Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Audit and update a project README, or bootstrap a new one. Detects tech stack, versions, and services.
.claude/skills/joshukraine-readme-refresh/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | 69% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 52% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 100% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 46% | 0% |
| case-20 | ✗→✓ | ▲ Improved | 60% | 0% |
Audit and update the project README, or bootstrap one if it doesn't exist. The README is a living document — it evolves with the code and should always reflect the current state of the project.
--bootstrap: Force bootstrap mode even if a README exists (useful to start fresh)README.md exists at the project root and --bootstrap was not passed: enter Refresh mode (Step 1).README.md does not exist or --bootstrap was passed: enter Bootstrap mode (Step 5).Gather facts from the codebase. Do not rely on the README's current claims — verify independently.
Tech stack and versions:
.ruby-version, Gemfile.lock (RUBY VERSION section), .tool-versionsGemfile.lock (search for rails ().node-version, .nvmrc, .tool-versions, package.json (engines)config/database.yml, Gemfile.lock (pg, mysql2, sqlite3)Gemfile or package.json for notable dependenciesServices and infrastructure:
fly.toml (Fly.io), netlify.toml, Procfile, render.yaml, app.json (Heroku), Dockerfile, docker-compose.ymlconfig/storage.ymlconfig/environments/production.rb (SMTP settings, action_mailer config), initializers for Postmark/SendGrid/etc.config/application.rb or initializers (Solid Queue, Sidekiq, etc.)Available commands:
bin/ directory: list executable scripts (bin/rails, bin/dev, bin/ci, bin/setup, etc.)Procfile.dev or Procfile: processes defined for local developmentMakefile or Taskfile: custom task definitionspackage.json scripts sectionConfiguration requirements:
.env.example or .env.template: expected environment variablesconfig/credentials.yml.enc or config/master.key: credentials setupconfig/database.yml: database setup requirementsCI/CD:
.github/workflows/: GitHub Actions workflows.circleci/, .travis.yml, Jenkinsfile: other CI configsRead the current README.md and compare each claim against the facts gathered in Step 1. Build a findings report with three categories:
Gemfile.lock shows Rails 8.0, or mentions a service that has been removed).bin/dev script for local development, a background job processor, a CI workflow).Do not flag:
CHECKPOINT: Present the findings report to the user. Format:
textREADME Refresh — Findings ========================== Outdated (N items): ✗ Ruby version: README says 3.3.0, project uses 4.0.1 ✗ Rails version: README says 7.1, Gemfile.lock shows 8.0.1 ✗ References Redis for caching, but project uses Solid Cache Missing (N items): + bin/dev script not documented (starts Procfile.dev with foreman) + Postmark configured for transactional email — not mentioned + GitHub Actions CI workflow exists — not documented Stale references (N items): ⚠ Setup step 3 references `rake db:seed` — project uses `bin/rails db:seed` ⚠ Link to API docs points to a 404 No issues found: (list any sections that are current and accurate)
Ask: "Want me to apply the mechanical fixes? I'll update versions, add missing sections, and remove stale references. I won't rewrite prose descriptions."
For each confirmed finding:
After applying changes:
Run the same scan as Step 1 to gather project facts.
Based on the scan results, generate a README with these sections (omit any that don't apply):
<!-- TODO: Add project description -->. Do not invent a description.bin/setup, bundle install, bin/rails db:prepare, .env configuration, etc. Only include steps that the project actually needs.bin/dev, bin/rails server, etc.), how to run tests, how to run the linter.docs/ directory exists, mention it and list key documents (PRD, domain model, etc.).CHECKPOINT: Present the generated README to the user for review.
Write the file to README.md at the project root. Do not commit — let the user review and commit when ready.
