claude-md
Skill2 starsChanged 4 months agoMITView on GitHub
Create a new CLAUDE.md or make targeted edits to an existing one
- Installs packages
16 other projects use a version of this.
Adopted by
Most recently taken up by
- ngmeyer/skillstheir own version, first seen Nov 1, 2026
- coroboros/agent-skillstheir own version, first seen Oct 29, 2026
- tanuuuuuuu/dotfilestheir own version, first seen Oct 20, 2026
- rcgsheffield/skillstheir own version, first seen Oct 13, 2026
- pravsjv/claude-skillstheir own version, first seen Oct 11, 2026
- poshan0126/dotclaudetheir own version, first seen Oct 8, 2026
Other versions16 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 16
- poshan0126/dotclaude★ 871same name
- Alexanderdunlop/skills★ 16same name
- fullymiddleaged/Clawness★ 5same name
- coroboros/agent-skills★ 4same name
- ngmeyer/skills★ 4same name
- sp-daewoon/chronicle★ 4same name
- vndee/engineering-skills★ 3same name
- frankxai/skills★ 2same name
- TeamSPWK/nova★ 2same name
- eugnmueller-87/claude-blueprint★ 1same name
6 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository13 other files it gives its agents.
- commit-pushSkill
- create-cliSkill
- debug-agentSkill
- domain-modelingSkill
- grillingSkill
- grill-with-docsSkill
- handoffSkill
- helpSkill
- prototypeSkill
- review-structureSkill
- schedule-tasksSkill
- skill-masterSkill
- wayfinderSkill
The file
We split this skill into the rules it gives an agent
4 rules. 0 are common, 1 is distinctive, and 3 are very specific to this project.
0
Common
Most of the 13 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 15
Know the repo: If you're creating from scratch, or the edit relies on repo knowledge you haven't verified, read references/explore.md and …
- Distinctive: line 17
Edit or create: If a CLAUDE.md exists, make targeted edits and steer the file toward the template below — don't rewrite from scratch unless …
- Very specific: line 18
Write the domain consumer guide: Read references/domain.md, then create or update agent-docs/domain.md from that seed using the verified …
- Very specific: line 84
Only create a file when you have real content for it — no empty templates.
“Common” is measured: kept by at least 6 in 10 of the 13 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Create or Improve a CLAUDE.md
Goal
CLAUDE.md is the only content guaranteed to appear in every Agent conversation in a repo — and the repo may hold code, business docs, a knowledge base, or any mix. Every line it carries is paid on every task, so it must carry only rules that are universally applicable (defined below) — that test drives every decision here. Such files usually land under ~100 lines, but length itself is not a rule: a longer file is fine when every line passes the test. Treat growth past that mark only as a cue to re-check the rules against it.
Process
- Know the repo. If you're creating from scratch, or the edit relies on repo knowledge you haven't verified, read
references/explore.mdand do the exploration first. - Map the domain docs. Inspect existing
CONTEXT.md,CONTEXT-MAP.md,PEOPLE.md, ADRs, andagent-docs/domain.md. Establish whether the repo is code or non-code: non-code means the deliverable is documents, knowledge, or operations rather than software — supporting scripts don't change this. Use multi-context only whenCONTEXT-MAP.mdexists or the repo has verified distinct domain contexts; code layout or monorepo structure alone is not evidence. Otherwise use single-context. - Edit or create. If a CLAUDE.md exists, make targeted edits and steer the file toward the template below — don't rewrite from scratch unless it's fundamentally broken. If none exists, write one following the template.
- Write the domain consumer guide. Read
references/domain.md, then create or updateagent-docs/domain.mdfrom that seed using the verified layout. Do not create emptyCONTEXT.md,CONTEXT-MAP.md, or ADR files. - Disclose the rest. Material that is only sometimes applicable goes in
agent-docs/, not the root CLAUDE.md. - Validate the whole file against the checklist.
Universally applicable
A rule belongs in CLAUDE.md only if it applies to every single task in the repo: would it apply when fixing a bug? Adding a feature? Drafting a document? Doing research? If the answer is "only sometimes", it belongs in a disclosure target or nowhere.
Belongs: "Use uv add instead of pip install" (every Python task); "Run make check before committing" (every change); "company/identity.md is the canonical source for all bios — never draft from scratch" (every writing task in that repo).
