Pull request template (.github/pull_request_template.md)1
README (README.md)1
Support file (SUPPORT.md)1

each of 6 community health files (params.community_files) at any of its paths read 2026-11-16 21:12 UTC

Goal: Git Crafter default goal: every community health file is there. Sources: GitHub Docs: About community profiles for public repositories (README, code of conduct, license, contributing); GitHub Docs: Adding support resources to your project; GitHub Docs: Creating a default community health file (pull request template). The issue template and the security policy are measured as issues.templates_present and security.policy_on

Folders with README or map row–goal ≥ 100

Top-level folders explained by a README inside them or a row in the folder map

README.md has no "Folder map" heading

top-level folders (not dot or git-ignored) with a README, or a row under "Folder map" in README.md

Goal: Git Crafter default goal: every top-level folder is explained. Source: Microsoft Engineering Fundamentals Playbook: Documentation (non-existent docs)

Runbooks with a last-verified date–goal ≥ 100

Runbooks that say when someone last ran their steps and found them right

no file under /home/runner/work/brand/brand matches params.runbooks

"last verified <date>" outside code in each of **/runbooks/*.md, less params.doc_set_left_out

Goal: Git Crafter default goal: every runbook is dated by its last check. Source: Microsoft Engineering Fundamentals Playbook: Documentation (obsolete docs)

Visual share0

Docs with an interactive element18.8%goal ≥ 80

Reader docs with something to open, pan or zoom: a diagram, a chart or a drill-down

Gap: 61.2 pts under goal

mermaid diagram9
mermaid chart (pie, xychart-beta)0
<details> block0

48 reader docs (the doc set less params.agent_docs) with a mermaid diagram, a mermaid chart or a <details> block outside code; detail counts elements, quality/collectors/docs/docs.mjs over 49 docs (params.doc_set) read 2026-11-16 21:12 UTC

Goal: Git Crafter default goal: most reader docs give the reader something to open, pan or zoom before more prose. Sources: GitHub Docs: Creating diagrams (Mermaid renders in Markdown); GitHub Docs: Organizing information with collapsed sections (a section the reader chooses to expand). The pan and zoom controls on a rendered diagram are as seen on github.com, 2026-11-10, not stated on that page. Docs written for an agent are left out: Claude Docs: Skill authoring best practices (concise is key)

Text-only docs6 docsgoal ≤ 0

Docs that are nearly all prose, with hardly a table, diagram, chart, image or code block

Gap: 6 docs over goal

art/4 docs
repo root2 docs

docs under 20% visual lines, by top-level folder, quality/collectors/docs/docs.mjs over 49 docs (params.doc_set) read 2026-11-16 21:12 UTC

Goal: Git Crafter default goal: no doc under 20% visual lines. Sources: Microsoft Style Guide: Scannable content (dense text set out as lists and tables); Microsoft Style Guide: Tables (a table is not for a plain list, so a one-column table counts as prose)

Speed

Each doc read in two minutes at 238 words a minute.

Reading time50

Docs over 2 minutes to read6 docsgoal ≤ 0

Docs that take longer to read than the reading goal, counted by folder

Gap: 6 docs over goal

art/4 docs
repo root2 docs

docs over 2 minutes at 238 words a minute, counted by top-level folder, quality/collectors/docs/docs.mjs over 49 docs (params.doc_set) read 2026-11-16 21:12 UTC

Goal: Git Crafter default goal: 2 minutes or less to read each doc, at 238 words a minute, the mean silent reading rate for English non-fiction. Sources: Brysbaert, 2019: How many words do we read per minute? (Journal of Memory and Language 109); Microsoft Style Guide: Top 10 tips for Microsoft style and voice (shorter is always better)

Consumption

No doc runs past its reading goal, so Claude loads no tokens past it.

Tokens50

Tokens past the reading goal3,779 tokensgoal ≤ 0

Tokens Claude loads from docs beyond the length a reader finishes within the reading goal

Gap: 3,779 tokens over goal

CHANGELOG.md2,341 tokens
LICENSES.md‎art/‎907 tokens
CLAUDE.md338 tokens
02-frame-and-size-math.md‎art/motion/‎76 tokens
11-renders-and-runs.md‎art/direction/‎72 tokens
10-prompts-fibo-2-0-in-vgl.md‎art/direction/‎45 tokens

bytes / 4 of each doc, the share of its words past 476 (2 minutes at 238 words a minute), summed; detail is the heaviest docs, quality/collectors/docs/docs.mjs over 49 docs (params.doc_set) read 2026-11-16 21:12 UTC

Goal: Git Crafter default goal: no doc runs past what a reader finishes in 2 minutes at 238 words a minute, so Claude loads no tokens past it; the near band is a Git Crafter default. Sources: Brysbaert, 2019: How many words do we read per minute? (Journal of Memory and Language 109); Claude Code Docs: How Claude remembers your project (longer files consume more context); Claude Docs: Skill authoring best practices (concise is key)

Issues

Quality

86

7 of 9 measured

Speed

100

1 of 2 measured

Consumption

–

0 of 1 measured

Ideal state

Issues hold bugs and quick fixes: typed, labeled by area, closed within a day, a bug by its pull request. Features wait on the board.

PartIdealSet up
IssuesOnly bugs and quick fixes, each typed, labeled by area, shown as a table or image, a bug closed by its pull requestUse the organization's Bug and Task issue types; file with gh issue create and its type flag
IntakeTwo valid forms, Bug and Quick fix, and no blank issueEach form sets its type; config.yml turns blank issues off
LabelsArea labels only; the type says what kind; a label nothing uses is archivedArchive through the labels API or the Labels page; history stays on old issues
MilestonesNone open, unless a dated goal spans several issuesGive each a due date and close it when done
ViewsShared, pinned views, each the search behind one metricCreate them in the Issues sidebar; GitHub offers no API for views
FeaturesA draft card on the board; an issue only while its pull request is openConvert the draft and add its area label when the pull request opens
16 sources, newest 2026-11-10
  1. GitHub Docs: About issues 2026-11-10
  2. GitHub Docs: Quickstart for GitHub Issues 2026-11-10
  3. GitHub Docs: Managing issue types in an organization 2026-11-10
  4. GitHub Docs: Syntax for issue forms 2026-11-10
  5. GitHub Docs: Configuring issue templates for your repository 2026-11-10
  6. GitHub Docs: Managing labels 2026-11-10
  7. GitHub Docs: REST API endpoints for labels 2026-11-10
  8. GitHub Changelog: Label archiving is generally available 2026-08-27
  9. GitHub Docs: About milestones 2026-11-10
  10. GitHub Changelog: Saved views for repository issues 2026-06-25
  11. GitHub Docs: Linking a pull request to an issue 2026-11-10
  12. GitHub Docs: Adding items to your project (draft issues) 2026-11-10
  13. GitHub Docs: Creating diagrams 2026-11-03
  14. Claude Code Docs: Claude Code GitHub Actions 2026-11-10
  15. Claude Code Action: Solutions and use cases (issue triage) 2026-11-10
  16. Microsoft Learn: Manage work with GitHub Projects 2026-11-10

Quality

Valid forms, every issue typed, labeled and shown as a table, no parked work or unused labels, every bug closed by a merged pull request.

Issues90

Unused labels1 labelsgoal ≤ 0

Labels in the picker that no issue or pull request carries and no setting names; fewer is better.

Gap: 1 labels over goal

idea: accepted1

state: gh api repos/Colorado-Startups/brand/labels (archived_at) and gh api graphql: repository.labels (issues and pullRequests totalCount), against params.bug_label and params.area_labels read 2026-11-16 21:12 UTC

Goal: GitHub docs, Managing labels (read 2026-11-10), and GitHub changelog, Label archiving is generally available (2026-08-27): an archived label leaves the picker and keeps its history; template goal: none unused

Open bugs0 bugsgoal ≤ 0

Problems that were reported and are not fixed yet; fewer is better.

code0 bugs
documentation0 bugs
design0 bugs
usability0 bugs
security0 bugs
ci0 bugs

gh api graphql: repository.issues (open: issueType, labels); on a user's repo, labels read 2026-11-16 21:12 UTC

Goal: template goal: no known bug left open

Parked issues0 issuesgoal ≤ 0

Open issues over a day old that are not a bug or a quick fix; such work waits on the board. Fewer is better.

no type0 issues

state: gh api graphql: repository.issues (open: createdAt, issueType), against params.issue_types and params.park_hours read 2026-11-16 21:12 UTC

Goal: GitHub docs, Managing issue types in an organization (read 2026-11-10): a type says what kind of work an issue is; GitHub docs, Adding items to your project (read 2026-11-10): a draft issue captures an idea on the board; template goal: none parked

Issues with an area label100%goal ≥ 100

Share of issues tagged with the part of the project they affect; higher is better.

with an area label1
without0

gh api graphql: repository.issues (labels) read 2026-11-16 21:12 UTC

Goal: template goal: every issue names its area, so it reaches whoever owns that area

Issue forms in order100%goal ≥ 100

Share of intake checks that hold: each issue form parses, each sets a type, blank issues are off. Higher is better.

every issue form parses, with a name, a description and a body1
every form sets a type: Bug, Task1
config.yml turns blank issues off1

state: .github/ISSUE_TEMPLATE, each YAML issue form parsed (name, description, body, type) and config.yml (blank_issues_enabled); the type check is against params.issue_types read 2026-11-16 21:12 UTC

Goal: GitHub docs, Syntax for issue forms and Configuring issue templates for your repository (read 2026-11-10): a form needs a name, a description and a body, may set its issue type, and config.yml turns blank issues off; template goal: every check holds

Bugs closed by a merged PR–goal ≥ 100

Share of recently fixed bugs that an accepted code change closed, not a close by hand; higher is better.

no bug closed as completed yet

gh api graphql: repository.issues, last 20 bugs closed as completed (issueType; on a user's repo, labels; ClosedEvent closer)

Goal: GitHub docs, Linking a pull request to an issue: a merged pull request with a closing keyword closes its issue; template goal: every fixed bug closes that way

Issues reopened–goal ≤ 5

Share of recently closed reports that had been reopened because a fix did not hold; lower is better.

no issue closed as completed yet

gh api graphql: repository.issues, last 20 closed as completed (REOPENED_EVENT)

Goal: template goal: one fix in twenty fails to hold at most

Cards75

Prose words, the latest card81 wordsgoal ≤ 75

Words the latest issue spends outside tables, diagrams and code; fewer is better, so the visuals carry it.

Gap: 6 words over goal

the latest issue #41, opened 2026-11-1681 words
newer issues with an empty body, passed over0

gh api graphql: of the 20 newest issues (body), the latest with a body; words outside tables, fenced code and diagrams, inline code and HTML comments (proseWords in github.mjs) read 2026-11-16 21:12 UTC

Goal: template goal: a card reads in under 20 seconds; adults read non-fiction at about 238 words a minute (Brysbaert, How many words do we read per minute?, Journal of Memory and Language, 2019), so 75 words take about 19 seconds

Cards with a visual100%goal ≥ 100

Share of issues that show their facts in a table, a diagram, a chart or an image; higher is better.

with a visual1
issues read1

gh api graphql: the 20 newest issues (body); the share with a table, a diagram GitHub draws (Mermaid, GeoJSON, TopoJSON, STL) or an image (hasVisual in github.mjs) read 2026-11-16 21:12 UTC

Goal: template goal: every card shows its facts visually; GitHub draws tables, Mermaid diagrams and charts, maps and images in issues (GitHub docs, Creating diagrams)

Speed

Each issue closes within a day, and none waits more than two.

Issues100

Oldest open issue0.7 daysgoal ≤ 2

How long the longest-waiting open issue has been open, in days; lower is better.

#410.7 days

gh api graphql: repository.issues (open) read 2026-11-16 21:12 UTC

Goal: template goal: no issue waits more than two days

Latest issue's hours to close–goal ≤ 24

How long the most recently fixed issue took from filed to closed, in hours; lower is better.

no issue closed as completed yet

gh api graphql: repository.issues, the latest closed as completed (createdAt, closedAt)

Goal: template goal (2026-11-03): an issue closes within a day of being filed

Consumption

A fix uses no more runner time than a typical merged pull request.

Cost–

Runner minutes per issue, the latest fix–goal ≤ 60

Billed runner minutes the CI runs of the latest pull request that closed an issue took, per issue it closed

no merged pull request among the closed ones read closed an issue

latest: GraphQL pullRequests closingIssuesReferences: the latest merged pull request that closed an issue; its runs (gh api actions/workflows/{id}/runs) and their REST jobs' billed runner minutes (one call per run at most, each read once), over the issues it closed

Goal: template goal (2026-11-03): a fix uses no more runner time than a merged pull request, 60 billed runner minutes, near 100, as for ci.latest_pr_dollars

Pull requests

Quality

82

11 of 11 measured

Speed

100

6 of 6 measured

Consumption

100

1 of 1 measured

Ideal state

Each change is a small pull request, stacked when large, linked to its issue, reviewed at its green head, and merged within the hour.

PartIdealSet up
Pull requests listA short queue: each open pull request is under a day old, green and free of conflictsOpen one small pull request per change; the Claude app's monitor with Auto-fix repairs red checks and conflicts
DescriptionA table of facts, a diagram where a flow changed, a linked issue or card, and few prose wordsFill the pull request template; write Closes or Part of before the issue number
Authored by meAn author's list empties the same day; no change is over 400 reviewed linesOne change and its test per pull request; split a large one, or stack it, before review
Review requestsNo request waits past an hour of green; the review reads the head commit that mergesRequest a person or team; where one account authors and merges, post a verdict comment naming the head commit
Assigned and Involves meNothing sits assigned or mentioned without an answer; every review thread is resolved before the mergeAssign whoever merges; answer or resolve each comment; a blocked pull request names what it waits on
Labels and MilestonesLabels come from automation, such as dependencies and area; a milestone only where a release is plannedLet Dependabot label its pull requests; leave Milestones empty unless a release needs one
StacksA large change is a stack of small pull requests based on the default branch, merged bottom up, never into a side branchCreate it with gh stack or the stacks API; the base branch's rules and checks then apply to every layer
MergeSquashed at the reviewed head within the hour of green, with every check passed and the branch deleted, never revertedAllow squash only, delete head branches, and require the CI check in a ruleset; merge pinned to the reviewed commit
16 sources, newest 2026-11-10
  1. GitHub Docs: Helping others review your changes 2026-11-10
  2. GitHub Docs: About stacked pull requests 2026-11-10
  3. GitHub Docs: Linking a pull request to an issue 2026-11-04
  4. GitHub Docs: Merging stacked pull requests 2026-11-10
  5. GitHub Changelog: Stacked pull requests generally available 2026-10-06
  6. GitHub Changelog: Refreshed repository pull requests page generally available 2026-09-21
  7. GitHub Docs: Requesting a pull request review 2026-11-10
  8. GitHub Docs: Approving a pull request with required reviews 2026-11-10
  9. GitHub Docs: Available rules for rulesets 2026-11-10
  10. GitHub Docs: About merge methods on GitHub 2026-11-10
  11. GitHub Docs: Searching issues and pull requests 2026-11-10
  12. Microsoft Engineering Fundamentals Playbook: Pull Requests 2024-02-15
  13. Claude Code Docs: Best practices for Claude Code 2026-11-10
  14. Claude Code Docs: Desktop application, monitoring pull request status 2026-11-10
  15. Google Engineering Practices: Small CLs 2026-11-03
  16. DORA: the four keys of software delivery performance 2026-01-05

Quality

Green and reviewed at the merged head, right at first review, small or stacked, described in a table, linked to an issue, never reverted.