Makefile, Dockerfile, docker-compose.yml, README.md).CLAUDE.md, read it for additional context about conventions, but do not document CLAUDE.md itself in the README (it is tooling-specific, not project documentation).This is a living-document maintenance command (→ See spec-driven-development.md §5 "Document Lifecycle"). The README is never frozen — it evolves with the code.
| Project Stage | README Refresh Focus | | ------------- | ----------------------------------------------------------------------------- | | Greenfield | Bootstrap mode. Generate initial README as setup docs solidify. | | MVP complete | Full refresh. The README likely drifted during rapid development. | | Mature | Light periodic refresh. Versions and dependencies are the main drift vectors. |
Suggested cadence: Run after phase boundaries, after /update-deps, or whenever the project feels like it has accumulated untracked changes. A good rule of thumb: if you'd be embarrassed for a new contributor to read the README, it's time.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 9,113 | 6,439 | -29% | 1 | 1 | 0% | 1,639 | 2,491 | +52% | 0 | 0 | — |
case-02 | pass→fail | 11,552 | 3,657 | -68% | 1 | 1 | 0% | 1,862 | 2,383 | +28% | 0 | 0 | — |
case-03 | pass→fail | 3,339 | 4,505 | +35% | 1 | 1 | 0% | 576 | 2,219 | +285% | 0 | 0 | — |
case-04 | pass→pass | 7,080 | 3,280 | -54% | 1 | 1 | 0% | 1,227 | 2,488 | +103% | 0 | 0 | — |
case-05 | pass→pass | 6,572 | 4,092 | -38% | 1 | 1 | 0% | 1,353 | 2,650 | +96% | 0 | 0 | — |
case-06 | fail→fail | 8,329 | 3,618 | -57% | 1 | 1 | 0% | 1,417 | 2,293 | +62% | 0 | 0 | — |
case-07 | fail→fail | 8,504 | 3,916 | -54% | 1 | 1 | 0% | 1,483 | 2,365 | +59% | 0 | 0 | — |
case-16 | pass→pass | 9,649 | 3,463 | -64% | 1 | 1 | 0% | 1,638 | 2,502 | +53% | 0 | 0 | — |
case-08 | fail→pass | 9,044 | 3,378 | -63% | 1 | 1 | 0% | 1,452 | 2,457 | +69% | 0 | 0 | — |
case-09 | fail→pass | 9,695 | 2,481 | -74% | 1 | 1 | 0% | 1,551 | 2,358 | +52% | 0 | 0 | — |
case-10 | fail→fail | 8,164 | 2,236 | -73% | 1 | 1 | 0% | 1,377 | 2,295 | +67% | 0 | 0 | — |
case-11 | fail→pass | 7,664 | 2,536 | -67% | 1 | 1 | 0% | 1,209 | 2,422 | +100% | 0 | 0 | — |
case-12 | pass→pass | 14,041 | 7,636 | -46% | 1 | 1 | 0% | 2,501 | 3,405 | +36% | 0 | 0 | — |
case-13 | pass→pass | 10,050 | 3,665 | -64% | 1 | 1 | 0% | 1,942 | 2,599 | +34% | 0 | 0 | — |
case-14 | fail→pass | 9,620 | 2,468 | -74% | 1 | 1 | 0% | 1,630 | 2,377 | +46% | 0 | 0 | — |
case-15 | pass→pass | 16,130 | 8,204 | -49% | 1 | 1 | 0% | 2,756 | 3,433 | +25% | 0 | 0 | — |
case-22 | pass→pass | 13,313 | 6,441 | -52% | 1 | 1 | 0% | 2,401 | 3,154 | +31% | 0 | 0 | — |
case-17 | pass→pass | 7,766 | 4,129 | -47% | 1 | 1 | 0% | 1,335 | 2,619 | +96% | 0 | 0 | — |
case-18 | pass→pass | 12,314 | 5,247 | -57% | 1 | 1 | 0% | 2,077 | 2,824 | +36% | 0 | 0 | — |
case-19 | pass→pass | 10,531 | 2,782 | -74% | 1 | 1 | 0% | 1,779 | 2,529 | +42% | 0 | 0 | — |
case-20 | fail→pass | 10,412 | 5,084 | -51% | 1 | 1 | 0% | 1,760 | 2,817 | +60% | 0 | 0 | — |
case-21 | pass→pass | 13,765 | 6,048 | -56% | 1 | 1 | 0% | 2,325 | 3,133 | +35% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 19 counted toward the lift figure. The other 3 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +14 percentage points is the difference between those two pass rates over the 19 comparable cases. 3 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.