| name | readme-editor |
| description | Create or edit a project's landing documentation set — README.md, AGENTS.md, CLAUDE.md, .github/copilot-instructions.md, and llms.txt — by first deciding which artifacts the repo actually needs, then grilling the user one question at a time with recommended defaults, building a compact README map, using official hyperlinks for named tools and technologies, and only then writing. Inspects the repo before asking anything discoverable. Use when the user wants to create, write, edit, rewrite, audit, polish, or "fix up" a README, AGENTS.md, CLAUDE.md, Copilot instructions, llms.txt, or any project landing/onboarding docs. |
README Editor
A good README is not just polished Markdown. Decide what each artifact owns and which artifacts this repo actually needs before writing.
Artifact split
| File | Audience | Owns |
|---|
README.md | Humans + builders evaluating or running the project | What it is, who it's for, 5-minute success, usage, scope, official links to tools, and links to deeper docs |
AGENTS.md | Coding agents editing the repo | Validation commands, module ownership, repo conventions, what not to assume |
CLAUDE.md | Claude Code specifically | Claude-only memory; usually @import AGENTS.md plus Claude-specific notes |
.github/copilot-instructions.md | GitHub Copilot | Copilot-only conventions |
llms.txt | Public AI doc indexers | Curated public docs index; only for projects with substantial public docs |
Default recommendation: README.md + AGENTS.md. Single-file README.md only for tiny demos with no agents working on them.
Process
- Inspect the repo before asking anything discoverable.
- Capture official links for public tools, runtimes, frameworks, languages, agents, and package managers named in the README.
- Choose the README style from repo evidence. Use references/readme-styles.md when the project is UI, CLI, SaaS/product, frontend app/tooling, backend/API, library, or docs-heavy.
- Grill the user, one question at a time, each with a recommended answer and a one-line reason.
- Build a README map and get explicit approval.
- Write, then report what changed.
Step 1 — Inspect first
Read what is discoverable before asking. Capture:
- Project type: library, app, CLI, web app, mobile, docs site, demo
- Language and primary framework from manifests such as
package.json, pyproject.toml, Cargo.toml, or go.mod
- Scripts that become "how to run/build/test"
- Existing
README.md, AGENTS.md, CLAUDE.md, .github/copilot-instructions.md, llms.txt
- Public vs internal signals:
LICENSE, registry publishing, deployment configs, private/internal language
- Official docs or home pages for named external tools and technologies
- Approximate size and activity: rough LOC, recent commit cadence
If something is discoverable, do not ask. State it and confirm in one line.
Step 2 — Choose the README style
Pick one dominant structure before asking subjective questions. Use repo evidence, not the user's labels alone:
- UI/design system or component library
- Frontend app
- Frontend framework or tooling
- CLI or developer tool
- SaaS/product or self-hosted app
- Backend/API framework
- Library/package
- Tiny demo
- Docs-heavy portal
Read references/readme-styles.md for the section order, first-success shape, link-out strategy, and anti-patterns for each style.
If the repo mixes styles, choose by the first reader and first success. Example: a monorepo with a hosted app and SDKs can have a product README at the root and package READMEs under packages/*.
Step 3 — Grill, one question at a time
Each question follows this shape:
Q: <one sharp question>
Recommended: <one answer>
Why: <one short reason>
Skip any question the inspection already answered. Order:
- Reader — Who opens this README first? developer evaluating, builder integrating, end user, or agent
- First success — What should a reader achieve in 5 minutes?
- Builder path — How does a builder extend, integrate, or contribute? or N/A
- Agent path — Are agents editing this repo? If yes,
AGENTS.md is in scope.
- Scope — What is intentionally out of scope? No fake roadmap or aspirational features.
- Deeper docs — Where do they live?
docs/, website, internal wiki, none
- Artifacts — Confirm the final list of files to write.
Step 4 — Build the README map
Produce this map and get approval before writing:
Reader:
README style:
First success:
Builder path:
Agent path:
Include:
Exclude:
Official links:
Link out:
Unknown:
Recommended artifacts:
See references/examples.md for filled maps across project types.
Step 5 — Write
Follow references/artifacts.md for the structure of each file.
When a usable file already exists, edit, don't replace. Preserve the user's voice; merge new structure into existing content.
Hard rules
- Don't invent. No fake production status, no aspirational features, no "revolutionary / seamless / robust / blazing-fast" filler.
- Don't bloat. A 200-line demo gets a 30-line README. No feature tables for a toy project.
- Don't bury agent rules. Agent guidance goes in
AGENTS.md, not deep inside README.md.
- Don't turn README into architecture docs. Link out to
docs/ or ARCHITECTURE.md if depth is needed.
- Don't guess shape-changing facts. If you can't tell whether the repo is public, internal, library, or app, ask one question.
- Don't cite unofficial docs when official docs exist. First README mention of an external tool, language, framework, runtime, package manager, agent, or platform gets an official hyperlink.