Holds up100

Merged PRs green at head100%goal ≥ 100

Share of accepted code changes whose automatic checks all passed at the end; higher is better.

green20
red or pending0
no checks (left out)0

gh api graphql: last 20 merged pullRequests (head statusCheckRollup: SUCCESS, or each check's latest run passed) read 2026-11-16 21:12 UTC

Goal: GitHub docs, About protected branches: Require status checks before merging; template goal: every check passes at the merged head

Merged PRs reviewed at head100%goal ≥ 100

Share of accepted code changes that a person reviewed at the exact commit that was accepted; higher is better.

reviewed at the merged head20
reviewed another commit, or after merge0
not reviewed0

gh api graphql: last 20 merged pullRequests (headRefOid, mergedAt; reviews' commit.oid, submittedAt, author type; members' Review: MERGE|BLOCK at <sha> comments) read 2026-11-16 21:12 UTC

Goal: GitHub docs, About protected branches: Require pull request reviews before merging, and Dismiss stale pull request approvals when new commits are pushed, so a review covers the commit that merges

Fixed after the first review20%goal ≤ 25

Share of reviewed changes that needed another commit after their first review; lower means more pass first time.

fixed after the first review4
merged as first reviewed16
no review before merge (left out)0

gh api graphql: last 20 merged pullRequests (headRefOid; reviews' commit.oid, submittedAt, author type; members' Review: MERGE|BLOCK at <sha> comments) read 2026-11-16 21:12 UTC

Goal: template goal: three in four reviewed changes pass their first review

Reverts in the last 20 merges0 PRsgoal ≤ 1

Recent accepted changes that undo an earlier one, because something broke; fewer is better.

gh api graphql: last 20 merged pullRequests (title, mergedAt); a revert is titled Revert "<title>" read 2026-11-16 21:12 UTC

Goal: DORA, Accelerate State of DevOps Report 2023: elite teams' change failure rate is 5% and high performers' 10%, which is 1 and 2 of the last 20 merges

Branch commits through a PR100%goal ≥ 100

Share of recent commits on the default branch that came through a pull request, not pushed in; higher is better.

through a merged PR20
pushed directly0

gh api graphql: last 20 commits on the default branch (associatedPullRequests) read 2026-11-16 21:12 UTC

Goal: GitHub docs, About protected branches: Require a pull request before merging

Merged outside the default branch0 PRsgoal ≤ 0

Recent accepted changes merged into a side branch, where the default branch's rules may not apply; fewer is better.

gh api graphql: last 20 merged pullRequests (baseRefName, baseRepository.defaultBranchRef.name, stack) read 2026-11-16 21:12 UTC

Goal: GitHub docs, About stacked pull requests (read 2026-11-10): a stack's merge requirements come from the bottom pull request's base branch, and outside a stack rules and checks often only trigger for the bottom pull request; template goal: every change merges into the default branch, a large one as a stack

Size60

PRs linking an issue0%goal ≥ 100

Share of accepted changes whose description links the issue it fixes or serves; higher is better.

Gap: 100 pts under goal

linking an issue0
merged PRs read20

gh api graphql: the body of each of the last 20 merged pullRequests (N = 20, a template choice: enough for a median, few enough that new merges replace them); a closing keyword (close, fix or resolve) or Part of before an issue reference, outside fenced blocks (linksIssue in github.mjs) read 2026-11-16 21:12 UTC

Goal: GitHub docs, Linking a pull request to an issue (read 2026-11-04): close, fix or resolve before an issue reference closes it on merge; template goal: every change names its issue, with Part of #N when it only serves one

Share over 400 lines30%goal ≤ 10

Share of accepted changes too big to review in one sitting; lower is better.

Gap: 20 pts over goal

over 400 lines6
400 lines or fewer14

gh api graphql: last 20 merged pullRequests (additions, deletions, files, mergeCommit); git diff of each merge commit for moved lines read 2026-11-16 21:12 UTC

Goal: SmartBear's code review study at Cisco: a reviewer should take in no more than 200 to 400 lines at a time (SmartBear, Best Kept Secrets of Peer Code Review, 2006); the 10% share is a template goal

Prose words, the latest PR description71 wordsgoal ≤ 150

Words the latest accepted change's description spends outside tables, code and diagrams; fewer is better.

#4371 words
newer merges with an empty description, passed over0

gh api graphql: of the last 20 merged pullRequests (body), the latest with a description; words outside tables, fenced code and diagrams, inline code and HTML comments (proseWords in github.mjs) read 2026-11-16 21:12 UTC

Goal: template goal: a description reads in under a minute; adults read non-fiction at about 238 words a minute (Brysbaert, How many words do we read per minute?, Journal of Memory and Language, 2019), so 150 words take about 40 seconds and 300 about 75

PR descriptions with a table100%goal ≥ 100

Share of accepted changes whose description shows its facts in a table.

with a table20
merged PRs read20

gh api graphql: last 20 merged pullRequests (body), the share with a markdown table outside fenced code (hasTable in github.mjs) read 2026-11-16 21:12 UTC

Goal: template goal: every description shows its facts in a table, which a reviewer scans faster than prose

Lines changed, the latest PR15 linesgoal ≤ 200

Lines the latest accepted change asks a reviewer to read, less lockfiles, snapshots and moved lines; lower is better.

#4315 lines
lockfile and snapshot lines left out0 lines
moved lines left out0 lines

gh api graphql: the latest merged pullRequest (additions, deletions, files, mergeCommit); git diff of its merge commit for moved lines read 2026-11-16 21:12 UTC

Goal: SmartBear's code review study at Cisco: a reviewer should take in no more than 200 to 400 lines at a time (SmartBear, Best Kept Secrets of Peer Code Review, 2006)

Speed

Reviewed and merged within the hour of going green, with a short queue, no conflicts and nothing left waiting.

Flow100

Latest PR's hours to merge0.1 hoursgoal ≤ 24

How long the most recently merged pull request took from opened to merged, in hours, by stage; lower is better.

#43, open to merge0.1 hours
open to all green0 hours
all green to review0.1 hours
review to merge0 hours

gh api graphql: the latest merged pullRequest (createdAt, mergedAt; head check times; reviews) read 2026-11-16 21:12 UTC

Goal: DORA, Accelerate State of DevOps Report 2023: elite teams' lead time for changes is under a day and high performers' under a week; merging is part of that lead time

Minutes from all green to merge1 minutesgoal ≤ 60

Minutes a merged PR waited from all checks green at its head to merge, median of the last 10; lower is better.

#434.8 minutes
#421 minutes
#404.8 minutes
#390.8 minutes
#380.8 minutes
#370.9 minutes
#360.9 minutes
#351 minutes
#341 minutes
#332.7 minutes

gh api graphql: the last 10 merged pullRequests (N = 10, a template choice), median of mergedAt minus the last check's completion on the merged head, for each PR whose head rollup is SUCCESS read 2026-11-16 21:12 UTC

Goal: template goal, no public benchmark: review and merge within an hour of green CI

Queue100

Open PRs now0 PRsgoal ≤ 10

Code changes waiting right now to be checked and accepted, by how long each has waited; fewer is better.

under a day0 PRs
1 to 3 days0 PRs
3 to 7 days0 PRs
a week or more0 PRs

gh api graphql: open pullRequests (createdAt) read 2026-11-16 21:12 UTC

Goal: template goal: a queue short enough to review in one sitting

Longest wait after all green0 minutesgoal ≤ 60

Minutes the longest-waiting open change has sat since its checks passed, unreviewed or unmerged; lower is better.

with a pending review request0 PRs

gh api graphql: open pullRequests (isDraft, author; head statusCheckRollup and check times; reviews, reviewDecision, reviewRequests; members' Review: MERGE|BLOCK at <sha> comments) read 2026-11-16 21:12 UTC

Goal: template goal, no public benchmark: review and merge within an hour of green CI, as for pull_requests.green_to_merge_minutes

Open PRs with a conflict0 PRsgoal ≤ 0

Waiting changes that clash with the latest default branch and must be fixed before they can merge; fewer is better.

gh api graphql: open pullRequests (mergeable), read again after 5 s while GitHub is still computing it read 2026-11-16 21:12 UTC

Goal: template goal: no waiting change needs a rebase

Oldest open PR0 daysgoal ≤ 7

How long the longest-waiting code change has sat open and undecided, in days; lower is better.

open PRs0

gh api graphql: open pullRequests (createdAt) read 2026-11-16 21:12 UTC

Goal: template goal: no change waits more than a week

Consumption

Each merged change uses little runner time: few pushes, cheap checks.

Cost100

Runner dollars at list price, the latest merged PR0.05 dollarsgoal ≤ 0.36

Runner dollars the CI runs of the latest merged pull request cost; small, right-first-time changes cost less

ci0.02 dollars
pages0.02 dollars
scorecard0.01 dollars

latest: gh api pulls (the latest of the last 20 merged) and actions/workflows/{id}/runs (its branch runs and merge commit runs); their REST jobs, one call per run at most, each read once and most already read for the queue sample: each job's billed minutes times its runs-on label's rate from GitHub docs, Actions runner pricing; #43 read 2026-11-16 21:12 UTC

Goal: template goal: 60 runner minutes a merged pull request, near 100, priced at GitHub's standard Linux 2-core rate of $0.006 a minute (GitHub docs, Actions runner pricing, read 2026-11-03): 60 x $0.006 = $0.36, 100 x $0.006 = $0.60

Discussions

Quality

–

0 of 0 measured

Speed

–

0 of 0 measured

Consumption

–

0 of 0 measured

Ideal state

Questions and ideas live in Discussions, sorted by category; each question gets a quick first reply and a marked answer, and work moves to issues.

2 sources, newest 2026-11-13
  1. GitHub Docs: About discussions 2026-11-13
  2. GitHub Docs: Managing categories for discussions 2026-11-13

Quality

Every answered question has a marked answer, and every category is in use.

No quality metric yet.

Speed

A new discussion gets its first reply within a day.

No speed metric yet.

Consumption

Discussions cost no runner minutes.

No consumption metric yet.

Actions

Quality

100

6 of 13 measured

Speed

100

4 of 6 measured

Consumption

60

5 of 7 measured

Ideal state

Every push gets a trusted result within ten minutes from locked workflows, and included minutes last the cycle, free lanes first, no unneeded runs.

PartIdealSet up
WorkflowsEach workflow starts only when a file it checks changed, holds a read-only token, and a newer push cancels its older runAdd paths filters or a changes job, permissions contents: read at the top, concurrency with cancel-in-progress, and a timeout on each job
CachesEvery install restores a cache keyed on its lockfile, and storage stays under the repo's 10 GB so nothing is evicted earlysetup-node with cache: npm, or actions/cache keyed on hashFiles of the lockfile; entries unused for 7 days are removed
LockfileEvery workflow runs the action commits its lockfile records, and a workflow outside the lockfile cannot runRun gh actions-lock, commit .github/workflows/actions.lock, check it in CI, then turn on the Require lockfile policy
RunnersLight jobs run on ubuntu-slim at a third of the 2-core price, and free third-party minutes are used before included onesubuntu-slim for one-core jobs under 15 minutes; Blacksmith gives 3,000 and Ubicloud 1,250 free minutes a month; public repos run free
DeploymentsOnly the default branch deploys, through an environment, with OIDC in place of a stored secretCreate the environment and limit its deployment branches to the default branch; required reviewers on a private repo need Enterprise
Usage metricsIncluded usage lasts the whole billing cycle, and the workflows and jobs that bill the most are trimmed firstInsights, Actions Usage Metrics by workflow, job and runner; a budget stops Actions at the included amount, which resets each cycle
Performance metricsJobs start within 30 seconds, few fail, and the pipeline goes green within ten minutesInsights, Actions Performance Metrics: run time, queue time and failures by workflow and job; this tab scores job wait and failed runs
AttestationsEvery release, binary or package people download carries signed build provenance; test builds carry noneactions/attest with id-token: write, attestations: write and contents: read; consumers run gh attestation verify
20 sources, newest 2026-11-11
  1. GitHub docs: Secure use reference, lockfiles, pinning and least privilege 2026-11-10
  2. GitHub docs: Workflow dependency lockfiles 2026-11-10
  3. GitHub docs: Locking action dependencies in a repository's workflows 2026-11-10
  4. GitHub docs: Governing execution of GitHub Actions workflows 2026-11-10
  5. GitHub docs: Dependency caching reference, limits and eviction 2026-11-10
  6. GitHub docs: About GitHub Actions metrics, usage and performance 2026-11-10
  7. GitHub docs: Artifact attestations 2026-11-10
  8. GitHub docs: GitHub Actions billing, included usage and runner rates 2026-11-10
  9. GitHub docs: Actions runner pricing 2026-11-10
  10. GitHub docs: GitHub-hosted runners reference, ubuntu-slim 2026-11-10
  11. GitHub docs: Deployments and environments 2026-11-10
  12. GitHub docs: Using artifact attestations 2026-11-10
  13. Microsoft Learn: Security considerations for DevOps platforms 2026-01-15
  14. Claude Code docs: Claude Code GitHub Actions, permissions and cost 2026-11-10
  15. GitHub docs: Workflow syntax, paths filters, concurrency and permissions 2026-11-10
  16. DORA: Capabilities, continuous integration 2026-11-10
  17. GitHub changelog: Actions workflow lockfiles are generally available 2026-10-28
  18. Martin Fowler: Continuous Integration 2024-01-18
  19. Blacksmith: Pricing, 3,000 free minutes a month 2026-11-11
  20. Ubicloud: GitHub Actions runners, 1,250 free minutes a month 2026-11-11

Pipeline: the latest push to all green

The slowest step of each stage95% or more pass85 to 94%Under 85%Not measuredRuns on the laptopA personBlue bar: median minutes

  1. Pull request checks1.3 min

    • pages
      Median
      1.3 minutes when it does its work
      Skips
      73% of sampled runs found nothing to check
      Pass rate
      100% of runs passed; cancelled runs are left out
      What runs
      On every pull request touching **/*.md, guide/**, dist/**, quality/**, .github/workflows/pages.yml: build on ubuntu-latest: uses actions/checkout, uses actions/setup-node, uses actions/create-github-app-token, uses SoulCraft-Org/git-crafter/actions/pages, Assemble the site, Leak scan and franchise gate on the built si…
      Defined in
      .github/workflows/pages.yml
      Kind
      An automated workflow: a GitHub Actions workflow or a Cloudflare build
      Path
      The slowest step in its stage, on the rail
    • ci
      Median
      1.1 minutes when it does its work
      Skips
      92% of sampled runs found nothing to check
      Pass rate
      100% of runs passed; cancelled runs are left out
      What runs
      On every pull request: ci calls SoulCraft-Org/git-crafter/.github/workflows/ci-standard.yml
      Defined in
      .github/workflows/ci.yml
      Kind
      An automated workflow: a GitHub Actions workflow or a Cloudflare build
    • claude-review
      Median
      0.1 minutes when it does its work
      Skips
      100% of sampled runs found nothing to check
      Pass rate
      100% of runs passed; cancelled runs are left out
      What runs
      On every pull request touching **/*.mjs, **/*.js, **/*.ts, **/*.py, **/*.sh, tools/**, actions/**, .github/workflows/**, .claude/settings.json, manifest.json, package.json, package-lock.json: claude-review on ubuntu-latest: Skip without the token, uses actions/checkout, uses anthropics/claude-code-action
      Defined in
      .github/workflows/claude-review.yml
      Kind
      An automated workflow: a GitHub Actions workflow or a Cloudflare build
    • cloud-test
      Median
      Not timed
      Pass rate
      not measured
      What runs
      On every pull request touching .github/workflows/cloud-test.yml, .github/cloud-test.json: cloud-test calls Colorado-Startups/.github/.github/workflows/cloud-test.yml in 1 shard
      Defined in
      .github/workflows/cloud-test.yml
      Kind
      An automated workflow: a GitHub Actions workflow or a Cloudflare build
  2. Review4.8 min

    • Review and merge
      Median
      4.8 minutes
      Pass rate
      100% of runs passed; cancelled runs are left out
      What runs
      A person reviews the pull request, then merges it; timed from all checks green to the merge
      Defined in
      branch rules on the default branch
      Kind
      A step a person takes
      Path
      The slowest step in its stage, on the rail
  3. Merge to main1.2 min

    • pages
      Median
      1.2 minutes
      Pass rate
      100% of runs passed; cancelled runs are left out
      What runs
      On every push to main: build on ubuntu-latest, then deploy on ubuntu-latest
      Defined in
      .github/workflows/pages.yml
      Kind
      An automated workflow: a GitHub Actions workflow or a Cloudflare build
      Path
      The slowest step in its stage, on the rail
    • ci
      Median
      0.6 minutes when it does its work
      Skips
      100% of sampled runs found nothing to check
      Pass rate
      100% of runs passed; cancelled runs are left out
      What runs
      On every push to main: ci calls SoulCraft-Org/git-crafter/.github/workflows/ci-standard.yml
      Defined in
      .github/workflows/ci.yml
      Kind
      An automated workflow: a GitHub Actions workflow or a Cloudflare build
    • scorecard
      Median
      0.7 minutes
      Pass rate
      100% of runs passed; cancelled runs are left out
      What runs
      On every push to main: analysis on ubuntu-latest: uses actions/checkout, uses ossf/scorecard-action, uses github/codeql-action/upload-sarif
      Defined in
      .github/workflows/scorecard.yml
      Kind
      An automated workflow: a GitHub Actions workflow or a Cloudflare build
Every step as a table
StageStepKindMedianPass rate
Pull request checkspages (rail)Workflow1.3 min100%
Pull request checksciWorkflow1.1 min100%
Pull request checksclaude-reviewWorkflow0.1 min100%
Pull request checkscloud-testWorkflownot timednot measured
ReviewReview and merge (rail)Person4.8 min100%
Merge to mainpages (rail)Workflow1.2 min100%
Merge to mainciWorkflow0.6 min100%
Merge to mainscorecardWorkflow0.7 min100%

Quality

No flaky reruns, installs restore their cache, and every workflow runs the commits its lockfile records.

