agentleFS
Sign inStart free
iadr-deviadr-dev/greenlight/CLAUDE.md

greenlight

CLAUDE.md2 starsChanged 5 days agoMITView on GitHub

greenlight is a Claude Code plugin for AI × BDD spec-driven development: a feature request becomes human-approved rules and examples, then red tests, then green code, under gates the AI cannot bypass. Plugin name and repo: greenlight. CLI: bdd. Project data: .bdd/.

The file

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

4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Run claude plugin validate .

    line 10
  2. Distinctive:

    Never weaken a test to make it pass.

    line 17
  3. Distinctive:

    Never present soft enforcement as hard.

    line 32
  4. Very specific:

    Only 6 skills (init, feature, approve, status, change, adopt).

    line 33

≈ “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

CLAUDE.md — greenlight

greenlight is a Claude Code plugin for AI × BDD spec-driven development: a feature request becomes human-approved rules and examples, then red tests, then green code, under gates the AI cannot bypass. Plugin name and repo: greenlight. CLI: bdd. Project data: .bdd/.

How to work

  • One task per branch and PR (lint-near-duplicates, …). The PR description lists the task's acceptance criteria and how each was verified (command output or test name).
  • Run claude plugin validate . --strict and claude plugin validate skills --strict before releases (a root-level CLAUDE.md warning on the plugin manifest is expected).
  • Squash-merge when checks pass. Update CHANGELOG.md under Unreleased in every PR.
  • Record new design decisions in docs/dev/decisions.md; platform facts in docs/dev/platform-notes.md (verified by experiment, with the date).
  • Dogfood: specify greenlight's own CLI behaviour as Gherkin in specs/ before implementing it, using the rules the plugin enforces (trace tags, vocabulary, red first).
  • Never weaken a test to make it pass; fix the code or raise the spec issue.

Engineering rules

  • tools/bdd is Python 3.11+, stdlib only (TOML via tomllib). No third-party imports at runtime. Dev/test-only dependencies go in requirements-dev.txt.
  • Cross-platform: Windows (hooks need Git for Windows, D11), macOS, Linux. Run python evals/platform-probe/probe.py after Claude Code upgrades. Use pathlib. Hooks run under Git Bash on Windows. Hooks and skills call ${CLAUDE_PLUGIN_ROOT}/tools/bdd/bdd, never python directly; the wrapper is POSIX sh, LF line endings (see .gitattributes), and runs Python with -I.
  • Hooks must be no-ops outside repos with .bdd/ and finish in < 200 ms (macOS/Linux) or < 300 ms (Windows).
  • Deterministic work (lint, hash, trace, status, classify) belongs in the CLI, never in a prompt.
  • File formats: TOML for human-edited files, JSON for tool-owned files (D1).
  • Human approvals are written only by bdd approve --from-hook (UserPromptSubmit). --self-reported exists for soft mode and must refuse when hard enforcement is active. Never present soft enforcement as hard.
  • Only 6 skills (init, feature, approve, status, change, adopt); phase logic goes in phases/<phase>.md; reusable knowledge goes in cards/ (frontmatter + Known pitfalls).
  • Agents: explicit tool lists, read artifacts not chat history, state their output files.

Definition of done (per task)

  • Acceptance criteria met and demonstrated.
  • Unit tests for CLI changes; CI green on all three OSes.
  • Docs and CHANGELOG updated.

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.

Check your setup, free