{
  "markdown": "<p align=\"center\">\n  <a href=\"LICENSE\"><img src=\"https://img.shields.io/badge/License-Apache_2.0-blue.svg\" alt=\"License\"></a>\n  <img src=\"https://img.shields.io/badge/Codex_CLI-Skill-5C5C5C.svg\" alt=\"Codex CLI Skill\">\n  <img src=\"https://img.shields.io/badge/Gemini_CLI-Skill-4285F4.svg\" alt=\"Gemini CLI Skill\">\n  <img src=\"https://img.shields.io/badge/Patterns-7_Architectures-orange.svg\" alt=\"7 Architecture Patterns\">\n  <a href=\"https://github.com/tae2089/harness/stargazers\"><img src=\"https://img.shields.io/github/stars/tae2089/harness?style=social\" alt=\"GitHub Stars\"></a>\n</p>\n\n<p align=\"center\">\n  <a href=\"#\"><img src=\"https://img.shields.io/badge/README-EN%20%7C%20KO%20%7C%20JA-lightgrey\" alt=\"i18n\"></a>\n</p>\n\n# Harness — Subagent Orchestration Meta-Framework\n\n**English** | [한국어](README_KO.md) | [日本語](README_JA.md)\n\nA meta-framework for designing specialist subagent teams in AI coding agents. Generates agent definitions, orchestrator skills, and all runtime scaffolding from a single natural-language domain description.\n\n## Available Skills\n\n| Skill | CLI | Agent Definition | Skills Path |\n|-------|-----|-----------------|-------------|\n| `codex-harness` | OpenAI Codex CLI | `.codex/agents/{name}.toml` | `.codex/skills/` |\n| `gemini-harness` | Google Gemini CLI | `.gemini/agents/{name}.md` | `.gemini/skills/` |\n\n---\n\n## codex-harness (OpenAI Codex CLI)\n\n### Installation\n\n**Personal (global):**\n```bash\ngit clone https://github.com/tae2089/harness.git\ncp -r harness/skills/codex-harness ~/.codex/skills/\n```\n\n**Team (per-repo):**\n```bash\ncp -r harness/skills/codex-harness .codex/skills/\n```\n\nAfter installation, say `\"build a codex harness\"` in a Codex CLI session to confirm `codex-harness` skill auto-triggers.\n\n> **First time?** Check [`references/usage-examples.md`](skills/codex-harness/references/usage-examples.md) first. 8 domain scenarios with trigger phrase mappings and a non-trigger table.\n\n### Core Principles (Codex CLI)\n\n- **sandbox_mode Permission Control:** Every agent requires an explicit `sandbox_mode`: `read-only` (Analyst/Architect) · `workspace-write` (Coder/Reviewer/QA) · `danger-full-access` (Operator/Deployer). No wildcard permissions.\n- **Plan Mode Required:** Activate with `/plan` or `Shift+Tab` before new builds and expansions.\n- **Main Agent as Single Broker:** No direct inter-subagent communication API. All collaboration brokered via `_workspace/`.\n- **3-Component Structure:** `.codex/agents/*.toml` + `.codex/skills/*/SKILL.md` + `AGENTS.md`.\n\n### Usage\n\n```\n/plan\nbuild a harness for an SSO authentication project\n```\n\n### Generated Artifacts\n\n```\n{project}/\n├── .codex/\n│   ├── agents/{name}.toml              # Agent definition (TOML: role, sandbox_mode, model)\n│   └── skills/{orchestrator}/\n│       ├── SKILL.md\n│       └── references/schemas/\n├── _workspace/\n│   ├── workflow.md\n│   ├── findings.md\n│   ├── tasks.md\n│   ├── checkpoint.json\n│   └── tasks/task_{agent}_{id}.json\n└── AGENTS.md\n```\n\n---\n\n## gemini-harness (Google Gemini CLI)\n\n### Installation\n\n```bash\ngemini skills install https://github.com/tae2089/harness.git --path skills\n```\n\nAfter installation, say `\"build a harness\"` in a Gemini CLI session to confirm `gemini-harness` skill auto-triggers.\n\n> **First time?** Check [`references/usage-examples.md`](skills/gemini-harness/references/usage-examples.md) first. 8 domain scenarios with trigger phrase mappings and a non-trigger table.\n\n### Core Principles (Gemini CLI)\n\n- **Strict Tool Permission Control:** `tools: [\"*\"]` is forbidden. All agents require `ask_user` and `activate_skill`. `invoke_agent` for orchestrators/supervisors only.\n- **Plan Mode Required:** Use `enter_plan_mode` before new builds and expansions (except yolo mode).\n- **Main Agent as Single Broker:** No `SendMessage`/`TeamCreate` API. All collaboration brokered via `_workspace/`.\n- **3-Component Structure:** `.gemini/agents/` + `.gemini/skills/` + `GEMINI.md`.\n\n### Usage\n\n```\n/gemini-harness build a harness for an SSO authentication project\n```\n\nOr trigger naturally:\n\n| Phrase Pattern | Mode |\n|----------------|------|\n| \"build/design/set up a harness\", \"automate {domain}\" | New build |\n| \"add {feature} to existing harness\", \"add agent\" | Expansion |\n| \"audit/inspect harness\", \"sync drift\" | Operations/maintenance |\n| \"re-run/fix/improve previous result\" | Operations (partial re-run) |\n\n### Generated Artifacts\n\n```\n{project}/\n├── .gemini/\n│   ├── agents/{name}.md                # Agent definition (role, tools, temperature)\n│   └── skills/{orchestrator}/\n│       ├── SKILL.md\n│       └── references/schemas/\n├── _workspace/\n│   ├── workflow.md\n│   ├── findings.md\n│   ├── tasks.md\n│   ├── checkpoint.json\n│   └── tasks/task_{agent}_{id}.json\n└── GEMINI.md\n```\n\n---\n\n## Shared: Core Concepts\n\n### Workflow Phases\n\n| Phase | Description |\n|-------|-------------|\n| Phase 0 | Audit current state and branch by mode (new / expand / operate) |\n| Phase 1 | Domain analysis and pattern matching (usage-examples.md scenario matching) |\n| Phase 2 | Virtual team design + permission mapping + architecture pattern selection |\n| Phase 3 | Agent definition generation |\n| Phase 4 | Orchestrator skill generation |\n| Phase 5 | Integration and orchestration (workflow.md · findings.md · tasks.md · checkpoint.json init) |\n| Phase 6 | Validation (trigger check, Resume, Zero-Tolerance, project manifest registration) |\n\n### 7 Pattern Selection Guide\n\n| Pattern | Best for |\n|---------|----------|\n| Pipeline | Sequential dependent tasks: design → implement → verify |\n| Fan-out/Fan-in | Parallel independent tasks with aggregation |\n| Expert Pool | Situational expert selection and invocation |\n| Producer-Reviewer | Generate → quality-check loop (PASS/FIX/REDO) |\n| Supervisor | Dynamic assignment via tasks.md claim |\n| Hierarchical | 2-tier delegation: team lead → worker (heterogeneous domains) |\n| Handoff | Dynamic routing to next specialist based on analysis result |\n\n### Naming Convention\n\nStage/Step names must be deliverable-noun kebab-case (`^[a-z][a-z0-9-]*$`). Placeholders like `main`, `step1`, `task` are blocked by workflow.md schema validation.\n\n### Zero-Tolerance Failure Protocol\n\nArbitrary skipping is absolutely forbidden. Max 2 retries (3 total) → unresolved → `Blocked` + user confirmation.\n\n---\n\n## Directory Structure\n\n```\nharness/\n└── skills/\n    ├── codex-harness/\n    │   ├── SKILL.md\n    │   └── references/\n    │       ├── usage-examples.md\n    │       ├── agent-design-patterns.md\n    │       ├── orchestrator-template.md\n    │       ├── orchestrator-procedures.md\n    │       ├── team-examples.md\n    │       ├── stage-step-guide.md\n    │       ├── skill-writing-guide.md\n    │       ├── skill-testing-guide.md\n    │       ├── qa-agent-guide.md\n    │       ├── evolution-protocol.md\n    │       ├── expansion-matrix.md\n    │       ├── schemas/\n    │       │   ├── models.md                     # ⚠️ Model ID source of truth\n    │       │   ├── agent-worker.template.toml\n    │       │   ├── agent-state-manager.template.toml\n    │       │   ├── agent-orchestrator.template.md\n    │       │   ├── task.schema.json\n    │       │   ├── checkpoint.schema.json\n    │       │   ├── workflow.template.md\n    │       │   ├── findings.template.md\n    │       │   ├── tasks.template.md\n    │       │   └── README.md\n    │       └── examples/\n    │           ├── full-bundle/sso-style.md\n    │           ├── team/01~05-*.md\n    │           └── step/01~05-*.md\n    └── gemini-harness/\n        ├── SKILL.md\n        └── references/                           # Same structure as codex-harness\n```\n\n## Reference Docs\n\n### codex-harness\n- `skills/codex-harness/SKILL.md` — Main skill definition + workflow + reference index\n- `references/schemas/models.md` — ⚠️ Model ID source of truth + `model_reasoning_effort` selection guide\n- `references/schemas/agent-worker.template.toml` · `agent-orchestrator.template.md` — Agent creation standard templates\n\n### gemini-harness\n- `skills/gemini-harness/SKILL.md` — Main skill definition + workflow + reference index\n- `references/schemas/models.md` — ⚠️ Model ID source of truth\n- `references/schemas/agent-worker.template.md` · `agent-orchestrator.template.md` — Agent creation standard templates\n\n### Shared References\n- `references/usage-examples.md` — 🚀 8 trigger phrases + mode mapping + non-trigger table + Phase matrix\n- `references/agent-design-patterns.md` — 7 patterns detail, agent definition structure, permission mapping\n- `references/orchestrator-template.md` — Orchestrator Step 0~5 pseudocode, checkpoint.json schema\n- `references/orchestrator-procedures.md` — Error handling decision tree, blocked_protocol, handle_handoff\n- `references/team-examples.md` — Pattern-based real-world case index\n- `references/stage-step-guide.md` — workflow.md specification, Stage/Step transition protocol\n- `references/skill-writing-guide.md` — Skill authoring patterns, data schema standards\n- `references/skill-testing-guide.md` — Trigger validation, Resume testing, Zero-Tolerance verification\n- `references/qa-agent-guide.md` — QA agent integration consistency verification\n- `references/evolution-protocol.md` — Harness evolution, operations/maintenance workflow\n- `references/expansion-matrix.md` — Phase selection matrix for existing expansion\n- `references/examples/full-bundle/sso-style.md` — Full artifact package canonical example\n- `references/examples/team/` · `references/examples/step/` — Pattern-based and structure-based detailed examples\n",
  "bytes": 9478,
  "sha": "0d65cbc815fd8200016b45e5af1c8d9fdad58bbe8ab163bd02a8e07c49be1b16",
  "repo_slug": "tae2089/harness",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_tae2089_harness_78a689b4/readme"
}