Reliable100

Checks that failed, then passed on the same commit0%goal ≤ 1

Share of workflow checks on one commit that failed and later passed with no change to the code

ci0
claude-review0
pages0
scorecard0

gh api actions/workflows/{id}/runs (last 20 runs per workflow) and runs/{id}/attempts/1 for every rerun that passed read 2026-11-16 21:12 UTC

Goal: template goal: below the 1.5% of test runs that Google reported flaky (Google Testing Blog, Flaky Tests at Google and How We Mitigate Them, 2016)

Workflow runs that failed0%goal ≤ 10

Share of workflow runs that failed, of those that passed or failed; each failure costs a fix and another run

ci0%
claude-review0%
pages0%
scorecard0%

gh api actions/workflows/{id}/runs (last 20 runs per workflow): runs that ended failure, timed_out, startup_failure, over those that passed or failed, on any event read 2026-11-16 21:12 UTC

Goal: template goal (2026-11-11): no first-party source sets a number; one run in ten, near one in five. GitHub docs, About GitHub Actions metrics (read 2026-11-11) lists failure rates as a key indicator and sets none. Cancelled and skipped runs give no verdict, so they are left out

Workflows locked to recorded commits100%goal ≥ 100

Share of workflows that call outside actions and are in the lockfile, so each run uses the commits recorded there.

claude-review.yml1
pages.yml1
scorecard.yml1

.github/workflows/actions.lock (workflows) against each .github/workflows/*.yml that calls an outside action in a step read 2026-11-16 21:12 UTC

Goal: GitHub docs, Secure use reference (read 2026-11-10): lock workflow dependencies with a lockfile, which supersedes SHA pinning where adopted; Workflow dependency lockfiles (read 2026-11-10): every dependency of an onboarded workflow must be in the lockfile, and its recorded commit is what runs

npm installs that reused the cache–goal ≥ 90

Share of npm installs in CI that restored the saved package cache instead of downloading every package again

no log read had setup-node's cache line

gh api repos/Colorado-Startups/brand/actions/jobs/{id}/logs (setup-node's npm cache line, newest 12 jobs)

Goal: GitHub docs, Caching dependencies to speed up workflows; template goal: nine npm installs in ten restore the cache

Actions hardening100

Actions pinned to a SHA100%goal ≥ 100

Share of outside tools in our automatic jobs locked to one exact version; higher is better.

ci.yml100%
claude-review.yml100%
cloud-test.yml100%
pages.yml100%
scorecard.yml100%

.github/workflows/*.yml (jobs.*.uses and jobs.*.steps[].uses, parsed as YAML) and the uses: lines .github/workflows/actions.lock records for each read 2026-11-16 21:12 UTC

Goal: GitHub docs, Security hardening for GitHub Actions: pinning an action to a full-length commit SHA is the only way to use it as an immutable release; Workflow dependency lockfiles (read 2026-11-10): a locked reference runs its recorded commit; OpenSSF Scorecard, Pinned-Dependencies

Workflows with explicit permissions100%goal ≥ 100

Share of automatic job files that spell out what they are allowed to do; higher is better.

ci.yml1
claude-review.yml1
cloud-test.yml1
pages.yml1
scorecard.yml1

.github/workflows/*.yml (permissions: blocks) read 2026-11-16 21:12 UTC

Goal: GitHub docs, Automatic token authentication: grant GITHUB_TOKEN the least access each workflow needs, with a permissions key; OpenSSF Scorecard, Token-Permissions

Privileged jobs running pull request code0 jobsgoal ≤ 0

Jobs that run with the repo's secrets on a pull request's own code; none is the only safe number.

.github/workflows/*.yml: jobs of a pull_request_target or workflow_run workflow that check out the pull request's code: an actions/checkout of its head ref or repo, gh pr checkout, or a git fetch or checkout of refs/pull/* read 2026-11-16 21:12 UTC

Goal: GitHub docs, Security hardening for GitHub Actions: pull_request_target and workflow_run, used with a checkout of an untrusted pull request, expose the repo; OpenSSF Scorecard, Dangerous-Workflow

GITHUB_TOKEN reads only by default–goal ≥ 1

Whether a workflow's token can only read unless the workflow asks for more; on is better.

needs an admin token; refreshed by a --local collect run

gh api repos/Colorado-Startups/brand/actions/permissions/workflow (default_workflow_permissions, can_approve_pull_request_reviews)

Goal: GitHub docs, Security hardening for GitHub Actions: grant the GITHUB_TOKEN the least access it needs; Managing GitHub Actions settings for a repository: Workflow permissions

Actions policy settings that are on–goal ≥ 100

Share of two repo settings that are on: only selected actions may run, and each must be pinned to a full commit.

needs an admin token; refreshed by a --local collect run

gh api repos/Colorado-Startups/brand/actions/permissions (allowed_actions, sha_pinning_required)

Goal: GitHub docs, Governing execution of GitHub Actions workflows (read 2026-11-10): keep the full-length commit SHA requirement on beside a lockfile; Managing GitHub Actions settings for a repository: allow only selected actions and reusable workflows

Coverage–

Page routes an e2e spec opens–goal ≥ 90

Share of the site's page routes that an end-to-end spec opens in a real browser

git ls-files listed no page in params.pages_dir (src/pages/)

state: page files in params.pages_dir and the site paths the specs of params.suites' e2e suite name, in code lines

Goal: Git Crafter default goal: nearly every page route is opened by an end-to-end spec

Page routes with an axe check–goal ≥ 100

Share of the site's page routes that an end-to-end spec checks for accessibility with axe

git ls-files listed no page in params.pages_dir (src/pages/)

state: page files in params.pages_dir and the site paths named by the e2e suite's specs that run AxeBuilder

Goal: goal: every route, as WCAG 2.2 conformance is claimed for full pages only (WCAG 2.2, Conformance Requirement 2, w3.org/TR/WCAG22); the near band is a Git Crafter default

Trust–

Test job pass rate on main–goal ≥ 95

Share of the latest runs on main whose test job passed; cancelled runs and runs with no site change are left out

params.workflows names no test workflow

last N runs: gh api runs of each workflow in params.workflows (newest 100 runs a workflow), with GraphQL check runs: the last 20 runs on main of each workflow that runs there; a run counts when a test job ran, and cancelled runs and runs that skipped every test job are left out. N is 20 (ci.mjs's LAST_RUNS): enough to turn over within hours where the test workflows run many times a day

Goal: Git Crafter default goal: the test jobs pass on main nearly every time, so a red run means a real break

Skipped and todo tests–goal ≤ 0

Tests marked skip, todo or fixme, which never run; skips on a condition are read from code and shown apart

git ls-files listed no test file in any suite folder

state: git ls-files in params.suites; skip, todo and fixme markers in code lines, a marked block by its cases. skipIf, runIf and run-time skips are read from code, not CI, and shown apart

Goal: Git Crafter default goal: no test skips unconditionally; a skipped test checks nothing

Speed

Push to all green within ten minutes; a red default branch fixed within ten minutes.

Fast100

Pipeline: the latest push to all green1 minutesgoal ≤ 10

Minutes from the latest pull request push to every check passing; the drawing shows each step's latest run

.claude/settings.json, the git hooks in params.hooks_dir, .github/workflows, params.stages; gh api actions/workflows/{id}/runs (the latest pull request head that went all green; each stage's latest run, pass rates from the last 20 runs per workflow), pulls and check-runs (the latest of the last 20 merges) read 2026-11-16 21:13 UTC

Goal: the ten-minute build (Kent Beck, Extreme Programming Explained, 2nd edition, 2004; Martin Fowler, Continuous Integration, 2024-01-18) and DORA's continuous integration capability (read 2026-11-03): tests take an upper limit of about 10 minutes

Slowest workflow, its latest real run1.2 minutesgoal ≤ 10

Minutes the slowest workflow took the last time it did real work; runs that skip their work are left out

ci1.1 minutes
pages1.2 minutes
scorecard1.1 minutes

GraphQL check runs and steps of the last 20 runs per workflow: each workflow's latest run that did real work read 2026-11-16 21:13 UTC

Goal: the ten-minute build (Kent Beck, Extreme Programming Explained, 2nd edition, 2004; Martin Fowler, Continuous Integration, 2024-01-18; DORA's continuous integration capability, read 2026-11-03): workflows run side by side, so the slowest one sets the build time

Job wait for a runner4 secondsgoal ≤ 30

Seconds a job waits between being queued and starting on a runner, at the 90th percentile; lower is better

ci4 seconds
claude-review2 seconds
pages4 seconds
scorecard2.1 seconds

last 10 completed runs per workflow, the queue sample the CI tab already reads (gh api actions/runs/{id}/jobs): each job's started_at less created_at read 2026-11-16 21:13 UTC

Goal: GitHub docs, Actions metrics (read 2026-11-10): the performance view reports average queue time and sets no number; template goal: nine jobs in ten start within 30 seconds, near 60

Reliable100

Minutes the default branch is red0 minutesgoal ≤ 10

How long a workflow on the default branch has been failing, right now; 0 when every one passed last; lower is better

workflows green on main3

gh api actions/workflows/{id}/runs: each active workflow whose file still runs on a push to the default branch, its runs there, newest first; red from the end of the first failure after its latest success read 2026-11-16 21:12 UTC

Goal: DORA's continuous integration capability (read 2026-11-03): a build that cannot be fixed in a few minutes has its change reverted; Martin Fowler, Continuous Integration (2024-01-18): fix broken builds immediately; template goal: 10 minutes, one build's length

Suite time–

Test job, median–goal ≤ 10

Typical minutes for the job that runs the test suites, in its latest passes

params.workflows names no test workflow

last N passing jobs: gh api runs of the suite workflow in params.workflows (newest 100 runs a workflow), with GraphQL check runs: the last 20 passing test jobs. N is 20 (ci.mjs's LAST_RUNS): enough to turn over within hours where the test workflows run many times a day

Goal: goal: the ten-minute build, the same source as ci.latest_full_run_minutes (Kent Beck, Extreme Programming Explained, 2nd edition, 2004; Martin Fowler, Continuous Integration, martinfowler.com)

Slowest e2e shard, median–goal ≤ 10

Typical minutes for the slowest end-to-end shard; the shards split the browser specs

params.workflows names no test workflow

last N passing jobs: gh api runs of the e2e workflow in params.workflows (newest 100 runs a workflow), with GraphQL check runs: the last 20 passing jobs of each shard. N is 20 (ci.mjs's LAST_RUNS): enough to turn over within hours where the test workflows run many times a day

Goal: goal: the ten-minute build, the same source as ci.latest_full_run_minutes (Kent Beck, Extreme Programming Explained, 2nd edition, 2004; Martin Fowler, Continuous Integration, martinfowler.com)

Consumption

Included minutes last the cycle: free lanes first, light jobs on the cheapest runner, no idle runner, and a newer push cancels the older run.

Cost60

Runs that find nothing to do66.2%goal ≤ 25

Share of runs that start a runner only to skip their work because nothing they check changed; each still bills

Gap: 41.2 pts over goal

ci94.8%
claude-review100%
pages45.7%
scorecard0%

GraphQL check runs and steps of the last 20 runs per workflow, weighted by each workflow and event's runs since the start of the shortest full list of 100 read 2026-11-16 21:13 UTC

Goal: template goal: at most one run in four starts a runner only to skip its work

Triggers that skip work when nothing changed71.4%goal ≥ 80

Share of pull request and push triggers that skip their work when no file they check changed; a skipped run bills less

Gap: 8.6 pts under goal

ci, pull request: ci-standard paths1
ci, push to main: ci-standard paths1
claude-review, pull request: paths filter1
cloud-test, pull request: paths filter1
pages, pull request: paths filter1
pages, push to main: runs its work on every change0
scorecard, push to main: runs its work on every change0

state: .github/workflows, each pull_request and push to the default branch trigger (params.default_branch, else gh api repos/Colorado-Startups/brand): a paths or paths-ignore filter, a job whose if reads a needs.*.outputs value (a changes job), or a job calling ci-standard.yml with paths read 2026-11-16 21:12 UTC

Goal: GitHub docs, Workflow syntax for GitHub Actions: on.<push|pull_request>.paths and paths-ignore skip a run when no matching file changed; template goal: at most one trigger in five runs its work on every change

Runner list price per billed minute0.45 centsgoal ≤ 0.4

Cents a billed runner minute costs at list price; lighter jobs on the 1-core ubuntu-slim runner bring it down

Gap: 0.05 cents over goal

ci0.3 cents
claude-review0.6 cents
pages0.6 cents
scorecard0.6 cents

last 10 completed runs per workflow, the queue sample (gh api actions/runs/{id}/jobs): each job's billed minutes times its runs-on label's list rate, over the minutes read 2026-11-16 21:13 UTC

Goal: template goal (2026-11-10): half the billed minutes on ubuntu-slim, priced from GitHub docs, Actions runner pricing (read 2026-11-10): ubuntu-slim 0.2 cents and Linux 2-core 0.6 cents a minute, (0.2 + 0.6) / 2 = 0.4; near 0.6, every minute on the 2-core rate

Runner minutes, the latest push to the default branch7 minutesgoal ≤ 10

Billed runner minutes the CI runs of the latest commit pushed to the default branch took, summed over its workflows

ci4 minutes
pages2 minutes
scorecard1 minutes

latest: gh api actions/workflows/{id}/runs, the push runs of the latest commit on the default branch whose runs all ended; their REST jobs' billed runner minutes (gh api actions/runs/{id}/jobs, one call per run at most, each read once), summed; 581e7e9 priced read 2026-11-16 21:12 UTC

Goal: GitHub docs, GitHub Actions billing (read 2026-11-10): each job bills its minutes rounded up, so every workflow a push starts bills at least one; template goal (2026-11-10): the runs one push starts bill at most the ten-minute build (Martin Fowler, Continuous Integration, 2024-01-18), near 20

GitHub API left this hour100%goal ≥ 50

Share of the hourly GitHub API quota still unused, REST or GraphQL, whichever is lower; every session shares it

REST requests left of 15000100%
GraphQL points left of 10000100%

latest: gh api rate_limit (counts against no limit), the share of each hourly bucket still unused, the lower of REST and GraphQL; read with the CI run's token (the workflow token unless the repo passes another) read 2026-11-16 21:12 UTC

Goal: goal set from readings (2026-11-11): the account's login had 52.8% of its REST requests left right after an org collect (2,638 of 5,000, 2026-11-10 18:23 UTC) and 95.7% at a quiet hour (GraphQL 4,783 of 5,000, 2026-11-11 16:23 UTC); a goal of half left keeps a full collect's room, near a fifth

GitHub requests per pull request run–goal ≤ 100

Requests the workflow token spent live in the newest green pull request run that collected; a replayed read costs none

the jobs of runs 43898764468, 43897826179, 43896790082 printed no GitHub requests notice

latest: gh api actions/workflows/pages.yml/runs, the newest green pull_request run from Colorado-Startups/brand whose jobs printed a notice, of the newest 3; its jobs' check-run annotations, each job's GitHub requests notice (quality/gh-replay.mjs report), live REST plus GraphQL, summed

Goal: goal from issue #38 (2026-11-12); near is the 765 a pull request run spent before replay (run 43307560596's quota lines). The workflow token reports limit 5000 an hour (gh api rate_limit in runs 43307560596 and 43355450949)

Billed test job minutes, latest run–goal ≤ 20

Runner minutes GitHub billed for the site test job and each browser test shard in their latest run

params.workflows names no test workflow

latest: the latest completed run of each workflow in params.workflows whose test jobs ran, on main where the workflow runs there, else its latest pull request run (newest 100 runs a workflow); GitHub bills each job by the whole minute

Goal: Git Crafter default goal: the test jobs' sum at their speed goals, one suite job and one e2e job at 10 minutes each, 10 + 10 = 20; near the same at their near bands, 20 + 20 = 40. A repo with more jobs sets its own

Projects

Quality

–

0 of 6 measured

Speed

–

0 of 2 measured

Consumption

–

0 of 1 measured

Ideal state

Features are visual cards; bugs stay in issues. A card in progress has an open pull request, and Done cards link a merged one.

