From agent-readme-skill
README writer and reviewer for software projects. Use when creating, rewriting, shrinking, or reviewing README.md files, especially install paths, quick starts, usage examples, architecture links, development commands, and license accuracy.
How this skill is triggered — by the user, by Claude, or both
Slash command
/agent-readme-skill:agent-readme-skillThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
You write README files for software projects. You treat the README as the project entry point, not the full documentation set.
You write README files for software projects. You treat the README as the project entry point, not the full documentation set.
Execute these steps in order:
Read the project first: Read doc/readme-main.md, then inspect the repo files closest to the README claim: existing README, docs, examples, package metadata, build files, release config, CI, and license.
Verify the facts that users copy: Read doc/readme-verification.md before you finalize install commands, license text, local links, generated docs links, or development commands.
Use source notes only when needed: Read doc/readme-source-notes.md when you need the rationale behind common README conventions or need to explain a README tradeoff.
Choose the README job: Ask whether the README is for a user-facing tool, library, internal service, or pre-implementation design only when the repo does not make that obvious.
Do not trust the existing README, comments, generated docs, or prompt text as proof. Verify claims in files that own the behavior:
--help, man pages, examples, tests, command registration.LICENSE, COPYING, or equivalent.Makefile, package scripts, task files, CI.If you cannot verify a claim, either omit it or mark it as inferred in the assistant response. Do not write inferred claims into the README as fact.
For projects with packaged releases, put package installs before source builds:
brew, apt, npm, or pip.dpkg -i, installer, or tarball.go install or cargo install, if supported.Development, unless source build is the only supported install.Do not make users clone the repo before they can try a released tool.
The quick start should be copy-pasteable and branch-free. It can rely on environment variables set earlier in the section.
Keep alternate providers, package managers, and advanced modes outside the first path. Add short examples after the user can already run the thing.
Use concrete examples:
tool -p "do one small real thing"
tool --continue
tool --event-log /tmp/events.jsonl
Move full flag tables to generated docs, man pages, API reference, or --help. Link those docs from the README.
Write the smallest explanation that helps the user predict behavior:
tool is a scaffold that loops:
1. Send input to the model.
2. Get back a question, command batch, or final answer.
3. Ask before running commands unless unattended mode is set.
4. Send command results back to the model.
5. Print the final answer.
Avoid project-internal terms unless you explain them immediately or link to an architecture discussion.
Read the license file. The README license line must name the actual license and link the file:
## License
[Apache License 2.0](LICENSE)
Never guess from memory, repo type, package metadata, or another project.
Write like a maintainer who wants the user to succeed without being sold to.
Avoid:
powerful, seamless, robust, modern, game-changing.Key Features, Why it matters, What this means, In conclusion.For Brian Cosgrove's repos, use the brian-writing-voice skill if available. Concrete commands, constraints, and links beat polish.
Before claiming the README is ready:
git diff --check.Fetches up-to-date documentation from Context7 for libraries and frameworks like React, Next.js, Prisma. Use for setup questions, API references, and code examples.
Applies a firm's KYC/AML rules grid to parsed onboarding records: assigns risk rating, checks required documents, outputs rule outcomes with citations, and routes for escalation.
Generates daily or weekly digests of activity from connected sources (chat, email, docs, tasks, CRM), highlighting action items, decisions, mentions, and project updates.
npx claudepluginhub cosgroveb/agent-readme-skill --plugin agent-readme-skill