AGENTS.md

1# claude-agents — multi-harness agentic plugin marketplace
2 
3Production-ready agentic-workflow building blocks: **94 plugins** (90 local + 4 external), **203 agents**, **175 skills**, **109 commands**. Native source-of-truth for Claude Code; also consumed by OpenAI Codex CLI, Cursor, OpenCode, and Gemini CLI from a single Markdown source.
4 
5This file is the canonical context file. Codex / Cursor / OpenCode read it directly. Claude Code reads it via `CLAUDE.md`, a symlink to this file. Gemini CLI reads it via `gemini-extension.json` (`contextFileName`) / `.gemini/settings.json`.
6 
7> **Read this file like a table of contents.** Detail lives in `docs/`. Authoring conventions live in `docs/authoring.md`. Per-harness setup and capability deltas live in [`docs/harnesses.md`](docs/harnesses.md). Gemini-specific setup is in `GEMINI.md` (also auto-loaded by Gemini CLI). This file should never grow beyond ~150 lines (per OpenAI's [harness-engineering](https://openai.com/index/harness-engineering/) practice).
8 
9## Map
10 
11- **[ARCHITECTURE.md](ARCHITECTURE.md)** — top-level architectural overview (adapter framework, source-of-truth invariant, capability matrix summary)
12- **[docs/architecture.md](docs/architecture.md)** — detailed design principles
13- **[docs/plugins.md](docs/plugins.md)** — full plugin catalog (94 plugins by category)
14- **[docs/agents.md](docs/agents.md)** — agent reference (203 agents, model tiers)
15- **[docs/agent-skills.md](docs/agent-skills.md)** — skill reference (progressive disclosure model)
16- **[docs/usage.md](docs/usage.md)** — commands, workflows, examples
17- **[docs/authoring.md](docs/authoring.md)** — portable-content style guide (read before adding plugins)
18- **[docs/harnesses.md](docs/harnesses.md)** — per-harness capability matrix
19- **[docs/plugin-eval.md](docs/plugin-eval.md)** — three-layer quality evaluation framework
20- **[docs/round-trip-results.md](docs/round-trip-results.md)** — real-CLI verification recipes
21- **[CONTRIBUTING.md](CONTRIBUTING.md)** — how to contribute
22 
23## Working in this repo
24 
25- Python tooling: **uv** (package manager), **ruff** (lint/format), **ty** (type check). Do not use pip / mypy / black.
26- Plugins live under `plugins/<name>/` with auto-discovery — see `docs/authoring.md` for frontmatter shapes.
27- Plugin names: lowercase, hyphen-separated. Never use `__` (it's the adapter namespace separator).
28- Never commit secrets. Never run destructive git (force-push, `reset --hard`, branch -D) without explicit ask.
29 
30## Quality gates (run these before pushing)
31 
32```bash
33make validate STRICT=1     # structural validation across all harness outputs
34make garden                # drift detection (dead links, stale artifacts, oversize skills)
35make test                  # full pytest suite (plugin-eval + tools/tests/)
36make smoke-test            # real-CLI subprocess tests against generated artifacts
37```
38 
39CI (`.github/workflows/validate.yml`) runs all four on every PR plus installs OpenCode + Gemini CLI for live verification.
40 
41## Regenerating per-harness artifacts
42 
43```bash
44make generate HARNESS=codex      # .codex/skills, .codex/agents, .codex/plugins/<p>/, .agents/plugins/marketplace.json
45make generate HARNESS=cursor     # .cursor-plugin/{marketplace,plugin}.json, .cursor/rules/
46make generate HARNESS=opencode   # .opencode/{skills,agents,commands,plugins}/, opencode.json
47make generate HARNESS=gemini     # skills/, agents/, commands/ at extension root
48make generate-all                # all four
49```
50 
51Generated artifacts are **committed** so each harness installs natively from a clone / GitHub URL (native-install commands in [`docs/harnesses.md`](docs/harnesses.md)). Run `make generate-all` before committing source changes — CI fails on drift. Source-of-truth lives only under `plugins/`; never hand-edit generated files.
52 
53## Skills (cross-harness)
54 
55175 skills under `plugins/*/skills/<n>/SKILL.md` — discoverable by every harness:
56 
57- **Claude Code**: auto-discovery via Anthropic's SKILL.md spec
58- **Codex CLI**: mirrored to `.codex/skills/<plugin>__<skill>/` (8 KB body cap; detail in `references/details.md`)
59- **OpenCode**: mirrored to `.opencode/skills/<plugin>-<skill>/` using hyphenated names for global install
60- **Cursor**: reads `.claude/skills/` directly (no re-emit)
61- **Gemini CLI**: native skills at `skills/<plugin>__<skill>/SKILL.md`
62 
63Top-level `skills/` is Gemini output; do not use it for OpenCode installs.
64 
65## Subagents (cross-harness)
66 
67203 subagents under `plugins/*/agents/<name>.md`. Per-harness transpilation:
68 
69- **Codex**: `.codex/agents/<plugin>__<agent>.toml` (drop `tools:`, map model alias to the GPT-5.x family, infer `sandbox_mode`)
70- **OpenCode**: `.opencode/agents/<plugin>__<agent>.md` with `mode: subagent` + `permission:` block (locked agents — those with source `tools: []` — get deny-everything except base `skill`/`task`)
71- **Gemini**: `agents/<plugin>__<agent>.md` (April 2026 subagent spec)
72- **Cursor**: reads `.claude/agents/` directly
73 
74## Why this file is short
75 
76Per OpenAI's harness-engineering practice: this file is a **map**, not an encyclopedia. Procedural detail lives in skills (loaded on demand by agents). Reference material lives in `docs/` (loaded when an agent navigates). A single bloated AGENTS.md crowds out the task, rots quickly, and is hard to verify mechanically. Keep it lean; push detail elsewhere.
77