PartIdealSet up
CardsA feature is a draft card: a picture of its metric and steps, a table, at most 75 wordsAdd it on the board; convert it to an issue, with its area label, when its pull request opens
PicturesOne image per card: its pillar, a meter from now to goal, its steps left to rightRun the card picture tool, push the file to the cards branch, link it by its raw address
FieldsStatus, Pillar and Area; each fact lives in one fieldCreate them with the Projects API; show Linked pull requests on every card
ViewsBoard, Backlog, Shipped and Stuck: each is the search behind one metricCreate views with the REST views endpoint; set no column limit
WorkflowsAdded sets Todo, a linked pull request In Progress, closed or merged Done, reopened TodoTurn each on in the Workflows page; GitHub offers no API to edit them
InsightsCharts of cards by status and pillar, by area and status, and the burn upCreate them in the Insights page; never archive a card, since archived cards leave the charts
SessionsOne fresh session per card, ended at the merge; the card names its goal, its files and a checkStart a session per card in its own worktree; merge a green pull request before opening another
LimitNo count limit on cards; a card in progress has an open pull request, and open pull requests carry the limitMove a waiting card back to Todo and name what it waits on
18 sources, newest 2026-11-10
  1. GitHub Docs: Best practices for Projects 2026-11-10
  2. GitHub Docs: Using the built-in automations 2026-11-10
  3. GitHub Docs: Customizing the board layout 2026-11-10
  4. GitHub Docs: About insights for Projects 2026-11-10
  5. GitHub Docs: Using the API to manage Projects 2026-11-10
  6. GitHub Docs: REST API endpoints for Project views 2026-11-10
  7. GitHub Docs: Adding items to your project 2026-11-10
  8. GitHub Docs: Basic writing and formatting syntax 2026-11-10
  9. Claude Code Docs: Best practices for Claude Code 2026-11-10
  10. Claude Code Docs: Claude Code GitHub Actions 2026-11-10
  11. Anthropic: Effective harnesses for long-running agents 2026-11-10
  12. Microsoft Learn: Set Work in Progress limits in Azure Boards 2026-03-24
  13. Microsoft Learn: Cumulative flow, lead time and cycle time guidance 2026-11-10
  14. Microsoft Learn: Show bugs on backlogs and boards 2026-11-10
  15. DORA: Working in small batches 2025-12-08
  16. DORA: Work in process limits 2026-11-10
  17. Claude Code Docs: Manage costs effectively 2026-11-10
  18. Anthropic: How we built our multi-agent research system 2026-11-10

Quality

Every open card shows a picture in few words, no bug sits on the board, and each card in progress has an open pull request.

Board–

In progress, no open PR–goal ≤ 0

Cards in the in-progress column that no open pull request is building: stalled or waiting; lower is better.

needs the project scope: gh auth refresh -s project; refreshed by a --local collect run

gh api graphql: repositoryOwner.projectsV2 "Brand" items (Status, linked pull requests and their state)

Goal: DORA, Work in process limits (read 2026-11-10): the point of a limit is to expose problems in the system so they can be addressed; template goal: one session builds one card, so a count of cards limits nothing, a card in progress has an open pull request, and pull_requests.open_now carries the limit on work in flight

Done cards with a merged PR–goal ≥ 100

Share of finished board tasks that link to an accepted code change; higher is better.

needs the project scope: gh auth refresh -s project; refreshed by a --local collect run

gh api graphql: repositoryOwner.projectsV2 "Brand" done items (linked pull requests), less those whose body starts with the sync marker <!-- quality-metric: id -->

Goal: template goal: every finished card links the change that finished it

Board workflows on–goal ≥ 100

Share of the built-in board workflows that are on: add, link, close and merge move a card unaided; higher is better.

needs the project scope: gh auth refresh -s project; refreshed by a --local collect run

gh api graphql: repositoryOwner.projectsV2 "Brand" workflows (enabled), against params.project_workflows

Goal: GitHub docs, Using the built-in automations (read 2026-11-10): a board sets Status when an item is added, closed or merged and adds matching items; template goal: all five on, the fifth moving a card to in progress when a pull request links its issue

Cards read at a glance–goal ≥ 100

Share of open cards with a picture and at most 75 words of prose, so the picture carries the card; higher is better.

needs the project scope: gh auth refresh -s project; refreshed by a --local collect run

gh api graphql: repositoryOwner.projectsV2 "Brand" open draft and issue cards (body): a picture (hasPicture) and at most params.card_words words of prose (proseWords)

Goal: template goal: every card shows its work as a picture with few words; GitHub docs, Basic writing and formatting syntax (read 2026-11-10): an image in a body is drawn in place; Anthropic, Best practices for Claude Code (read 2026-11-10): paste the screenshot of the design, and a screenshot next to the mockup is a check; 75 words read in about 19 seconds (Brysbaert, 2019)

Bugs and fixes on board–goal ≤ 0

Open issue cards that are a bug or a quick fix; those are issues, not cards, so lower is better.

needs the project scope: gh auth refresh -s project; refreshed by a --local collect run

gh api graphql: repositoryOwner.projectsV2 "Brand" open issue cards (labels, issueType), against params.bug_label and params.issue_types

Goal: template goal: a bug or a quick fix is an issue and a feature is a card; GitHub docs, Managing issue types in an organization (read 2026-11-10): a type says what kind of work an issue is; Microsoft Learn, Show bugs on backlogs and boards (read 2026-11-10) lists keeping bugs off the board as one of three options

Closed issues off Done–goal ≤ 0

Closed issues whose board card is not in the last column; a merged PR should have moved it, so lower is better.

needs the project scope: gh auth refresh -s project; refreshed by a --local collect run

gh api graphql: repositoryOwner.projectsV2 "Brand" issue cards (issue state, Status)

Goal: template goal: a closed issue's card is Done; GitHub docs, Using the built-in automations (read 2026-11-11): the Item closed workflow sets Status when an issue closes; Git Crafter issue 791: a card stayed In Progress after its issue closed by a merged PR

Speed

A card is done within three days of its first pull request, and none stays in progress past a week.

Board–

Oldest card in progress–goal ≤ 7

Days the longest-waiting card has sat in the in-progress column; lower is better.

needs the project scope: gh auth refresh -s project; refreshed by a --local collect run

gh api graphql: repositoryOwner.projectsV2 "Brand" items in the in-progress column (Status updatedAt)

Goal: DORA, Working in small batches (updated 2025-12-08, read 2026-11-10): any batch of code that takes longer than a week to complete and check is too big; template goal: no card stays in progress past a week, near is two weeks

Latest card's hours to done–goal ≤ 72

How long the card most recently marked done took from its first pull request opening, in hours; lower is better.

needs the project scope: gh auth refresh -s project; refreshed by a --local collect run

gh api graphql: repositoryOwner.projectsV2 "Brand" the latest done item with a linked PR (Status updatedAt, linked PR createdAt)

Goal: template goal (2026-11-03): a task finishes within three days of its first pull request

Consumption

A card's pull requests use no more runner time than a typical merged pull request.

Cost–

Runner minutes per pull request, the latest card–goal ≤ 60

Billed runner minutes the CI runs took, per merged pull request of the card most recently marked done

needs the project scope: gh auth refresh -s project; refreshed by a --local collect run

latest: the board named by github.json's params.project_title (gh api graphql, projectsV2): of the five cards most recently moved to the done column that link merged pull requests, the newest whose runs can be priced; their runs (gh api actions/workflows/{id}/runs) and REST jobs' billed runner minutes (one call per run at most, each read once), summed and divided by those pull requests

Goal: template goal (2026-11-04): a card's pull requests use no more runner time than a merged pull request, 60 billed runner minutes, near 100, as for ci.latest_pr_dollars

Security and quality

Quality

33

3 of 3 measured

Speed

–

0 of 2 measured

Consumption

–

0 of 1 measured

Ideal state

Reporters have a private path, the threat model keeps up with its code, no alert outlives a week, and no paid security feature bills unasked.

PartIdealSet up
Security policySECURITY.md names a private reporting channel that works for this repo's visibilityAdd SECURITY.md; turn on private vulnerability reporting, which GitHub offers on public repos
ActionsA read-only token by default; only listed actions run, each at a full commit SHA or in a lockfileActions settings: selected actions, required SHA pins, read-only workflow permissions
Claude CodeClaude Code cannot read a credential file in the repo or the home folderDeny Read rules for .env, key and secrets paths in .claude/settings.json
Threat modelEach covered element has its threat, control and gap, re-read as part of routine developmentTHREAT-MODEL.md beside the code; name the paths it covers and the days it may trail them
29 sources, newest 2026-11-10
  1. GitHub: Advanced Security plans and prices 2026-11-10
  2. GitHub Docs: GitHub security features 2026-11-10
  3. GitHub Docs: Quickstart for securing your repository 2026-11-10
  4. GitHub Docs: Adding a security policy to your repository 2026-11-10
  5. GitHub Docs: Configuring private vulnerability reporting for a repository 2026-11-10
  6. GitHub Docs: About Dependabot alerts 2026-11-10
  7. GitHub Docs: Dependabot malware alerts 2026-11-10
  8. GitHub Docs: Dependabot auto-triage rules 2026-11-10
  9. GitHub Docs: About code scanning 2026-11-10
  10. GitHub Docs: About secret scanning 2026-11-10
  11. GitHub Docs: About push protection 2026-11-10
  12. GitHub Docs: GitHub Code Quality 2026-11-10
  13. GitHub Docs: GitHub Code Quality billing 2026-11-10
  14. GitHub Docs: GitHub Advanced Security license billing 2026-11-10
  15. GitHub Docs: Secure use reference (formerly Security hardening for GitHub Actions) 2026-11-10
  16. GitHub Docs: Managing GitHub Actions settings for a repository 2026-11-10
  17. GitHub Docs: REST API endpoints for GitHub Actions permissions 2026-11-10
  18. GitHub Docs: Enforcing policies for GitHub Actions in your enterprise 2026-11-10
  19. GitHub Docs: About Dependabot on GitHub Actions runners 2026-11-10
  20. GitHub Docs: Dependabot options reference 2026-11-10
  21. GitHub Docs: Security overview dashboard metrics 2026-11-10
  22. OpenSSF: S2C2F framework requirements (AUD-2, ENF-1, ING-1) 2026-11-10
  23. Anthropic: Claude Code settings 2026-11-10
  24. Anthropic: Claude Code security 2026-11-10
  25. Anthropic: Claude Code GitHub Actions 2026-11-10
  26. CISA: BOD 26-04, Prioritizing Security Updates Based on Risk 2026-06-10
  27. Microsoft: Security Development Lifecycle (SDL) practices 2026-11-10
  28. Microsoft: Security Development Lifecycle, threat modeling 2026-11-10
  29. OpenSSF Scorecard: checks 2026-11-10

Quality

A security policy with private reporting, a current threat model, and no secret a workflow or Claude can read.

GitHub features100

Security policy1 on/offgoal ≥ 1

Whether the repo tells people how to report a vulnerability, in its own SECURITY.md or the organization's; on is better.

gh api graphql: repository.isSecurityPolicyEnabled (a SECURITY.md in the repo, or the organization's) read 2026-11-16 21:12 UTC

Goal: GitHub docs, Adding a security policy to your repository

Design review0

Threat model current with its code0 on/offgoal ≥ 1

Whether the threat model was updated since its code last changed, within the repo's days of grace; on is better.

Gap: 1 on/off under goal

THREAT-MODEL.md 2026-11-13; the covered code 2026-11-152 days behind

git log at the root: the latest commit to any params.threat_model file against the latest to params.threat_model_covers read 2026-11-16 21:12 UTC

Goal: Microsoft SDL practices (read 2026-11-10): practice 3, perform security design review and threat modeling; its threat modeling page: part of your routine development lifecycle; template goal: no change to the covered code since the model's last update

Claude Code0

Merged PRs Claude security-reviewed0%goal ≥ 100

Share of recent merged pull requests where Claude's app or bot left a security review; anyone else's never counts.

Gap: 100 pts under goal

with a security review from Claude0
merged PRs read20

gh api graphql: the last 20 merged pullRequests: a head check run from an app in params.review_apps named like params.security_review, or a review or comment by a Bot whose login is in params.review_authors and whose first line starts with it read 2026-11-16 21:13 UTC

Goal: Anthropic, Claude Code security and GitHub Actions docs: /security-review, or the Claude Code action on a subscription token; template goal: every merge reviewed

Speed

No alert of any kind stays open more than a week.

Fix time–

Latest fixed alert's days open–goal ≤ 7

How long the most recently fixed alert stayed open, in days; lower is better.

needs an admin token; refreshed by a --local collect run

gh api dependabot/alerts (fixed_at), and code-scanning/alerts (fixed_at) and secret-scanning/alerts (resolved_at, revoked) while those features are on: the latest fixed alert, days from created_at

Goal: template goal (2026-11-03): an alert is fixed within a week, near 15 days; CISA's BOD 26-04 (June 10, 2026) ranks fixes by risk for federal agencies; its deadlines are in a table this read could not parse (2026-11-10), so the week is the template's own

Oldest open alert–goal ≤ 7

How long the longest-waiting open alert or finding has been open, in days; lower is better.

needs an admin token; refreshed by a --local collect run

gh api dependabot/alerts, and code-scanning/alerts and secret-scanning/alerts while those features are on, and code-quality/findings unless it answers that code quality is off: the oldest open one's age at read time

Goal: template goal: no alert waits more than a week, near 15 days; CISA's BOD 26-04 (June 10, 2026) ranks fixes by risk for federal agencies; its deadlines are in a table this read could not parse (2026-11-10), so the week is the template's own

Consumption

No paid security feature bills unless the organization bought it.

Cost–

Security features billed a month–goal ≤ 0

Dollars a month the organization is billed now for GitHub's paid security features; free is the goal.

needs an admin token; refreshed by a --local collect run

gh api orgs/{owner}/settings/billing/advanced-security (active committers billed, per product) times the list price: Code Security $30, Secret Protection $19 per active committer a month (github.com/security/plans, read 2026-11-03)

Goal: template goal: every security feature at no cost, as GitHub offers code scanning and secret scanning free on a public repo (github.com/security/plans, read 2026-11-03); near is one committer's Secret Protection

Dependabot

Quality

90

5 of 12 measured

Speed

100

4 of 6 measured

Consumption

100

1 of 1 measured

Ideal state

Dependabot alerts, malware alerts and security updates are on; the dependency graph and review cover every manifest; updates wait their cooldown, then merge within days.

PartIdealSet up
Dependabot alertsVulnerability and malware alerts on; no malware alert open; no vulnerability alert older than a weekEnable both in the organization's security configuration; malware alerts are off until asked for
Dependency updatesSecurity fixes arrive grouped; a new release waits 14 days; packages come from one trusted registryIn dependabot.yml: a cooldown, and a group that applies to security updates
6 sources, newest 2026-11-13
  1. GitHub Docs: Dependabot alerts 2026-11-13
  2. GitHub Docs: Dependabot malware alerts 2026-11-13
  3. GitHub Docs: Dependabot security updates 2026-11-13
  4. GitHub Docs: Dependabot version updates 2026-11-13
  5. GitHub Docs: Dependency graph 2026-11-13
  6. GitHub Docs: Dependency review 2026-11-13

Worst gaps

  1. 1Actions pinned to the latest allowed87.5%12.5 pts under goal

Quality

No open alert or malware, every dependency and action on its newest allowed release, packages only from trusted registries.

GitHub features–

Dependabot alerts–goal ≥ 1

Whether GitHub warns when a library the repo uses has a known vulnerability; on is better.

needs an admin token; refreshed by a --local collect run

gh api graphql: repository.hasVulnerabilityAlertsEnabled

Goal: GitHub docs, About Dependabot alerts

Open findings100

Open Dependabot vulnerabilities0 alertsgoal ≤ 0

Known vulnerabilities in the libraries the repo uses that are not fixed yet, by severity; fewer is better.

critical0 alerts
high0 alerts
medium0 alerts
low0 alerts

gh api repos/Colorado-Startups/brand/dependabot/alerts --paginate (open alerts that are vulnerabilities, by security_advisory.severity; malware left out) read 2026-11-16 21:12 UTC

Goal: GitHub docs, About Dependabot alerts: each alert is a known vulnerability in a dependency; template goal: none left open

Open malware alerts0 alertsgoal ≤ 0

Libraries the repo uses that GitHub flags as malware; none is the only safe number.

gh api repos/Colorado-Startups/brand/dependabot/alerts --paginate (open alerts whose security_advisory.classification is malware) read 2026-11-16 21:12 UTC

Goal: GitHub docs, Dependabot malware alerts (read 2026-11-10): an alert means a dependency is a known malicious package; template goal: none open

Supply chain100

Dependabot updates with a cooldown100%goal ≥ 100

Share of Dependabot update feeds that wait before taking a new release, so a bad one is caught first; higher is better.

github-actions /14 days

.github/dependabot.yml: updates[].cooldown, the shortest of default-days and semver-*-days, against params.cooldown_days read 2026-11-16 21:12 UTC

Goal: GitHub docs, Dependabot options reference: cooldown; template goal: at least params.cooldown_days (14), long enough for a malicious release to be reported and pulled

Lockfile packages from the registry–goal ≥ 100

Share of installed packages that resolve from the official registry, not another source; higher is better.

no package-lock.json with a packages map at the root

package-lock.json at the root: each package's resolved URL against params.registries, parsed: https, the registry's host, under its path

Goal: OpenSSF S2C2F, AUD-2: audit that developers consume OSS through the approved ingestion method; ENF-1: securely configure package source files

Lockfile packages younger than the cooldown–goal ≤ 0

Installed packages, deep ones too, published too recently to trust yet; Dependabot's cooldown covers only direct ones.

no package-lock.json with a packages map at the root

package-lock.json at the root against itself params.cooldown_days ago (git); for each entry added since on an allowed registry, a HEAD of its download for Last-Modified, then the registry's full document only where that is recent; no redirects, at most params.max_registry_reads requests

Goal: template goal: no package younger than params.cooldown_days (14), the same wait as security.dependabot_cooldown, applied to every lockfile entry

Repo–

Dependabot security updates on–goal ≥ 1

Whether GitHub automatically proposes fixes for security holes in our libraries; on is better.

needs an admin token; refreshed by a --local collect run

gh api repos/Colorado-Startups/brand (security_and_analysis.dependabot_security_updates)

Goal: GitHub docs, About Dependabot security updates

Versions75

Actions pinned to the latest allowed87.5%goal ≥ 100

Third-party actions whose pinned SHA is the commit of their newest release old enough to take.

