Link

1 repoAPI data 2026-11-15 05:38 UTCsite data 2026-11-15 05:38 UTCbuilt 2026-11-15 05:38 UTC

Ideal state

A process is a map of at most six steps where quality, speed and consumption are decided, each scored by one Quality metric.

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

Delivery

A change, from the approved plan to merged on the default branch

Ideal state

Small changes, planned first, green on first push, reviewed at the head and merged within a day, judged by lead time and failure, not volume.

4 sources, newest 2026-11-04
  1. Claude Code: best practices 2026-11-04
  2. Google engineering practices: speed of code reviews 2026-11-04
  3. Trunk Based Development: short-lived feature branches 2026-11-04
  4. DORA's software delivery performance metrics 2026-01-05
1

Plan

critical path

Quality

6.7%

PRs linking an issue

goal ≥ 100%

No per-repo data for this step.

Plan in an issue before code, and split the work into changes that each land on their own.

https://code.claude.com/docs/en/best-practices
  • GitHub Issues
  • Linked pull requests
2

Build

critical path

Quality

33.3%

Share over 400 lines

goal ≤ 10%

No per-repo data for this step.

Build in small changes, each pull request short enough to review in one sitting.

https://code.claude.com/docs/en/best-practices
  • Short-lived branches
  • Git hooks
3

Check

critical path

Consumption

0.03 dollars

Runner dollars at list price, the latest merged PR

goal ≤ 0.36 dollars

No per-repo data for this step.

CI on the latest pull request costs no dollars: it runs within the included Actions minutes.

https://dora.dev/guides/dora-metrics/
  • GitHub Actions standard runners, ubuntu-slim for light jobs
  • Included Actions usage, a budget that stops at it
  • Free lanes: public repos, third-party free minutes
  • Pinned actions

Adopt Git Crafter

A repo, from copying the starters to running Git Crafter's CI and pages and staying current

Ideal state

A repo copies two starters and three CI files, pins Git Crafter to a full SHA, requires its check, and takes each release within days.

4 sources, newest 2026-11-05
  1. GitHub Actions: reuse workflows 2026-11-05
  2. GitHub Actions: secure use reference 2026-11-05
  3. GitHub: about rulesets 2026-11-05
  4. GitHub: Dependabot version updates 2026-11-05
1

Copy

critical path

Quality

77.8%

Actions pinned to a SHA

goal ≥ 100%

No per-repo data for this step.

Copy the starters, CI and pages callers and Dependabot entry; pin Git Crafter to a full SHA with its version.

https://docs.github.com/en/actions/reference/security/secure-use
  • templates/quality/config.json
  • templates/process/config.json
  • templates/ci/ci.yml
  • templates/ci/pages.yml
  • templates/ci/dependabot.yml

Measure

A repo's numbers, from collected to scored, published, goaled and fixed

Ideal state

Metrics are read live in CI, scored against sourced goals, published on every push, and each miss becomes a labeled issue a pull request closes.

4 sources, newest 2026-11-05
  1. Google SRE book: service level objectives 2026-11-05
  2. DORA's software delivery performance metrics 2026-01-05
  3. GitHub Pages: custom workflows 2026-11-05
  4. GitHub: link a pull request to an issue 2026-11-05
1

Collect

critical path

Consumption

0.03 dollars

Runner dollars at list price, the latest merged PR

goal ≤ 0.36 dollars

No per-repo data for this step.

Metrics are read live in CI for cents a pull request; a failed read is not measured, never 0.

https://dora.dev/guides/dora-metrics/
  • actions/pages in CI
  • quality/collect.mjs --local for session data
2

Score

critical path

Quality

0 dates

Dated claims older than their area

goal ≤ 0 dates

No per-repo data for this step.

A metric meets, nears or misses a goal that names its source, and every figure carries a recent read date.

https://sre.google/sre-book/service-level-objectives/
  • quality/score.mjs
  • Each goal's goal_source
4

Goal

critical path

Quality

–

Issues with an area label

goal ≥ 100%

params.area_labels in targets/github.json is empty: list the labels that name an area

No per-repo data for this step.

