agentleFS
ekirtonekirton/Poule/CLAUDE.md

Poule

CLAUDE.md0 starsChanged 8 months agoApache-2.0View on GitHub

Each layer answers "what" for the layer below and "how" for the layer above. A specification at any level describes the "what" for the next level down — customer needs define "what," system decomposition defines "how," and that decomposition becomes the "what" for component-level specs. Use /diagnose to investigate a bug and determine which SDD layer to fix first. Use /sdd <layer> "description" to execute a fix from that layer downward. Use /requirements, /features, /architecture, /specification, /tasks, /tests, or /implementation…

  • Commits and pushes

The file

We split this CLAUDE.md into the rules it gives an agent

9 rules. About 0 are common, 6 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    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
  1. Very specific:

    Use /diagnose to investigate a bug and determine which SDD layer to fix first.

    line 30
  2. Distinctive:

    Use imperative mood: "Fix extraction crash on empty input", not "Fixed" or "Fixes".

    line 36
  3. Distinctive:

    If the commit addresses a specific task or feature, reference it (e.g., "Fix #42").

    line 38
  4. Distinctive:

    Never use generic messages like "wip", "fix", "update", or "changes".

    line 40
  5. Distinctive:

    Do not push to a remote branch after every commit.

    line 44
  6. Distinctive:

    When the user says the branch is ready:

    line 48
  7. Very specific:

    If the branch is out of date with main, rebase before pushing: git rebase origin/main.

    line 54
  8. Very specific:

    Always create new branches from an up-to-date main: git checkout main && git pull origin main && git checkout -b <new-branch>.

    line 56
  9. Distinctive:

    Do not use external Mermaid rendering services (mermaid.ai, mermaid.ink, Mermaid Chart MCP).

    line 62

≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
CLAUDE.md

Directory structure

Each layer answers "what" for the layer below and "how" for the layer above. A specification at any level describes the "what" for the next level down — customer needs define "what," system decomposition defines "how," and that decomposition becomes the "what" for component-level specs.

Layer 1 : Stakeholder intent

  • doc/requirements/ : business goals, user needs, immutable constraints

Layer 2 : Behavioral specification

  • doc/features/ : what and why, design rationale, acceptance criteria

Layer 3 : Design specification

  • doc/architecture/ : how, at design level
  • specification/ : derived from architecture, authorative for implementation

Layer 4 : Implementation specification

  • tasks/ : detailed plan for implementation

Layer 5 : Implementation

  • src/
  • test/
  • commands/ : slash command prompt files (agentic workflows)

SDD Layer Enforcement

Use /diagnose to investigate a bug and determine which SDD layer to fix first. Use /sdd <layer> "description" to execute a fix from that layer downward. Use /requirements, /features, /architecture, /specification, /tasks, /tests, or /implementation to enter a single phase. Use /free to disable enforcement.

Commit Messages

Each commit is preserved on merge, so every message must be meaningful on its own.

  • Use imperative mood: "Fix extraction crash on empty input", not "Fixed" or "Fixes".
  • Subject line under 72 characters. Lead with what changed and why, not how.
  • If the commit addresses a specific task or feature, reference it (e.g., "Fix #42").
  • One logical change per commit. Do not bundle unrelated changes.
  • Never use generic messages like "wip", "fix", "update", or "changes".

Pull Request Process

Do not push to a remote branch after every commit. Push only when the branch is ready to merge — the user makes many commits before a branch is ready.

Never create a PR or enable auto-merge without explicit user approval. CI runs automatically on push and auto-merge is configured on PRs, so an unauthorized PR can merge to main before the user reviews it. Always ask the user before creating a PR.

When the user says the branch is ready:

  1. Review the commit log and set a descriptive PR title: git log --oneline origin/main..HEAD, then gh pr edit <number> --title "...".
  2. Push the branch and open a PR: git push origin <branch> && gh pr create
  3. Enable auto-merge so it merges once CI passes: gh pr merge <number> --auto --rebase

If the branch is out of date with main, rebase before pushing: git rebase origin/main.

Always create new branches from an up-to-date main: git checkout main && git pull origin main && git checkout -b <new-branch>. Never branch from a feature branch — this causes "out of date with base branch" errors on GitHub when the parent branch merges first.

Diagram Visualization

When a user asks to visualize proofs, dependencies, or proof states, use the visualize_* MCP tools (visualize_proof_state, visualize_proof_tree, visualize_dependencies, visualize_proof_sequence). These tools write a proof-diagram.html file to the project directory as a side effect. After calling a visualization tool, tell the user to open proof-diagram.html in their browser.

Do not use external Mermaid rendering services (mermaid.ai, mermaid.ink, Mermaid Chart MCP). All diagram rendering is local via the HTML file.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.