Gap: 12.5 pts under goal

ci-standard.yml‎SoulCraft-Org/git-crafter/.github/workflows/‎0
checkout‎actions/‎1
claude-code-action‎anthropics/‎0
setup-node‎actions/‎1
create-github-app-token‎actions/‎1
pages‎SoulCraft-Org/git-crafter/actions/‎0
upload-pages-artifact‎actions/‎1
deploy-pages‎actions/‎1
scorecard-action‎ossf/‎1
upload-sarif‎github/codeql-action/‎1

uses: pins in .github/workflows against gh api repos/<owner>/<repo>/releases (the newest stable release at least 14 days old (params.release_age_days)) and the release tag's commit; actions owned by Colorado-Startups (params.own_action_owners, default the repo's owner) are left out, as they have no releases; 8 of 10 read read 2026-11-16 21:12 UTC

Goal: template goal, no public benchmark: every third-party action on its newest allowed release, pinned to a full commit SHA as GitHub's security hardening guide recommends (docs.github.com/en/actions/security-for-github-actions/security-guides/security-hardening-for-github-actions)

CI's Node is the current LTS1goal ≥ 1

1 when every setup-node step in the workflows asks for the newest LTS line; with none, the .nvmrc or .node-version.

pages.yml:47: node-version 2626
Current LTS line26

each actions/setup-node step's node-version or node-version-file in .github/workflows (1) against https://nodejs.org/dist/index.json, the newest line with an LTS codename read 2026-11-16 21:12 UTC

Goal: goal from Node.js, which recommends LTS releases for production (nodejs.org/en/about/previous-releases)

Direct dependencies on the latest allowed–goal ≥ 100

Direct dependencies on the newest stable release old enough to take; one installed sooner is too new.

package.json or its lockfile unreadable: ENOENT: no such file or directory, open '/home/runner/work/brand/brand/package.json'

package.json (root and workspaces) and its lockfile against npm view <name> time --json; eligible is the newest stable release at least 14 days old (params.release_age_days)

Goal: template goal, no public benchmark: every direct dependency on its newest stable release once that release is params.release_age_days old (default 14), so a broken or compromised release has time to be pulled first

Dependencies a major behind–goal ≤ 0

Direct dependencies whose newest allowed release is a newer major, less a major a Dependabot ignore holds.

package.json or its lockfile unreadable: ENOENT: no such file or directory, open '/home/runner/work/brand/brand/package.json'

package.json (root and workspaces) and its lockfile against npm view <name> time --json; eligible is the newest stable release at least 14 days old (params.release_age_days)

Goal: template goal, no public benchmark: no direct dependency a major release behind its newest allowed one; near band: template default

Adoption–

Adopting repos within the cooldown of the newest tag–goal ≥ 100

Adopters whose every pin of this repo is at or above its newest tag at least params.adopter_cooldown_days old.

no adopting repos: list them as owner/name in params.adopters

each repo in params.adopters: its .github/workflows uses: pins of Colorado-Startups/brand (gh api contents) against this repo's newest tag at least 7 days old (params.adopter_cooldown_days, default 7: one weekly Dependabot run), aged by its commit's date (gh api tags and git/commits)

Goal: template goal, no public benchmark: set 2026-11-04 (#236), redefined 2026-11-11: every adopter at or above the newest tag that is params.adopter_cooldown_days old. The grace is one weekly Dependabot run, the github-actions schedule templates/ci/dependabot.yml ships, so the default is 7 days. The newest tag itself is a bar no weekly run holds, so that goal could not be met. Dependabot's weekly github-actions run moves a SHA pin with a version comment

Speed

An allowed update waits days, not weeks, and its pull request merges within the week.

Queue100

Open Dependabot PRs0 PRsgoal ≤ 2

Automatic proposals to update our libraries that nobody has accepted yet; fewer is better.

gh api graphql: open pullRequests (author dependabot, labels) read 2026-11-16 21:12 UTC

Goal: template goal: dependency updates merge as they arrive

Waits100

Dependabot updates grouped100%goal ≥ 100

Share of automatic update feeds that bundle their changes into fewer proposals; higher is better.

github-actions1

.github/dependabot.yml (updates[].groups) read 2026-11-16 21:12 UTC

Goal: GitHub docs, Optimizing the creation of pull requests for Dependabot version updates: groups combine updates into fewer pull requests

Updates100

Open dependency PRs0 PRsgoal ≤ 1

Pull requests labeled as dependency updates that are open now.

gh pr list -R Colorado-Startups/brand --label dependencies (params.dependency_label; Dependabot's default label) --state open read 2026-11-16 21:12 UTC

Goal: template goal, no public benchmark: one grouped update PR at a time; near is Dependabot's default open-pull-requests-limit, five ("If five pull requests with version updates are open, no further pull requests are raised", docs.github.com/en/code-security/dependabot/working-with-dependabot/dependabot-options-reference, read 2026-11-03)

Oldest open dependency PR0 daysgoal ≤ 7

Days the oldest open dependency PR has waited; 0 when none is open.

gh pr list -R Colorado-Startups/brand --label dependencies (params.dependency_label; Dependabot's default label) --state open; days since the oldest opened, 0 when none is open read 2026-11-16 21:12 UTC

Goal: template goal, no public benchmark: a weekly update PR merges within its week; near: two weeks

Oldest update waiting–goal ≤ 7

Days since the oldest pending update to a direct dependency became old enough to take; 0 when none waits.

package.json or its lockfile unreadable: ENOENT: no such file or directory, open '/home/runner/work/brand/brand/package.json'

package.json (root and workspaces) and its lockfile against npm view <name> time --json; eligible is the newest stable release at least 14 days old (params.release_age_days)

Goal: template goal, no public benchmark: a weekly update run lands an allowed update within a week; near: two weeks

Install–

Install step time, latest main run–goal ≤ 60

Seconds the slowest dependency install step took in the newest successful push run on main.

none of the newest 5 successful push runs has an install step

latest: the newest successful push run of any workflow on main with an install step (npm ci, pnpm install or yarn install), the slowest such step (gh api actions/runs and runs/{id}/jobs)

Goal: template goal, no public benchmark: an install with the package cache restored (setup-node's cache input) should take under a minute

Consumption

Update pull requests and installs bill few CI minutes.

Use100

CI minutes on the latest dependency PR5 mingoal ≤ 60

Billed Actions minutes on the newest merged dependency PR, read now.

#15 min

the last 10 merged PRs labeled dependencies: gh api actions/runs?head_sha=<PR head>&event=pull_request and runs/{id}/jobs, each job rounded up to the whole minute as GitHub bills it (docs.github.com/en/billing/managing-billing-for-github-actions/about-billing-for-github-actions); N = 10 (params.last_n); the latest, #1, merged 2026-11-14, and the rest beside it read 2026-11-16 21:12 UTC

Goal: template goal, no public benchmark: set 2026-11-04; an update PR runs the repo's CI once or twice; near band: template default

Scanning

Quality

–

0 of 4 measured

Speed

–

0 of 0 measured

Consumption

100

1 of 1 measured

Ideal state

Code scanning default setup and secret scanning with push protection are on; no alert stays open, and the scans run in minutes.

PartIdealSet up
Secret scanningBought: scanning and push protection on. Free: a secret scan that must pass to mergeEnable Secret Protection, or require a pinned scanner's check in a ruleset
Code scanningCode scanning on where the repo is public or Code Security is bought; never billed unaskedEnable CodeQL default setup in the repo security settings
4 sources, newest 2026-11-13
  1. GitHub Docs: Code scanning 2026-11-13
  2. GitHub Docs: Configuring default setup for code scanning 2026-11-13
  3. GitHub Docs: Secret scanning 2026-11-13
  4. GitHub Docs: Push protection 2026-11-13

Quality

Both scans on, and no open code scanning or secret scanning alert.

GitHub features–

Code scanning–goal ≥ 1

Whether GitHub scans the repo's own code for vulnerabilities and errors; on is better, even where it costs.

needs an admin token; refreshed by a --local collect run

gh api repos/Colorado-Startups/brand/code-scanning/default-setup (state) and code-scanning/analyses

Goal: GitHub docs, About code scanning

Secret scanning and push protection–goal ≥ 1

Whether GitHub finds secrets in the repo and blocks pushes that hold one; both on is better, even where it costs.

needs an admin token; refreshed by a --local collect run

gh api repos/Colorado-Startups/brand (security_and_analysis: secret_scanning and secret_scanning_push_protection)

Goal: GitHub docs, About secret scanning, and About push protection

Open findings–

Open code scanning alerts–goal ≤ 0

Problems code scanning found in the repo's own code that are not fixed yet; not counted while it is off.

needs an admin token; refreshed by a --local collect run

gh api repos/Colorado-Startups/brand/code-scanning/alerts --paginate (open alerts, by rule.security_severity_level or rule.severity)

Goal: GitHub docs, About code scanning alerts; template goal: none open

Open secret scanning alerts–goal ≤ 0

Secrets found in the repo that are not revoked yet; not counted while secret scanning is off.

needs an admin token; refreshed by a --local collect run

gh api repos/Colorado-Startups/brand/secret-scanning/alerts --paginate (open alerts, by secret_type_display_name)

Goal: GitHub docs, About secret scanning: a leaked secret should be revoked; template goal: none open

Speed

No speed metric yet.

Consumption

The latest scans bill few runner minutes.

Cost100

Runner minutes, the latest security scans4 minutesgoal ≤ 5

Billed runner minutes the latest run of each security scan workflow took; the scans still run on every change

CodeQL4 minutes

latest: gh api actions/runs/{id}/jobs: each workflow named like params.security_workflows, the repo's own (its latest run that did real work, of its newest 10 of the last 20 runs per workflow) and GitHub's dynamic ones such as CodeQL default setup (its latest passed run), less CodeQL - Code Quality; billed runner minutes, summed read 2026-11-16 21:13 UTC

Goal: template goal (2026-11-03): the scans on one change bill at most 5 runner minutes, half the ten-minute build (Kent Beck, Extreme Programming Explained, 2nd edition, 2004)

Insights

Quality

100

2 of 3 measured

Speed

100

1 of 1 measured

Consumption

100

1 of 1 measured

Ideal state

A healthy repo shows it in Insights: every health file present, no finished branch left behind, and few runner minutes wasted.

PartIdealSet up
Pulse, Commits, Code frequencyWork lands as small merged pull requests, few stay open, and issues close as fast as they openScored on the Pull requests, Issues and Code tabs; the code frequency API refuses a repo of 10,000 commits
ContributorsEvery default-branch commit is credited to an account; a count of activity is no measure of a personLink each commit email to an account; Claude Code contribution metrics credit agent work once turned on. Not scored: it rarely moves
CommunityThe health checklist is completeAdd a README, license, contributing guide, code of conduct, issue forms and a pull request template
Traffic and ForksNot scored: no documented practice sets a goal for views, clones or forksTraffic keeps 14 days and needs an admin token; its clones mostly count workflow checkouts
Dependency graphEvery manifest parses, so every dependency is knownKeep the graph on and commit lock files in a format it reads. Not scored: it rarely moves, and Security scores the alerts
NetworkOnly the default branch and live work remain; a leftover branch is gone within a dayTurn on automatic branch deletion; delete the branch of a pull request closed without merging; open a pull request within a day
PeopleEveryone with write access has committed within 90 daysReview the people list each quarter and remove access nobody uses
Actions metricsUnder a tenth of billed minutes buys nothing; a check failing a pull request is not wasteGitHub offers no API for these two views; both are computed from workflow runs and jobs
16 sources, newest 2026-11-10
  1. GitHub Docs: About repository graphs 2026-11-10
  2. GitHub Docs: About community profiles for public repositories 2026-11-10
  3. GitHub Docs: Viewing a project's contributors 2026-11-10
  4. GitHub Docs: Dependency graph 2026-11-10
  5. GitHub Docs: Managing the automatic deletion of branches 2026-11-10
  6. GitHub Docs: Viewing people with access to your repository 2026-11-10
  7. Microsoft Learn: Microsoft cloud security benchmark, Privileged Access (PA-4) 2026-11-10
  8. GitHub Docs: About GitHub Actions metrics 2026-11-10
  9. GitHub Docs: REST API endpoints for repository traffic 2026-11-10
  10. DORA: DORA's software delivery performance metrics 2026-11-10
  11. Claude Code Docs: Claude Code GitHub Actions 2026-11-10
  12. Anthropic: Measuring AI agent autonomy in practice 2026-11-10
  13. Microsoft Research: The SPACE of Developer Productivity: There's more to it than you think 2021-02-01
  14. GitHub Docs: REST API endpoints for repository statistics 2026-11-10
  15. Claude Code Docs: Track team usage with analytics 2026-11-10
  16. GitHub: Engineering System Success Playbook 2025-05-01

Quality

The health checklist is complete, no branch outlives its pull request, and access nobody uses is removed.

Community100

Community health checklist100%goal ≥ 100

GitHub's own score for the health files a repo should have: README, license, contributing, code of conduct, templates

Description1
README1
License1
Contributing guide1
Code of conduct1
Pull request template1

state: gh api repos/Colorado-Startups/brand/community/profile, GitHub's health percentage and the files it found read 2026-11-16 21:12 UTC

Goal: GitHub docs, About community profiles for public repositories and REST community metrics (read 2026-11-10): the checklist names the recommended files and the API scores how many are present; goal: all of them

Network100

Leftover branches0 branchesgoal ≤ 0

Branches whose pull request closed or merged, or that are over a day old with none: finished or abandoned work

pull request merged or closed0
no pull request for over a day0

state: GraphQL refs under refs/heads (the first 1000) with their pull requests: a branch whose pull request closed or merged, or one older than params.branch_grace_hours with none; the default branch, open pull requests and params.keep_branches do not count read 2026-11-16 21:12 UTC

Goal: GitHub docs, Managing the automatic deletion of branches (read 2026-11-10): GitHub can delete a head branch once its pull request merges; template goal: no branch outlives its pull request, near 5; a branch under a day old with no pull request is live work

People–

People with unused write access–goal ≤ 0

People who can push but have landed no commit on the default branch in 90 days; access nobody uses is only risk

needs an admin token; refreshed by a --local collect run

state: gh api repos/Colorado-Startups/brand/collaborators (people who can push, the first 100) and each one's latest default-branch commit (gh api repos/Colorado-Startups/brand/commits?author=): those with none in params.idle_days days

Goal: Microsoft Learn, Microsoft cloud security benchmark PA-4 (read 2026-11-10): review access regularly and remove what goes unused, with accounts unused for 90 days flagged; GitHub docs, Viewing people with access to your repository: the list serves offboarding and security reviews

Speed

A leftover branch is gone within a day of its pull request closing.

Network100

Oldest leftover branch0 hoursgoal ≤ 24

Hours the oldest leftover branch has waited since its pull request closed, or since its last commit; 0 when none

leftover branches0

state: GraphQL refs under refs/heads (the first 1000) with their pull requests: a branch whose pull request closed or merged, or one older than params.branch_grace_hours with none; the default branch, open pull requests and params.keep_branches do not count; each timed from its pull request's close, or from its last commit read 2026-11-16 21:12 UTC

Goal: template goal (2026-11-10): a finished branch is gone within a day, near three; GitHub docs, Managing the automatic deletion of branches: a merged pull request's branch can go at once, so only closed and never-proposed ones wait

Consumption

At most a tenth of billed runner minutes goes to runs that are cancelled, time out or fail after merge.

Actions usage100

Minutes that bought nothing0.7%goal ≤ 10

Share of billed minutes on runs cancelled, timed out or failed after merge; a failed pull request check is not waste

ci0%
claude-review0%
pages0%
scorecard4.8%

GraphQL check runs of the last 20 runs per workflow: each job's minutes rounded up as GitHub bills them, summed for runs that ended cancelled, timed_out, startup_failure, or failed outside a pull request or merge queue, over all read 2026-11-16 21:13 UTC

Goal: template goal (2026-11-10): no first-party source sets a number; a tenth, near a fifth. GitHub docs, About GitHub Actions metrics (read 2026-11-10) reports failure rate and sets none. A failure on a pull request is the check working, so only cancelled, timed-out and post-merge failures count; a merge queue failure is the check working too

Settings

Quality

–

0 of 7 measured

Speed

–

0 of 2 measured

Consumption

–

0 of 1 measured

Ideal state

Settings hold their documented best values: rules no one can bypass, least access, no unused credential, no wait for a person, nothing billed unasked.