Each miss becomes a typed issue with an area label if one pull request fixes it, else a board card.

https://sre.google/sre-book/service-level-objectives/
  • The work queue
  • Area labels

Code

A repo's code, from measured to simplified, deleted, explained, checked and held small

Ideal state

Code stays small and plain: the worst function is split first, unused lines are deleted, shared functions say what they do.

6 sources, newest 2026-11-10
  1. Microsoft Learn: Code metrics, cyclomatic complexity 2026-10-20
  2. GitHub Docs: Code Quality metrics and scores reference 2026-11-10
  3. Claude Code Docs: Tools reference 2026-11-10
  4. Anthropic: Effective context engineering for AI agents 2025-09-29
  5. PEP 257: Docstring Conventions 2024-04-17
  6. Claude Code Docs: Best practices for Claude Code 2026-11-10
4

Explain

Quality

100%

Shared functions with a header comment

goal ≥ 90%

No per-repo data for this step.

Every shared function says what it does where it starts.

https://peps.python.org/pep-0257/
  • A header comment or docstring

Triage

A finding, from intake to the issue or card that a merged pull request closes

Ideal state

A finding becomes a typed issue closed within a day, or a card on the board; each closes through its merged pull request.

5 sources, newest 2026-11-10
  1. GitHub Actions billing 2026-11-04
  2. GitHub Docs: About issues 2026-11-10
  3. GitHub Docs: Linking a pull request to an issue 2026-11-10
  4. GitHub Docs: Best practices for Projects 2026-11-10
  5. GitHub Docs: Using the built-in automations 2026-11-10

Secure

A repo's security, from keys no session reads to alerts closed within a week

Ideal state

No session reads a key, scans run in minutes, the threat model stays current, reports have a private path, and alerts close within a week.

4 sources, newest 2026-11-10
  1. Claude Code: configure permissions 2026-11-04
  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 2026-11-03

Document

A doc, from written visual first to linted, mapped to its files, published, read and loaded

Ideal state

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

5 sources, newest 2026-11-10
  1. GitHub Docs: Creating diagrams 2026-11-10
  2. GitHub Docs: Style guide 2026-11-10
  3. GitHub Docs: About the repository README file 2026-11-10
  4. Microsoft Style Guide: Scannable content 2023-06-20
  5. Claude Code Docs: Best practices 2026-11-10
2

Lint

critical path

Quality

–

Content-lint WARN in docs

goal ≤ 0 findings

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

No per-repo data for this step.

Every doc passes the content lint before it merges.

https://docs.github.com/en/contributing/style-guide-and-content-model/style-guide
  • The repo's content lint (docs params.lint)
  • Git Crafter: tools/voice_lint.py
6

Load

Consumption

0 tokens

Tokens past the reading goal

goal ≤ 0 tokens

No per-repo data for this step.

Claude loads no tokens past a doc's reading goal.

https://code.claude.com/docs/en/best-practices
  • Agent docs: CLAUDE.md, skills and rules

Contribute

An outside contribution, from idea to production, with AI agents reviewing, merging and deploying it

Ideal state

An outside idea becomes a merged change within a day, no fork code runs with a secret, and no person waits in the loop.

4 sources, newest 2026-11-12
  1. GitHub Docs: About discussions 2026-11-12
  2. GitHub Docs: Approving workflow runs from forks 2026-11-12
  3. GitHub Docs: Events that trigger workflows 2026-11-12
  4. Cloudflare Docs: Rollbacks 2026-11-12

Guard

An agent's tool call, through the repo's guards, to run or blocked

Ideal state

Guards block one-way doors and machine policy, each block names its fix, false alarms become tests, and a guard change adds one regression case.

2 sources, newest 2026-11-04
  1. Claude Code: configure permissions 2026-11-04
  2. Claude Code: hooks reference 2026-11-04
1

Call

critical path

Quality

–

Guard blocks per 100 tool calls

goal ≤ 0.05 blocks

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

No per-repo data for this step.

Deny rules and hooks block only one-way-door commands; most calls pass untouched.

https://code.claude.com/docs/en/permissions
  • Claude Code permissions
  • PreToolUse hooks
2

Block

critical path

Speed

–

Seconds the latest block cost

goal ≤ 40 s

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

No per-repo data for this step.