Doesn't belong: "Use React Query for data fetching" (only when adding data fetching); "lead files follow sales/pipeline/_LEAD-TEMPLATE.md" (only when creating leads); style rules a linter or formatter can enforce; inventories of commands, skills, hooks, or automations (discoverable via .claude/, and stale the week after you write them).
Template
# Project Name
[One sentence describing what this repo is]
## Quick Reference
[The handful of commands or canonical-file pointers needed on every task.]
[Code repo, typically:]
- **Build**: `[command]`
- **Test**: `[command]`
- **Lint**: `[command]`
[Docs/ops repo: pointers to canonical sources, e.g.]
- **Identity/Bio** → `company/identity.md` (canonical — never draft from scratch)
## Non-Obvious Structure
[Don't list directories — agents discover those via tools.]
[Document only what file and directory names alone don't reveal:]
[- Generated or machine-managed paths (so agents don't edit or hand-create them)]
[- Monorepo package relationships and dependency direction]
[- Naming conventions that imply behavior (e.g. *.server.ts = server-only)]
[- Canonical vs superseded distinctions (e.g. positioning/ current, archive/ old)]
## Key Patterns
[2-3 critical patterns unique to this repo, as pointers to canonical files (file:line for code) — never copied content]
## Domain docs
[Single-context: Single-context layout. Before working with or producing domain-specific output, read `agent-docs/domain.md` and apply the repo's context.]
[Multi-context: Multi-context layout. Before working in or producing output for a domain area, read `agent-docs/domain.md` to locate and apply that area's context.]
[Non-code repo: extend the chosen line's trigger to also fire on people, e.g. "Before working with domain-specific material, or producing output for or about a specific person (client, stakeholder), read `agent-docs/domain.md` …".]
## Documentation
- `agent-docs/architecture.md` - System design and component relationships
- `agent-docs/[area].md` - Read before creating or editing anything in `[folder]/` ([one-line contents hint])
[Only list files that exist. Do not repeat `agent-docs/domain.md` here; the triggered pointer above already discloses it.]
[Write scoped entries as triggers ("Read before …") — see Disclosure targets.]
## Critical Rules
[Only universally applicable rules; max 5-10; nothing a linter can enforce]Disclosure targets
Material that matters only for certain tasks belongs in agent-docs/ at the repo root, e.g. architecture.md, conventions.md, or content.md. Use portable context pointers rather than scoped CLAUDE.md files, whose automatic loading is Claude Code-specific.
Write each Documentation entry as a trigger — the condition that requires reading the file — not a description of its contents: `agent-docs/<area>.md` - Read before creating or editing anything in `<folder>/` (contents hint). A folder is the most common scope, but any determinate condition works ("Read before creating a lead"); reserve descriptive entries for genuinely cross-cutting references like architecture.md. The trigger is the portable replacement for scoped CLAUDE.md auto-loading. Widen the verb to match the folder's side effects — "creating, editing, or running" where scripts send email or spend money.
Only create a file when you have real content for it — no empty templates. The root CLAUDE.md's Documentation section lists every disclosure file except agent-docs/domain.md, which has its own triggered pointer.
Domain consumer guide
agent-docs/domain.md is the routing layer, not the domain source. Build it from references/domain.md, adapting every path and the single- or multi-context layout to evidence from the repo. Contexts may live anywhere — never assume a code-specific path such as src/. Keep actual domain knowledge in CONTEXT.md, people context in PEOPLE.md, and actual decisions in ADRs; do not duplicate any of these into agent-docs/domain.md.
Checklist
The change is done only when every box checks:
- Every directive universally applicable
- If past ~100 lines, the test above was re-run and every line still passes
- No style rules a linter or formatter could enforce
- No copied content — point at canonical files (
file:linefor code) - No inventories of commands, skills, hooks, or automations
- Every command and file path presented as existing was verified; optional conventional paths are identified as optional
- "Only sometimes" material moved to a disclosure target, not deleted or inlined
- Scoped Documentation entries are trigger-phrased ("Read before …"), verbs matching the folder's side effects
- CLAUDE.md contains the correct single- or multi-context Domain docs pointer, extended with the people trigger in a non-code repo
agent-docs/domain.mdrecords the verified layout without duplicating domain knowledge- No empty
CONTEXT.md,CONTEXT-MAP.md,PEOPLE.md, or ADR files were created
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