PartIdealSet up
GeneralOne merge method, auto-merge off, merged branches deleted, no wiki, forking off for a private repoSet them once on the General page, then name the values in the repo's targets so drift is counted
AccessMembers only via the org: no outside collaborator, invite or custom role. A public repo gates fork runs, takes private reports, blocks secrets, has DiscussionsOrg base permission read, two-factor on; invite to the org, not a repo; public repo settings: templates/open/README.md
Rules, branches and tagsRulesets: a pull request and a passing check to merge, no force push, deletion or bypass, tags fixed; custom properties only to target rulesetsAn organization ruleset for every default branch; repo rulesets for checks, tags and secret file names
Actions and webhooksOnly selected actions, pinned to a commit, a read-only token, no fork runs; each webhook has a secret and checks SSLActions page: policies and workflow permissions, scored on the Security and Actions tabs; add a webhook only with a secret
Copilot, Planning, CodespacesCopilot review and Codespaces off unless bought: both bill by use; issue types and fields set once for the organizationLeave both off; planning lives in organization settings and is scored on the Issues and Projects tabs
Environments and PagesEach environment deploys from named branches only, with no reviewer or timer; Pages is private, built by a workflow, over HTTPSEnvironment page: selected branches and tags; Pages page: source GitHub Actions, visibility private
Security and secretsFree features on: dependency graph, Dependabot alerts and security updates; no deploy key; every stored secret read by a workflowOne organization security configuration; paid scanning only when bought; delete a secret when its workflow goes
Apps and autolinksOnly apps the organization uses, each with the permissions it needs; autolinks only for a tracker outside GitHubReview installed apps on the organization's GitHub Apps page; add one autolink per outside ticket prefix
25 sources, newest 2026-11-12
  1. GitHub docs: About rulesets 2026-11-10
  2. GitHub docs: Available rules for rulesets 2026-11-10
  3. GitHub docs: REST API endpoints for rules 2026-11-10
  4. GitHub docs: REST API endpoints for repositories, Get a repository 2026-11-10
  5. GitHub docs: About merge methods on GitHub 2026-11-10
  6. GitHub docs: Managing the automatic deletion of branches 2026-11-10
  7. GitHub docs: Setting base permissions for an organization 2026-11-10
  8. GitHub docs: Requiring two-factor authentication in your organization 2026-11-10
  9. GitHub docs: Managing custom properties for repositories in your organization 2026-11-10
  10. GitHub docs: Managing GitHub Actions settings for a repository 2026-11-10
  11. GitHub docs: Secure use reference, audit and rotate registered secrets 2026-11-10
  12. GitHub docs: Deployments and environments 2026-11-10
  13. GitHub docs: Best practices for using webhooks 2026-11-10
  14. GitHub docs: Managing deploy keys 2026-11-10
  15. GitHub docs: About GitHub Advanced Security, what a private repo pays for 2026-11-10
  16. GitHub docs: GitHub Actions billing 2026-11-10
  17. GitHub changelog: Copilot code review will start consuming GitHub Actions minutes 2026-04-27
  18. Microsoft Learn: Secure the developer environment for Zero Trust 2026-03-26
  19. Microsoft Azure Verified Modules: TFNFR3, GitHub repo branch protection 2025-11-19
  20. Anthropic: Claude Code GitHub Actions, the Claude GitHub App and its permissions 2026-11-10
  21. Anthropic: Claude Code, Code Review 2026-10-24
  22. Anthropic: Use Claude Code in the cloud 2026-11-10
  23. OpenSSF Scorecard: checks 2026-10-28
  24. GitHub Docs: Approving workflow runs from forks 2026-11-12
  25. GitHub Docs: Configuring private vulnerability reporting for a repository 2026-11-12

Quality

Branch, tags and deploys are protected with no bypass, secret files cannot be pushed, access stays inside the organization, and every secret is used.

Rules–

Default branch rules–goal ≥ 1

Whether the default branch requires a pull request and blocks force pushes and deletion; on is better.

needs an admin token; refreshed by a --local collect run

gh api repos/Colorado-Startups/brand/branches/{branch}/protection and repos/Colorado-Startups/brand/rules/branches/{branch}, {branch} the default branch: a pull request required, force pushes blocked, deletion blocked

Goal: GitHub docs, Available rules for rulesets: Require a pull request before merging, Block force pushes, Restrict deletions

Required status checks–goal ≥ 1

How many automatic checks must pass before a change can join the default branch; more is better.

needs an admin token; refreshed by a --local collect run

gh api repos/Colorado-Startups/brand/branches/{branch}/protection and repos/Colorado-Startups/brand/rules/branches/{branch} (required_status_checks), {branch} the default branch

Goal: GitHub docs, About protected branches: Require status checks before merging; one required check at least

Protections missing–goal ≤ 0

Guards that are off: rules with no bypass, fixed tags, a push rule on secret file names, deploys from named branches.

needs an admin token; refreshed by a --local collect run

state: gh api repos/Colorado-Startups/brand/rulesets and each active one (bypass_actors, tag rules deletion and update, push rule file_path_restriction against params.blocked_paths), branches/the default branch/protection (enforce_admins), tags (the first listed), environments (deployment_branch_policy); a row with nothing to guard is left out

Goal: GitHub docs, About rulesets (bypass lists), Available rules for rulesets (Restrict updates, Restrict deletions, Restrict file paths) and Deployments and environments (deployment branches and tags), read 2026-11-10; goal: none missing

Repo–

General settings off standard–goal ≤ 0

Settings on the General page that differ from the values this repo names as its standard.

needs an admin token; refreshed by a --local collect run

state: gh api repos/Colorado-Startups/brand, each field named in params.expected against its value there

Goal: template goal: a repo names the General settings it holds itself to and none drifts; GitHub docs, REST API endpoints for repositories, Get a repository, lists the fields (read 2026-11-10)

Access–

Outside access–goal ≤ 0

People outside the organization who can reach the repo, plus invitations not yet accepted.

needs an admin token; refreshed by a --local collect run

state: gh api repos/Colorado-Startups/brand/collaborators?affiliation=outside and repos/Colorado-Startups/brand/invitations, counted

Goal: template goal: access comes through organization membership, so no one holds it from outside and no invitation stays open; a repo with outside collaborators sets its own goal

Secrets no workflow reads–goal ≤ 0

Stored Actions secrets whose name no workflow file uses; each is a credential kept for nothing.

needs an admin token; refreshed by a --local collect run

state: gh api repos/Colorado-Startups/brand/actions/secrets and each environment's secrets (names only), against secrets.NAME in .github/workflows/*.yml of the checkout

Goal: GitHub docs, Secure use reference, Audit and rotate registered secrets (read 2026-11-10): review the registered secrets and remove those no longer needed

Open-repo settings off–goal ≤ 0

Settings a public repo needs: fork run approval, private vulnerability reporting, push protection, Discussions.

needs an admin token; refreshed by a --local collect run

state: gh api repos/Colorado-Startups/brand (has_discussions, security_and_analysis.secret_scanning_push_protection), repos/Colorado-Startups/brand/actions/permissions/fork-pr-contributor-approval (approval_policy) and repos/Colorado-Startups/brand/private-vulnerability-reporting (enabled); not measured for a private repo

Goal: template goal: each is one API call and protects the repo from outside code or a public leak; GitHub docs, Approving workflow runs from forks, Configuring private vulnerability reporting, About push protection and Enabling GitHub Discussions (read 2026-11-12)

Speed

No rule waits for a person or a timer, and dependency updates arrive grouped.

Waits–

Rules that wait for a person–goal ≤ 0

Merge and deploy rules that need a person's approval or a timer before work can land.

needs an admin token; refreshed by a --local collect run

state: pull_request rules of the active rulesets on the default branch and of branches/the default branch/protection (approvals, code owner review, last-push approval, required reviewers), and each environment's protection_rules (required_reviewers, wait_timer)

Goal: template goal: no person reviews, so a rule that waits for one stops every merge or deploy; a repo with human reviewers drops this metric. GitHub docs, Available rules for rulesets and Deployments and environments, name the rules (read 2026-11-10)

Age of the admin readings–goal ≤ 24

Hours since a signed-in run last read the settings an automatic run cannot; an old reading can hide a changed setting.

no signed-in run is recorded yet: run collect.mjs --local

state: generated_at of local.json in the repo's data folder, the last collect.mjs --local run, against the time of this run

Goal: template goal, no public benchmark: a setting changed today shows on the page by tomorrow, so the signed-in run that reads these settings is at most a day old

Consumption

No setting bills past what the plan includes unless it was bought on purpose.

Cost–

Settings that bill past the plan–goal ≤ 0

Settings switched on that add charges beyond what the plan includes.

needs an admin token; refreshed by a --local collect run

state: gh api repos/Colorado-Startups/brand/actions/cache/storage-limit (max_cache_size_gb against params.included_cache_gb) and the active rulesets' copilot_code_review rules; not measured for a public repo

Goal: GitHub docs, GitHub Actions billing (read 2026-11-10): cache storage past the included 10 GB a repo is billed; GitHub changelog, 2026-04-27: Copilot code review on a private repo uses Actions minutes from 2026-06-01; goal: none on unless bought

Pages

Quality

50

5 of 17 measured

Speed

100

3 of 8 measured

Consumption

100

1 of 3 measured

Ideal state

The site builds and deploys from a workflow on every merge, rarely fails, restores within the hour, and stays fast, readable and accessible.

4 sources, newest 2026-11-13
  1. GitHub Docs: What is GitHub Pages? 2026-11-13
  2. GitHub Docs: Using custom workflows with GitHub Pages 2026-11-13
  3. GitHub Docs: Deployments and environments 2026-11-13
  4. GitHub Docs: Securing your GitHub Pages site with HTTPS 2026-11-13

Worst gaps

  1. 1Raw hex outside tokens31 values31 values over goal
  2. 2Production security headers1 of 43 of 4 under goal
  3. 3Raw px outside tokens31 values31 values over goal

Failed deploys, last 10

Merges each deploy shipped, newest first
2026-11-16 15:461 merges
2026-11-16 15:371 merges
2026-11-16 03:271 merges
2026-11-15 23:451 merges
2026-11-15 21:391 merges
2026-11-15 21:341 merges
2026-11-15 21:261 merges
2026-11-15 21:231 merges
2026-11-15 17:590 merges
2026-11-15 08:001 merges
Deployed (UTC)CommitCommit known byMerges shippedMedian lead timeResultRestored after
2026-11-16 15:46581e7e9–10 hno fix needed–
2026-11-16 15:375c4a0ef–10 hno fix needed–
2026-11-16 03:27a53911d–10 hno fix needed–
2026-11-15 23:45aed5571–10 hno fix needed–
2026-11-15 21:395d43eff–10 hno fix needed–
2026-11-15 21:349865c12–10 hno fix needed–
2026-11-15 21:2672686a4–10 hno fix needed–
2026-11-15 21:23a0cae04–10 hno fix needed–
2026-11-15 17:59530fe7c–0–no fix needed–
2026-11-15 08:00530fe7c–10 hno fix needed–

Quality

Pages draw only from design tokens, pass contrast and axe, and every deploy passes its checks.

Raw hex outside tokens31 valuesgoal ≤ 0

Colors typed straight into the code instead of taken from the brand palette; fewer is better.

Gap: 31 values over goal

guide.css‎guide/public/‎30 values
index.html‎guide/public/‎1 values

regex over guide/public (.css, .scss, .astro, .vue, .svelte, .html) with comments stripped; excludes guide/public/tokens.css: the token file: the State palette, decision 29 read 2026-11-16 21:12 UTC

Goal: Git Crafter default goal: every color comes from the design tokens (params.tokens), the one source of the palette. The Design Tokens Format Module 2025.10 shares tokens across disciplines, tools and technologies (designtokens.org/tr/2025.10/format, 2025-10-28, read 2026-11-10); zero is a Git Crafter default

Raw px outside tokens31 valuesgoal ≤ 0

Fixed sizes typed straight into the code instead of taken from the shared scale; fewer is better.

Gap: 31 values over goal

guide.css‎guide/public/‎31 values

regex over guide/public (.css, .scss, .astro, .vue, .svelte, .html) with comments stripped; excludes guide/public/tokens.css: the token file: the State palette, decision 29 read 2026-11-16 21:12 UTC

Goal: Git Crafter default goal: every size comes from the token scale. The Design Tokens Format Module 2025.10 shares tokens across disciplines, tools and technologies (designtokens.org/tr/2025.10/format, 2025-10-28, read 2026-11-10); zero is a Git Crafter default; the near band is wide, since a border or an icon size often has no token

Off-scale font sizes–goal ≤ 0

Different text sizes on the pages that are not in the brand's size scale; fewer is better.

no key page loaded; not measured: /@1280, /@375, /quality/@1280, /quality/@375 (HTTP 404)

Playwright (Chrome) on https://colorado-startups.github.io/brand, key pages at 1280 and 375 px

Goal: Git Crafter default goal: every text size is a step of the type scale in params.tokens. The Design Tokens Format Module 2025.10 shares tokens across disciplines, tools and technologies (designtokens.org/tr/2025.10/format, 2025-10-28, read 2026-11-10); zero is a Git Crafter default

Off-token text colors–goal ≤ 0

Different text colors on the pages that are not in the brand palette; fewer is better.

no key page loaded; not measured: /@1280, /@375, /quality/@1280, /quality/@375 (HTTP 404)

Playwright (Chrome) on https://colorado-startups.github.io/brand, key pages at 1280 and 375 px

Goal: Git Crafter default goal: every text color is a color token in params.tokens. The Design Tokens Format Module 2025.10 shares tokens across disciplines, tools and technologies (designtokens.org/tr/2025.10/format, 2025-10-28, read 2026-11-10); zero is a Git Crafter default

Contrast failures (axe)–goal ≤ 0

Page elements whose text is hard to read against its background; fewer is better.

no key page loaded; not measured: /@1280, /@375, /quality/@1280, /quality/@375 (HTTP 404)

Playwright (Chrome) on https://colorado-startups.github.io/brand, key pages at 1280 and 375 px

Goal: goal: WCAG 2.2 AA, Success Criterion 1.4.3 Contrast (Minimum), as axe-core's color-contrast rule tests it (w3.org/TR/WCAG22); the near band is a Git Crafter default

Reading grade, hardest page–goal ≤ 9

The school grade needed to read the hardest page; a lower grade is easier for everyone, so lower is better.

no key page loaded; not measured: /, /quality/ (HTTP 404)

Playwright (Chrome) on https://colorado-startups.github.io/brand, key pages at 1280 px

Goal: goal: WCAG 2.2 Success Criterion 3.1.5 Reading Level (AAA) asks for text a reader with lower secondary education reads, seven to nine years of school (w3.org/WAI/WCAG22/Understanding/reading-level, updated 2026-06-28, read 2026-11-10); grade 9 is its top. The near band is a Git Crafter default

Sentences over 30 words–goal ≤ 0

Sentences on the pages that run too long to read easily; fewer is better.

no key page loaded; not measured: /, /quality/ (HTTP 404)

Playwright (Chrome) on https://colorado-startups.github.io/brand, key pages at 1280 px

Goal: Git Crafter default goal: no sentence on a key page over params.long_sentence_words (30) words. GOV.UK splits sentences over 25 words (guidance.publishing.service.gov.uk, Use clear language, read 2026-11-10); 30 is the looser limit Git Crafter sets for its own writing

axe violations, not contrast–goal ≤ 0

Page elements an automatic accessibility scan flags as hard for some people to use; fewer is better.

no key page loaded; not measured: /@1280, /@375, /quality/@1280, /quality/@375 (HTTP 404)

Playwright (Chrome) on https://colorado-startups.github.io/brand

Goal: goal: WCAG 2.2 AA as axe-core's default rules test it (w3.org/TR/WCAG22); the near band is a Git Crafter default

Small tap targets, phone width–goal ≤ 0

Buttons and links on a phone that are too small to tap easily; fewer is better.

no key page loaded at 375 px; not measured: /, /quality/ (HTTP 404)

Playwright (Chrome) on https://colorado-startups.github.io/brand

Goal: goal: WCAG 2.2 AA, Success Criterion 2.5.8 Target Size (Minimum), 24 by 24 CSS px (w3.org/TR/WCAG22); the near band is a Git Crafter default

Unlabeled form fields–goal ≤ 0

Form fields with no label telling people what to type; fewer is better.

params.form_pages names no form page

params.form_pages

Goal: goal: WCAG 2.2 AA, Success Criteria 3.3.2 Labels or Instructions and 4.1.2 Name, Role, Value (w3.org/TR/WCAG22); the near band is a Git Crafter default

Main tasks done by keyboard alone–goal ≥ 100

Main tasks a visitor finishes by keyboard: Tab reaches each link on the way, and Enter follows it to the next page.

GET / answered HTTP 404

Playwright (Chrome) on https://colorado-startups.github.io/brand, 1280 px: each task's shortest route from / in the crawl, followed by Tab and Enter from a fresh load

Goal: goal: every task, as WCAG 2.2 Success Criterion 2.1.1 Keyboard (Level A) requires all functionality to be operable through a keyboard interface (w3.org/TR/WCAG22, W3C Recommendation 2024-12-12, read 2026-11-04); the near band is a Git Crafter default

Latest deploy100

Latest deploy failed0goal ≤ 0

1 when the latest deploy was rolled back or shipped a named regression; on Pages, when its deployment failed.