A block names its fix, so the session moves on in seconds.

https://code.claude.com/docs/en/hooks
  • The guard's own fix line
3

Message

critical path

Consumption

–

Tokens the latest block cost

goal ≤ 250K tokens

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

No per-repo data for this step.

A block's message is short: the rule and the fix, nothing more.

https://code.claude.com/docs/en/hooks
  • Short hook output
4

False alarm

Quality

–

Guard blocks that were false alarms

goal ≤ 5%

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

No per-repo data for this step.

A block on a safe command is a false alarm, and it becomes a guard test.

https://code.claude.com/docs/en/hooks
  • A guard test per false alarm
5

Change

Quality

0 PRs

Enforcement PRs reverted or bugged

goal ≤ 0 PRs

No per-repo data for this step.

A guard fix lands with one regression case under the normal review, and a new rule needs a real incident.

https://code.claude.com/docs/en/hooks
  • One regression case
  • The normal review

Decide

A question or blocker, from where it arises to Claude's call or the owner's queue, then cleared for good

Ideal state

Claude makes every call its delegation allows and moves on; only questions and human steps reach the owner, and a false block is fixed once.

3 sources, newest 2026-11-10
  1. Claude Code: configure permissions 2026-11-04
  2. Claude Code: hooks reference 2026-11-04
  3. Claude Code Docs: Best practices 2026-11-10
1

Arise

critical path

Quality

–

Blockers that cost more than they help

goal ≤ 0 blockers

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

No per-repo data for this step.

A question or blocker surfaces once with its fix named; none costs more than it helps.

https://code.claude.com/docs/en/hooks
  • The bash guard's fix line
  • rules/delegation.md
2

Claude decides

critical path

Consumption

–

Tokens the latest block cost

goal ≤ 250K tokens

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

No per-repo data for this step.

Within its delegation Claude decides, records reason and reversal in the PR, and moves on.

https://code.claude.com/docs/en/best-practices
  • Decide without asking
  • The PR description
3

Owner's queue

Speed

–

Permission prompts per 100 tool calls

goal ≤ 0 prompts

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

No per-repo data for this step.

Only questions and human steps reach the owner; any other prompt is a defect to remove.

https://code.claude.com/docs/en/permissions
  • The owner's queue
  • One question listing every item
4

Clear

critical path

Quality

–

Guard blocks that were false alarms

goal ≤ 5%

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

No per-repo data for this step.

Few blocks are false alarms; each one on a safe command becomes a guard fix by PR.

https://code.claude.com/docs/en/hooks
  • A guard fix by PR
  • One regression case

Ideal state

Every repo runs one CI standard on hosted runners, pinned to a SHA, and skips unchanged work so included minutes last the month.

2 sources, newest 2026-11-04
  1. GitHub Actions: reuse workflows 2026-11-04
  2. GitHub: about billing for GitHub Actions 2026-11-04

The CI standard

Runner dollars at list price, the latest merged PR

0.03 dollars

ci0.02 dollars
pages0.01 dollars

templates/ci/README.md

link3 workflows

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 checks0.6 min

  2. Review

  3. Merge to main1.1 min

Every step as a table
StageStepKindMedianPass rate
Pull request checkspages (rail)Workflow0.6 min80%
Pull request checksciWorkflow0.3 min95%
ReviewPinned review and merge (rail)Personnot timednot measured
Merge to mainpages (rail)Workflow1.1 min100%
Merge to mainciWorkflow0.3 min100%
Merge to mainscorecardWorkflow0.5 min100%

gh api graphql: HEAD:.github/workflows, HEAD:.githooks and HEAD:.github/cloud-test.json; each workflow's latest run timed, its pass rate over its last 20 runs read 2026-11-15 05:38 UTC

Ideal state

Many fast unit tests, fewer broad ones, few end-to-end; every suite runs in CI on every PR, and main stays green.

1 source, newest 2018-02-26
  1. Martin Fowler: The Practical Test Pyramid 2018-02-26

No suite runs on a local machine. A session dispatches one to cloud-test.

link0 suites

Test cases by suite: not measured: the site's data has no tests.cases

No workflow named for tests.

gh api graphql: HEAD:.github/workflows, HEAD:.githooks and HEAD:.github/cloud-test.json; each workflow's latest run timed, its pass rate over its last 20 runs read 2026-11-15 05:38 UTC