agentleFS
UtsavBalar1231UtsavBalar1231/oh-my-claudeagent/AGENTS.md

oh-my-claudeagent

AGENTS.md3 starsChanged 32 days agoMITView on GitHub

- Use Bun, not npm scripts: bun install --frozen-lockfile. CI pins Bun 1.4.4. just setup only installs pre-commit hooks; it does not install dependencies. - Run just ci before claiming full verification. Its order is lint -> typecheck -> validate -> test -> test-mod -> test-opencode -> smoke. Recipes use bash, including Git Bash on Windows. - just typecheck checks separate mod, Bun runtime and OpenCode projects. - just validate checks docs and runtime contracts. Focus with just validate --check…

The file

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

5 rules. About 0 are common, 1 is distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

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

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

    Use Bun, not npm scripts: bun install --frozen-lockfile.

    line 5
  2. Very specific:

    Run just ci before claiming full verification.

    line 7
  3. Very specific:

    After OpenCode UI or RPC changes, run just build-opencode and include opencode/tui.generated.js with the source changes.

    line 35
  4. Very specific:

    Read CONTRIBUTING.md for prose and prompt contracts.

    line 53
  5. Distinctive:

    Update a directory's AGENTS.md when changing its structure.

    line 55

≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.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
AGENTS.md

Working in this repository

Commands and verification

  • Use Bun, not npm scripts: bun install --frozen-lockfile. CI pins Bun 1.4.4. just setup only installs pre-commit hooks; it does not install dependencies.
  • Run just ci before claiming full verification. Its order is lint -> typecheck -> validate -> test -> test-mod -> test-opencode -> smoke. Recipes use bash, including Git Bash on Windows.
  • just typecheck checks separate mod, Bun runtime and OpenCode projects.
  • just validate checks docs and runtime contracts. Focus with just validate --check <group>: claims, formats, hooks, mod, tree, engine, mcp. The engine group skips without claude; a passing validator can therefore be incomplete.
  • Bun specs are colocated *.spec.ts. Run one with bun test --timeout=30000 ./servers/tools/ast.spec.ts, optionally -t '<test name>'. just test selects explicit roots; bare bun test also collects mod tests that need just test-mod (claude plugin test .). OpenCode specs run separately with just test-opencode.
  • Full checks need ast-grep (or sg) and Claude Code (CI pins 2.1.316). OpenCode integration specs skip without opencode on PATH; CI installs 2.2.5. just smoke uses a mock model; tests/plugin-evals/ calls a real model and is manual.

Runtime boundaries

  • src/core/ is shared with the Claude mod: keep it free of node:/bun: imports and Bun globals. Mod code uses the engine through Host in hooks/host.ts. servers/, statusline/ and scripts/ run on Bun, outside the mod engine.
  • .mcp.json launches servers/omca.ts. Settings hooks in hooks/hooks.json call its omca_hook tool; servers/hooks/registry.ts must reach each handler. Mod features register through hooks/register.ts, the only file that calls on().
  • Hooks may deny, ask or steer, never allow. Keep PermissionRequest unregistered; allow decisions belong to native permission settings. See hooks/AGENTS.md.
  • opencode/index.ts is the server plugin; opencode/rpc.ts is the import-safe contract for the optional TUI. Read opencode/AGENTS.md before adapter work. OpenCode has no Stop veto; historical evidence is not current-checkout verification.
  • After OpenCode UI or RPC changes, run just build-opencode and include opencode/tui.generated.js with the source changes. Never edit it by hand. just test-opencode checks artifact freshness before running specs.
  • .claude-plugin/types/ is an engine-generated snapshot. Refresh with just types using the Claude CLI version pinned in CI; validation rejects a different generator version. Do not edit manually; pre-commit leaves these files byte-identical.

Prompts and instructions

  • Query on-disk inventories rather than asserting counts. Agents are agents/*.md: analyzer, architect, build-fixer, executor, explorer, orchestrator, planner, researcher, reviewer, viewer. When adding one, also update templates/claudemd.md, docs/references.md and this list.
  • Skills are skills/*/SKILL.md. Keep plan and start-work user-only (disable-model-invocation: true) and inline (no context: fork) so orchestration retains the invoking session's Agent tool. See skills/AGENTS.md for prompt boundaries.
  • Restrict agent tools with disallowedTools, not tools: an allowlist blocks MCP inheritance. Use model tier aliases in frontmatter, not generation IDs.
  • Read CONTRIBUTING.md for prose and prompt contracts. Keep model-facing text about current behavior; preserve machine-consumed headings, tokens and frontmatter. Update a directory's AGENTS.md when changing its structure.

Packaging

  • Claude marketplace installs use the packaged plugin branch, not the development tree. scripts/package.ts ships only git-tracked files and excludes dependencies, root manifests, tests, opencode/ and most scripts. Keep shipped imports inside that tree; scripts/setup-statusline.ts is the script exception.
  • For an authorized release, use just release <version> rather than manual version edits. It commits and tags both trees but never pushes; add the matching CHANGELOG heading first. Prerequisites and push order are in CONTRIBUTING.md.

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.