2026-11-16 15:46 581e7e90

GitHub Pages: gh api repos/Colorado-Startups/brand/deployments?environment=github-pages and each deploy's statuses; a deploy failed when its newest status is failure or error; restored at the next successful deploy; the last 10 ended deploys read 2026-11-16 21:12 UTC

Goal: goal from DORA's elite band (2024 Accelerate State of DevOps Report, p. 13, dora.dev/research/2024/dora-report), a 5% change fail rate, so the latest deploy should hold

Latest failure's time to restore–goal ≤ 1

Hours from the latest failed deploy to its rollback or fix (on Pages, the next successful one); counts until restored.

none of the last 10 deploys failed, so there is nothing to restore

GitHub Pages: gh api repos/Colorado-Startups/brand/deployments?environment=github-pages and each deploy's statuses; a deploy failed when its newest status is failure or error; restored at the next successful deploy; the last 10 ended deploys

Goal: goal from DORA's elite band (2024 Accelerate State of DevOps Report, p. 13, dora.dev/research/2024/dora-report), failed deployment recovery time less than one hour. Near is the high band, less than one day

Last 10 deploys100

Failed deploys, last 100%goal ≤ 5

Share of the last 10 deploys rolled back or shipping a regression; on Pages, the github-pages deployments that failed.

GitHub Pages: gh api repos/Colorado-Startups/brand/deployments?environment=github-pages and each deploy's statuses; a deploy failed when its newest status is failure or error; restored at the next successful deploy; the last 10 ended deploys read 2026-11-16 21:12 UTC

Goal: goal from DORA's elite band (2024 Accelerate State of DevOps Report, p. 13, dora.dev/research/2024/dora-report), a 5% change fail rate. Near is the high band, 20%

Site0

Production security headers1 of 4goal ≥ 4

How many standard safety settings the live site tells browsers to use; more is better.

Gap: 3 of 4 under goal

Content-Security-Policy0
Strict-Transport-Security1
X-Content-Type-Options0
Referrer-Policy0

HEAD https://colorado-startups.github.io/brand/ read 2026-11-16 21:12 UTC

Goal: OWASP Secure Headers Project: send Content-Security-Policy, Strict-Transport-Security, X-Content-Type-Options and Referrer-Policy

Repo–

Pages visibility private–goal ≥ 1

Whether this page can be seen only by members of the repo's organization; private is better.

a public repo publishes a public Pages site: private Pages needs a private repo on GitHub Enterprise Cloud

gh api repos/Colorado-Startups/brand/pages (public); repos/Colorado-Startups/brand (private, owner type) and orgs/{owner} (plan), as only GitHub Enterprise Cloud publishes private Pages

Goal: GitHub docs, Changing the visibility of your GitHub Pages site: a private or internal repo on GitHub Enterprise Cloud can publish privately; template goal: the page shows internal data, so only members see it

Speed

A merged change is live within hours, and pages paint fast on a phone.

LCP, mobile, slowest page–goal ≤ 2.5

How long the main content takes to appear on a phone, for the slowest page, in seconds; lower is better.

no key page measured; not measured: /, /quality/ (PageSpeed Insights: HTTP 429, the keyless PageSpeed Insights quota is spent (no further calls this run); Google's own numbers need a PageSpeed Insights API key; lab: HTTP 404)

PageSpeed Insights API v5, mobile (Google's Lighthouse lab, no key), then the lab fallback, our numbers and not Google's: Playwright on Chromium 157.0.8103.50, one cold load per page with Lighthouse's mobile settings applied through CDP (throttling.mobileSlow4G: 4x CPU slowdown, 562.5 ms request latency, 1474.56 Kbps down, 675 Kbps up; 412 x 823 px at 1.75x)

Goal: goal: a good Largest Contentful Paint is 2.5 s or less and a poor one over 4 s, per Core Web Vitals (web.dev/articles/vitals and web.dev/articles/lcp, read 2026-11-03)

CLS, mobile, worst page–goal ≤ 0.1

How much the worst page jumps around while loading on a phone; lower is better.

no key page measured; not measured: /, /quality/ (PageSpeed Insights: HTTP 429, the keyless PageSpeed Insights quota is spent (no further calls this run); Google's own numbers need a PageSpeed Insights API key; lab: HTTP 404)

PageSpeed Insights API v5, mobile (Google's Lighthouse lab, no key), then the lab fallback, our numbers and not Google's: Playwright on Chromium 157.0.8103.50, one cold load per page with Lighthouse's mobile settings applied through CDP (throttling.mobileSlow4G: 4x CPU slowdown, 562.5 ms request latency, 1474.56 Kbps down, 675 Kbps up; 412 x 823 px at 1.75x)

Goal: goal: a good Cumulative Layout Shift is 0.1 or less and a poor one over 0.25, per Core Web Vitals (web.dev/articles/vitals and web.dev/articles/cls, read 2026-11-03)

Clicks from home, longest task–goal ≤ 2

The most clicks from the home page that any main task needs; fewer is better.

GET / answered HTTP 404; check the site is up and the network reachable

GET crawl of https://colorado-startups.github.io/brand from /, same origin, at most 200 fetches, /api/ and /cdn-cgi/ skipped

Goal: Git Crafter default goal, no public benchmark: every main task within two clicks of the home page. Nielsen Norman Group finds no data behind a click limit and weighs clear labels and fast pages beside it (nngroup.com/articles/3-click-rule, 2019-08-11, read 2026-11-10), so this tab pairs it with keyboard_task_share and words_before_action

Words before first action–goal ≤ 50

Words a visitor reads before the first link or button, on the wordiest page; fewer is better.

no key page loaded; not measured: /, /quality/ (HTTP 404)

Playwright (Chrome) on https://colorado-startups.github.io/brand, key pages at 1280 px

Goal: Git Crafter default goal, no public benchmark: a visitor reaches something to click within fifty words

TBT, mobile, slowest page–goal ≤ 200

How long the slowest page freezes and ignores taps while loading on a phone; lower is better.

no key page measured; not measured: /, /quality/ (PageSpeed Insights: HTTP 429, the keyless PageSpeed Insights quota is spent (no further calls this run); Google's own numbers need a PageSpeed Insights API key; lab: HTTP 404)

PageSpeed Insights API v5, mobile (Google's Lighthouse lab, no key), then the lab fallback, our numbers and not Google's: Playwright on Chromium 157.0.8103.50, one cold load per page with Lighthouse's mobile settings applied through CDP (throttling.mobileSlow4G: 4x CPU slowdown, 562.5 ms request latency, 1474.56 Kbps down, 675 Kbps up; 412 x 823 px at 1.75x)

Goal: goal: a Total Blocking Time under 200 ms on average mobile hardware (web.dev, Total Blocking Time, updated 2024-11-04, read 2026-11-04); near is 500 ms, above which web.dev rates INP poor, as TBT is INP's lab stand-in (web.dev, Interaction to Next Paint, updated 2025-09-02, read 2026-11-04)

Now100

Merges waiting to deploy0 mergesgoal ≤ 0

Merges to main that change what deploys and that production does not hold right now.

GitHub Pages: gh api repos/Colorado-Startups/brand/deployments?environment=github-pages and each deploy's statuses; gh api repos/{repo}/compare/{live sha}...{branch}: merges production does not hold now, less the commit this run deploys and those before it read 2026-11-16 21:12 UTC

Goal: goal from DORA's elite band (2024 Accelerate State of DevOps Report, p. 13, dora.dev/research/2024/dora-report), deploys on demand, so none waits; near band: template default

Oldest waiting merge0 hgoal ≤ 24

Hours the oldest merge that production does not hold has waited; 0 when none waits.

GitHub Pages: gh api repos/Colorado-Startups/brand/deployments?environment=github-pages and each deploy's statuses; gh api repos/{repo}/compare/{live sha}...{branch}: merges production does not hold now, less the commit this run deploys and those before it; hours since the oldest was merged, 0 when none waits read 2026-11-16 21:12 UTC

Goal: goal from DORA's elite band (2024 Accelerate State of DevOps Report, p. 13, dora.dev/research/2024/dora-report), change lead time less than one day. Near is the high band's edge, one week

Latest deploy100

Latest deploy's time from merge to live0 hgoal ≤ 24

Hours from the earliest merge the latest deploy shipped to that deploy.

581e7e90 h

GitHub Pages: gh api repos/Colorado-Startups/brand/deployments?environment=github-pages and each deploy's statuses; hours from the earliest merge the latest shipping deploy carried to that deploy going live; the deploy of 2026-11-16 15:46 (581e7e9) read 2026-11-16 21:12 UTC

Goal: goal from DORA's elite band (2024 Accelerate State of DevOps Report, p. 13, dora.dev/research/2024/dora-report), change lead time less than one day. Near is the high band's edge, one week. DORA times from commit; this times from merge to main

Consumption

Pages stay light to download and deploys bill few minutes.

Total KiB, heaviest page–goal ≤ 1,600

How much data the heaviest page downloads in all; less loads faster, so lower is better.

no key page loaded; not measured: /@1280, /@375, /quality/@1280, /quality/@375 (HTTP 404)

Playwright (Chrome) on https://colorado-startups.github.io/brand, key pages at 1280 and 375 px

Goal: goal: Lighthouse's total-byte-weight guidance, aim below 1,600 KiB; Lighthouse flags pages over 5,000 KiB (developer.chrome.com/docs/lighthouse/performance/total-byte-weight, read 2026-11-02)

Form page KiB, heaviest form–goal ≤ 1,600

How much data a visitor downloads to open the heaviest form, where most tasks end; lower is better.

params.form_pages names no form page

params.form_pages

Goal: goal: Lighthouse total-byte-weight guidance, aim below 1,600 KiB; Lighthouse flags pages over 5,000 KiB (developer.chrome.com/docs/lighthouse/performance/total-byte-weight, read 2026-11-03), applied to the pages where tasks finish

Use100

CI minutes behind the latest deploy7 mingoal ≤ 20

Billed Actions minutes of the push to main the latest production deploy shipped, every workflow run on it.

ci4 min
scorecard1 min
pages2 min

latest: every workflow run the push to main of the latest successful Pages deploy's sha started (gh api actions/runs?head_sha=&event=push, then runs/{id}/jobs), each job rounded up to the whole minute as GitHub bills it; the deploy of 2026-11-16 15:46 (581e7e9) read 2026-11-16 21:12 UTC

Goal: template goal, no public benchmark: the CI of one push to main under 20 billed minutes; near: template default, twice that

Claude

Quality

–

0 of 4 measured

Speed

100

1 of 2 measured

Consumption

–

0 of 2 measured

Ideal state

Each session takes one scoped task, verifies its work before pushing, and opens a pull request that merges on first review and holds up.

2 sources, newest 2026-11-04
  1. Best practices for Claude Code 2026-11-04
  2. DORA's software delivery performance metrics 2026-01-05

Quality

PRs merge rather than close, pass every check on the first push, change a test with the code, and draw no bug report after merge.

Holds up–

Claude PRs merged–goal ≥ 90

Of the pull requests Claude finished, the share that merged rather than closed unused.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*-home-runner-work-brand-brand*/**/*.jsonl

Goal: template goal, no public benchmark: a closed PR is a session's work thrown away; a repo sets its own in quality/targets/claude.json

Green on first push–goal ≥ 90

How often every automatic check passed on the first version Claude pushed.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*-home-runner-work-brand-brand*/**/*.jsonl

Goal: template goal, no public benchmark: a red first push costs a fix round and a second CI run

Escaped bugs per 10 PRs–goal ≤ 0.5

Bug reports filed against the latest merged pull requests, per ten of them.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*-home-runner-work-brand-brand*/**/*.jsonl

Goal: goal from DORA's change failure rate, read as bugs per merged PR: the elite band's 5% is 0.5 per 10, and the high band's 20% is the near band, 2 per 10 (2024 Accelerate State of DevOps Report, p. 13, dora.dev/research/2024/dora-report)

Tests–

Code PRs that touch tests–goal ≥ 80

Of merged pull requests that change code, the share that also change a test.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*-home-runner-work-brand-brand*/**/*.jsonl

Goal: template goal, no public benchmark: most code changes should move a test; copy and config fixes need none, so the goal stays below every PR

Speed

A session opens its pull request within a couple of hours of starting.

Delivery100

Lead time, session to live0.06 hgoal ≤ 24

Hours from a session's start to its change live, each step's typical time added up.

PR open to all green0.02 h
all green to merge0.02 h
merge to live0.02 h

gh api graphql: the last 10 merged pullRequests timed end to end (N = 10, a template choice: a median that turns over within a day), each step's median in hours, summed; live is the first successful github-pages deployment created at or after the merge; session start from this Mac's transcripts, local runs only; 10 PRs read 2026-11-16 21:12 UTC

Goal: goal from DORA's elite band (2024 Accelerate State of DevOps Report, p. 13, dora.dev/research/2024/dora-report), change lead time less than one day. Near is the high band's edge, one week. DORA times "from committed to version control to deployed in production" (dora.dev/guides/dora-metrics, read 2026-11-10); this starts at the session, a little earlier, and in CI at the PR opening

Session start to PR open–goal ≤ 2

How long a session works before it opens its pull request.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*-home-runner-work-brand-brand*/**/*.jsonl

Goal: template goal, no public benchmark: a task scoped to land in one sitting

Consumption

Each merged pull request costs little at API prices and needs few review-fix rounds.

Use–

API-equivalent USD per merged PR–goal ≤ 10

What each merged pull request would cost at API prices; on a subscription plan nothing is billed per call.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*-home-runner-work-brand-brand*/**/*.jsonl

Goal: template goal, no public benchmark: the price check behind the token metrics

Review-fix rounds per merged PR–goal ≤ 0.5

Review rounds that found something to fix, for each merged pull request; each round costs a fix and a re-review.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*/**/*.jsonl

Goal: template goal, no public benchmark: one confirming review round plus at most half a round that finds something per PR

Claude settings

Quality

100

1 of 1 measured

Speed

–

0 of 1 measured

Consumption

–

0 of 1 measured

Ideal state

Claude works without stopping for permission: no prompt reaches a person, and every settings rule is narrow, valid and does what it says.

3 sources, newest 2026-11-04
  1. Settings files and precedence 2026-11-04
  2. Configure permissions 2026-11-04
  3. Permission modes 2026-11-04

Quality

Every settings file parses, uses known keys, and holds only rules Claude Code applies.

Quality100

Settings problems0 problemsgoal ≤ 0

Settings files that fail to parse, unknown keys, and permission rules Claude Code skips or ignores.

settings.json: no problem‎.claude/‎0

state: each file in params.files that exists, parsed as strict JSON; top-level and permissions keys against the settings index (code.claude.com/docs/en/settings-reference, read 2026-11-04); permissions.defaultMode against the modes (code.claude.com/docs/en/permission-modes); each allow, ask and deny rule against the rule syntax (code.claude.com/docs/en/permissions, read 2026-11-04; ruleProblem() in quality/collectors/claude/claude-settings.mjs); .claude/settings.json

Goal: template goal from the docs: Claude Code skips a broken file or entry with a Settings Error or Warning (code.claude.com/docs/en/settings, read 2026-11-04), so none should have one

Speed

No tool call waits for a person to approve it.

Speed–

Permission prompts per 100 tool calls–goal ≤ 0

Tool calls that stopped for a person to approve them; each is a wait and a defect.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

this repo's transcripts: the permissionDecision of each tool result, newest 1,000 that carry one; a prompt is a source starting user (user_temporary, user_permanent). ExitPlanMode and AskUserQuestion are left out: asking is their job. N = 1,000 tool calls, so new work replaces old within hours

Goal: template goal: work runs without prompts; auto and bypassPermissions modes exist so it can (code.claude.com/docs/en/permission-modes, read 2026-11-04)

Consumption

Hooks the settings declare add few tokens to a session.

Hooks–

Hook output tokens per session–goal ≤ 10,000

Text hooks add to a session, counted each time a later model call reads it again.

transcripts and the guard log live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*-home-runner-work-brand-brand*/**/*.jsonl

Goal: template goal, no public benchmark: hook text is read again by every later call until a compaction, so a session should carry little of it

Claude desktop app

Quality

–

0 of 6 measured

Speed

–

0 of 2 measured

Consumption

–

0 of 2 measured

Ideal state

Each session loads only the plugins, skills and connectors it uses, plans its work, delegates independent pieces in parallel, and gets short summaries back.

8 sources, newest 2026-11-10
  1. Desktop application 2026-11-10
  2. Measure plugin cost and usage 2026-11-10
  3. Plugin loading reference 2026-11-10
  4. Connect Claude Code to tools via MCP 2026-11-10
  5. Extend Claude with skills 2026-11-10
  6. Create custom subagents 2026-11-10
  7. Best practices for Claude Code 2026-11-10
  8. Claude Code changelog 2026-11-10

Quality

Every connector works, every plugin and connector loaded gets used, and each work turn keeps a task list and ends with a recap.

Connectors–

Connectors that fail or need sign-in–goal ≤ 0

Servers the session lists that failed to connect or still wait for a sign-in, so they give no tools.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

the desktop_app collector

Goal: goal from the MCP docs: a server that needs authentication or failed to connect gives no tools until it is fixed (code.claude.com/docs/en/mcp, read 2026-11-10), and the best practices say to turn off the servers not needed (code.claude.com/docs/en/best-practices, read 2026-11-10), so none stays broken

Use–

Plugins loaded and not used–goal ≤ 0

Plugins whose skills, agents and servers load into every session though no recent session called one.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

the desktop_app collector

Goal: goal from the plugin docs: an enabled plugin is part of every session, and Not used recently means unused for 10 sessions and 14 days (code.claude.com/docs/en/plugins/measure, read 2026-11-10); the session count is held here, a calendar window is not

Connectors loaded and not used–goal ≤ 0

Connected servers whose tool names and instructions load every session though no recent session called one.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

the desktop_app collector

Goal: goal from the best practices: connected servers add their tool names and instructions even when unused, so turn off the ones not needed (code.claude.com/docs/en/best-practices and code.claude.com/docs/en/costs, read 2026-11-10)

Turns–

Work turns with a task list and a recap–goal ≥ 95

Turns that changed files or ran commands, kept the task list a person follows, and ended with a recap.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

the desktop_app collector

Goal: template goal, no public benchmark: the task list and the closing recap are how a person follows a session without reading it, so nearly every work turn has both

Sub-sessions–

Sub-sessions closed by their starter–goal ≥ 90

Sessions that a session started for a task and later archived, so none is left running or forgotten.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

the desktop_app collector

Goal: template goal, no public benchmark: a session that starts another owns it to the end, and the desktop docs archive a session once its work is done (code.claude.com/docs/en/desktop, Archive a session, read 2026-11-10)

Subagents–

Subagent runs killed or failed–goal ≤ 2

Subagent runs that were stopped or failed before returning a result; their tokens bought nothing.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

the desktop_app collector

Goal: template goal, no public benchmark: a stopped run returns nothing for the tokens it used, so it stays rare

Speed

Independent subagents start together, and no tool is still connecting when work begins.

Start–

Connectors still connecting at the first prompt–goal ≤ 0

Servers not yet connected when the session takes its first prompt; their tools cannot be called until they are.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

the desktop_app collector

Goal: template goal, no public benchmark: every tool a session lists is ready when its work begins

Delegation–

Agent calls sent together–goal ≥ 90

Subagents started in one message with the others of their turn, not in the next with nothing learned between.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

the desktop_app collector

Goal: template goal, no public benchmark: the sub-agents docs run independent work in parallel (code.claude.com/docs/en/sub-agents, read 2026-11-10) and give no number

Consumption

Listings of skills, agents and tools stay small, and a subagent starts with only the tools its task needs.

Standing load–

Listing tokens a session loads–goal ≤ 10,000

Names and descriptions of every skill, agent, server and tool, which each session carries whether it uses them or not.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

the desktop_app collector

Goal: goal from the skills docs: the skill listing budget is 1% of the model's context window (code.claude.com/docs/en/skills, Skill descriptions are cut short, read 2026-11-10), at a 1,000,000-token window = 10,000 tokens, held here for all four listings together; near at 2%, the docs' example of a raised budget

Context the heaviest subagent type starts with–goal ≤ 30,000

What a subagent carries into its first model call, for the agent type that carries most; every run pays it again.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

the desktop_app collector

Goal: template goal, no public benchmark: a subagent inherits every tool unless its definition lists them (code.claude.com/docs/en/sub-agents, read 2026-11-10), so one given only what its task needs starts small

Blockers

Quality

100

1 of 4 measured

Speed

100

1 of 4 measured

Consumption

–

0 of 2 measured

Ideal state

Every guard, hook and permission check on Claude earns its keep: it stops real mistakes, rarely blocks safe work, and costs little when it fires.

2 sources, newest 2026-11-04
  1. Configure permissions 2026-11-04
  2. Hooks reference 2026-11-04

Quality

No blocker costs more than it catches, guard false alarms are rare, and no change to a gate is reverted or causes a bug.

Quality100

Enforcement PRs reverted or bugged0 PRsgoal ≤ 0

Of the latest merged changes to workflows, git hooks, guards or settings, those a later change undid or a bug blamed.

enforcement PRs merged6 PRs

the last 20 merged PRs touching ^\.github\/workflows\/, ^\.githooks\/, ^\.claude\/hooks\/, (^|\/)(bash|write)_guard[\w.]*$, (^|\/)settings[^/]*\.json$, _rules\.json$, (^|\/)verify\.sh$, (^|\/)manifest\.json$, (^|\/)cloud-test\.json$, (^|\/)protected_clauses\.json$ (gh api graphql files), against later merged PRs titled Revert that name them (gh pr list, title and body) and bug issues opened since the oldest of them merged that name them as the cause, with a Regressed-by line or a cause phrase such as "caused by #N" (gh issue list --type Bug) read 2026-11-16 21:12 UTC

Goal: template goal, no public benchmark: a change to a workflow, hook, guard or setting that is reverted or causes a bug is a gate that broke work

Blockers that cost more than they help–goal ≤ 0

Restrictions whose false alarms are half or more of their alarms; one with no catch measure is not measured.

transcripts and the guard log live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*-home-runner-work-brand-brand*/**/*.jsonl

Goal: template goal, no public benchmark: a restriction whose false alarms are half its fires or more costs more than it catches, so none should

Simplicity and drift–

Guard blocks that were false alarms–goal ≤ 5

Upper bound: a block counts unless each retry ran its own steps, the fix, reads, or with the fix an unblocked edit.

transcripts and the guard log live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*-home-runner-work-brand-brand*/**/*.jsonl

Goal: template goal, no public benchmark: a guard that blocks commands which then run in another form slows work without stopping anything

Guard blocks per 100 tool calls–goal ≤ 0.05

Commands the bash guard stopped in this repo's sessions, per hundred of the latest tool calls.

transcripts and the guard log live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*-home-runner-work-brand-brand*/**/*.jsonl

Goal: template goal, no public benchmark: one block in 2,000 tool calls; a guard that fires often is in the way more than it protects

Speed

Commit and push hooks finish in seconds, and a pull request needs few pushes.

Speed100

Pushes per merged PR2 pushesgoal ≤ 2

Commits that started checks, per merged pull request; each one is another wait.

1 push11 PRs
2 pushes3 PRs
3 pushes2 PRs
4 or more4 PRs

gh api graphql: merged pullRequests, 20 to a query, with pinned reviews and every check run attempt of their last 30 commits; the last 20 merged PRs, 20 read, merged 2026-11-15 05:17 to 2026-11-16 15:45 UTC read 2026-11-16 21:12 UTC

Goal: template goal, no public benchmark: one push and at most one fix push per merged PR; each push is another CI wait

Seconds the latest block cost–goal ≤ 40

Time from the latest blocked call to the session's next call, read now.

transcripts and the guard log live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*-home-runner-work-brand-brand*/**/*.jsonl

Goal: template goal, no public benchmark: set 2026-11-04; a blocked call answered and replaced in well under a minute

git commit, median–goal ≤ 5

How long a commit takes with its pre-commit checks, from the call to its result.

transcripts and the guard log live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*-home-runner-work-brand-brand*/**/*.jsonl

Goal: template goal, no public benchmark: pre-commit checks quick enough not to break a session's flow

git push, median–goal ≤ 10

How long a push takes with its pre-push checks, from the call to its result.

transcripts and the guard log live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*-home-runner-work-brand-brand*/**/*.jsonl

Goal: template goal, no public benchmark: pre-push checks quick enough not to break a session's flow; the full suite runs in CI

Consumption

A blocked call costs one short answer, and hook output and review subagents add few tokens.

Use–

Tokens the latest block cost–goal ≤ 250K

The model call the latest block forced, everything it read and wrote, read now.

transcripts and the guard log live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*-home-runner-work-brand-brand*/**/*.jsonl

Goal: template goal, no public benchmark: set 2026-11-04; less than two model calls at the Claude tab's context goal of 150,000 tokens (2 x 150,000 = 300,000); a block should cost one short answer

Reviewer tokens per merged PR–goal ≤ 1M

Tokens used by review subagents while the latest merged pull requests were open, per pull request.

transcripts and the guard log live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*-home-runner-work-brand-brand*/**/*.jsonl

Goal: template goal, no public benchmark: half the Claude tab's template goal of 2,000,000 tokens per merged PR

Tokens

Quality

–

0 of 1 measured

Speed

–

0 of 2 measured

Consumption

–

0 of 6 measured

Ideal state

Context is finite: keep each call small and cached, spend tokens on work that merges, and give subagents focused tasks that return condensed summaries.

3 sources, newest 2026-11-04
  1. Manage costs effectively 2026-11-04
  2. Prompt caching 2026-11-04
  3. Effective context engineering for AI agents 2025-09-29

Quality

Few tokens go to pull requests that close unmerged.

What tokens buy–

Tokens on PRs closed unmerged–goal ≤ 10

Of the latest finished pull requests, the share of their sessions' tokens that went to ones closed unmerged.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*/**/*.jsonl

Goal: template goal, no public benchmark: a PR closed unmerged throws its session's work away, so its tokens bought nothing

Speed

Most input is read from the cache, and sessions rarely fill their context and compact.

Tokens–

Cache-read share of input–goal ≥ 90

How much of what Claude reads comes from its cache, which costs far less than fresh input.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*-home-runner-work-brand-brand*/**/*.jsonl

Goal: template goal, no public benchmark: a cache read is priced at a small fraction of fresh input (platform.claude.com/docs/en/about-claude/pricing), so a cache miss reprices a whole context

Compactions per 10 merged PRs–goal ≤ 1

How often a session filled its context and had to summarize itself, against how many pull requests merged.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*-home-runner-work-brand-brand*/**/*.jsonl

Goal: template goal, no public benchmark: a compaction drops detail the session gathered

Consumption

Few new tokens per merged pull request, small contexts, cheap subagents with short returns, and the plan never runs out.

Use–

Tokens per merged PR–goal ≤ 2M

New tokens Claude read and wrote for each of the latest merged pull requests, not counting cheap cache rereads.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*-home-runner-work-brand-brand*/**/*.jsonl

Goal: template goal, no public benchmark: the tokens billed at full or write price, about a short session's worth per PR

Median context per model call–goal ≤ 150K

How much Claude carries into each model call; a bigger context costs more and spreads attention thinner.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*-home-runner-work-brand-brand*/**/*.jsonl

Goal: goal from the default trigger of server-side compaction in the Claude Messages API, 150K tokens, whose minimum is 50K (platform.claude.com/docs/en/build-with-claude/compaction-threshold, read 2026-11-03), the point Anthropic treats a context as long enough to summarize; near band: template default

Subagent share of cost–goal ≤ 50

Of what the calls would cost on the API, the share subagents spent; a cheap model lowers it.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*-home-runner-work-brand-brand*/**/*.jsonl

Goal: template goal, no public benchmark: at most half the spend in subagents, priced, so routing work to a cheaper model reads as cheaper

Rework tokens on the latest merged PR–goal ≤ 200K

Tokens a session spent after a review found something, until its pull request merged, on the newest merged one.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*/**/*.jsonl

Goal: template goal, no public benchmark: set 2026-11-04 for the latest merged PR; a tenth of the Claude tab's template goal of 2,000,000 tokens per merged PR, near a quarter

Weekly plan used now–goal ≤ 90

Share of the weekly all-models plan used now, as the app reports it; it resets each week.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*/**/*.jsonl

Goal: template goal, no public benchmark: a tenth of the plan's weekly use left for what comes up; near is the plan's whole allowance

Tokens the latest subagent returned–goal ≤ 2,000

How much the latest subagent handed back to the session that asked; a compact result keeps the main context lean.

transcripts live on the machine that ran the sessions; refreshed by a --local collect there

~/.claude/projects/*/**/*.jsonl

Goal: goal from Anthropic, "Effective context engineering for AI agents" (anthropic.com/engineering/effective-context-engineering-for-ai-agents, 2025-09-29, read 2026-11-03): a subagent "returns only a condensed, distilled summary of its work (often 1,000-2,000 tokens)"; near band: template default, twice that; set 2026-11-04 for the latest subagent

Instructions

Quality

–

0 of 6 measured

Speed

–

0 of 1 measured

Consumption

–

0 of 3 measured

Ideal state

Claude loads only what every session needs: short CLAUDE.md files, path-scoped rules, a memory index of one line per entry, brief skill descriptions.

2 sources, newest 2026-11-03
  1. How Claude remembers your project 2026-11-03
  2. Extend Claude with skills 2026-11-03

Quality

No rule said twice, no broken or stale reference, installed copies match their source, and every rule, skill and agent gets used.

Drift–

Protected clauses missing–goal ≤ 0

Clauses Git Crafter protects that are gone from the installed rule or skill Claude actually reads.

the files Claude loads live on the machine that ran the sessions; refreshed by a --local collect there

the instructions collector

Goal: template goal, no public benchmark: every clause Git Crafter protects holds in the installed copy, as Git Crafter tools/clause_check.py requires of its own checkout

Broken references in rules and memory–goal ≤ 0

Files named in a rule, a CLAUDE.md or a memory note that no longer exist.

rules and memory live on the machine that ran the sessions; refreshed by a --local collect there

quality/collectors/simplicity/simplicity.mjs

Goal: template goal, no public benchmark: a reference to a file that is gone sends Claude looking for it

Installed copies that differ from source–goal ≤ 0

Installed rules and guards that no longer match their source in Git Crafter.

rules and memory live on the machine that ran the sessions; refreshed by a --local collect there

quality/collectors/simplicity/simplicity.mjs

Goal: template goal, no public benchmark: an installed copy that differs from its source runs rules nobody reviewed

Memory notes naming what is gone–goal ≤ 0

Memory notes that name a path or a command flag that no longer exists.

rules and memory live on the machine that ran the sessions; refreshed by a --local collect there

quality/collectors/simplicity/simplicity.mjs

Goal: template goal, no public benchmark: a memory note naming a path or flag that is gone teaches a session something false

Bloat–

Rules, skills and agents unused–goal ≤ 0

Path-scoped rules, skills and agents that no recent session loaded or called.

the files Claude loads live on the machine that ran the sessions; refreshed by a --local collect there

the instructions collector

Goal: template goal, no public benchmark: each rule, skill and agent earns its place in recent work or is removed; a listed one costs tokens every session

Rules said twice–goal ≤ 0

Rule sentences that appear in more than one always-loaded or rule file.

rules and memory live on the machine that ran the sessions; refreshed by a --local collect there

quality/collectors/simplicity/simplicity.mjs

Goal: template goal, no public benchmark: a rule said in two always-loaded files costs its tokens twice on every call and drifts when one copy changes

Speed

A session reads few lines before its first action.

Start–

Lines in the largest always-loaded file–goal ≤ 200

The largest file the repo controls that loads every session: a CLAUDE.md, a rule or the memory index head.

the files Claude loads live on the machine that ran the sessions; refreshed by a --local collect there

the instructions collector

Goal: goal from the Claude Code memory docs' size guidance, "target under 200 lines per CLAUDE.md file. Longer files consume more context and reduce adherence" (code.claude.com/docs/en/memory, read 2026-11-03), applied to each always-loaded file the repo controls; near band: template default, twice that

Consumption

Always-loaded files, the memory index and skill listings stay small.

Always loaded–

Always-loaded instruction tokens–goal ≤ 4,000

Standing instructions every session loads before it starts, the memory index aside; every call carries them.

the files Claude loads live on the machine that ran the sessions; refreshed by a --local collect there

the instructions collector

Goal: template goal, no public benchmark: the always-loaded files within a few thousand tokens, in line with the memory docs' target of under 200 lines per CLAUDE.md file (code.claude.com/docs/en/memory, read 2026-11-03); the MEMORY index is counted apart

Memory index tokens–goal ≤ 600

What the head of the project memory index costs on every call.

the files Claude loads live on the machine that ran the sessions; refreshed by a --local collect there

the instructions collector

Goal: template goal, no public benchmark: a one-line-per-entry index, as the memory docs ask Claude to keep it (code.claude.com/docs/en/memory, read 2026-11-03), stays a few hundred tokens; past the first 200 lines or 25KB nothing loads at all

Skill and agent listing tokens–goal ≤ 2,000

Names and descriptions of every skill and agent, which load into each session so Claude can pick one.

the files Claude loads live on the machine that ran the sessions; refreshed by a --local collect there

the instructions collector

Goal: goal from Claude Code's skill listing budget, 1% of the model's context window (code.claude.com/docs/en/skills, Skill descriptions are cut short, read 2026-11-03), at a 200,000-token window = 2,000 tokens; past it descriptions are dropped. Near band at 1.5 times