{"owner":"SuperClaude-Org","repo":"SuperClaude_Framework","hasSkills":true,"hasMcp":false,"mcpConfig":null,"found":["AGENTS.md","CLAUDE.md"],"skills":{"AGENTS.md":"# Repository Guidelines\n\n## Project Structure & Module Organization\n- `src/superclaude/` holds the Python package and pytest plugin entrypoints.\n- `tests/` contains Python integration/unit suites; markers map to features in `pyproject.toml`.\n- `pm/`, `research/`, and `index/` house TypeScript agents with standalone `package.json`.\n- `skills/` holds runtime skills (e.g., `confidence-check`); `commands/` documents scripted Claude commands.\n- `docs/` provides reference packs; start with `docs/developer-guide` for workflow expectations.\n\n## Build, Test, and Development Commands\n- `make install` installs the framework editable via `uv pip install -e \".[dev]\"`.\n- `make test` runs `uv run pytest` across `tests/`.\n- `make doctor` or `make verify` check CLI wiring and plugin health.\n- `make lint` and `make format` delegate to Ruff; run after significant edits.\n- TypeScript agents: inside `pm/`, run `npm install` once, then `npm test` or `npm run build`; repeat for `research/` and `index/`.\n\n## Coding Style & Naming Conventions\n- Python: 4-space indentation, Black line length 88, Ruff `E,F,I,N,W`; prefer snake_case for modules/functions and PascalCase for classes.\n- Keep pytest markers explicit (`@pytest.mark.unit`, etc.) and match file names `test_*.py`.\n- TypeScript: rely on project `tsconfig.json`; keep filenames kebab-case and exported classes PascalCase; align with existing PM agent modules.\n- Reserve docstrings or inline comments for non-obvious orchestration; let clear naming do the heavy lifting.\n\n## Testing Guidelines\n- Default to `make test`; add `uv run pytest -m unit` to scope runs during development.\n- When changes touch CLI or plugin startup, extend integration coverage in `tests/test_pytest_plugin.py`.\n- Respect coverage focus on `src/superclaude` (`tool.coverage.run`); adjust configuration instead of skipping logic.\n- For TypeScript agents, add Jest specs under `__tests__/*.test.ts` and keep coverage thresholds satisfied via `npm run test:coverage`.\n\n## Commit & Pull Request Guidelines\n- Follow Conventional Commits (`feat:`, `fix:`, `refactor:`) as seen in `git log`; keep present-tense summaries under ~72 chars.\n- Group related file updates per commit to simplify bisects and release notes.\n- Before opening a PR, run `make lint`, `make format`, and `make test`; include summaries of verification steps in the PR description.\n- Reference linked issues (`Closes #123`) and, for agent workflow changes, add brief reproduction notes; screenshots only when docs change.\n- Tag reviewers listed in `CODEOWNERS` when touching owned directories.\n\n## Plugin Deployment Tips\n- Use `make install-plugin` to mirror the development plugin into `~/.claude/plugins/pm-agent`; prefer `make reinstall-plugin` after local iterations.\n- Validate plugin detection with `make test-plugin` before sharing artifact links or release notes.\n","CLAUDE.md":"# CLAUDE.md\n\nThis file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.\n\n## 🐍 Python Environment Rules\n\n**CRITICAL**: This project uses **UV** for all Python operations. Never use `python -m`, `pip install`, or `python script.py` directly.\n\n### Required Commands\n\n```bash\n# All Python operations must use UV\nuv run pytest                    # Run tests\nuv run pytest tests/pm_agent/   # Run specific tests\nuv pip install package           # Install dependencies\nuv run python script.py          # Execute scripts\n```\n\n## 📂 Project Structure\n\n**Current v4.3.0 Architecture**: Python package with 30 commands, 20 agents, 7 modes\n\n```\n# Claude Code Configuration (v4.3.0)\n# Installed via `superclaude install` to user's home directory\n~/.claude/\n├── settings.json\n├── commands/sc/         # 30 slash commands (/sc:research, /sc:implement, etc.)\n│   ├── pm.md\n│   ├── research.md\n│   ├── implement.md\n│   └── ... (30 total)\n├── agents/              # 20 domain-specialist agents (@pm-agent, @system-architect, etc.)\n│   ├── pm-agent.md\n│   ├── system-architect.md\n│   └── ... (20 total)\n└── skills/              # Skills (confidence-check, etc.)\n\n# Python Package\nsrc/superclaude/\n├── __init__.py          # Public API: ConfidenceChecker, SelfCheckProtocol, ReflexionPattern\n├── pytest_plugin.py     # Auto-loaded pytest integration (5 fixtures, 9 markers)\n├── pm_agent/            # confidence.py, self_check.py, reflexion.py, token_budget.py\n├── execution/           # parallel.py, reflection.py, self_correction.py\n├── cli/                 # main.py, doctor.py, install_commands.py, install_mcp.py, install_skill.py\n├── commands/            # 30 slash command definitions (.md files)\n├── agents/              # 20 agent definitions (.md files)\n├── modes/               # 7 behavioral modes (.md files)\n├── skills/              # Installable skills (confidence-check, etc.)\n├── hooks/               # Claude Code hook definitions\n├── mcp/                 # MCP server configurations (10 servers)\n└── core/                # Core utilities\n\n# Project Files\ntests/                   # Python test suite (136 tests)\n├── unit/                # Unit tests (auto-marked @pytest.mark.unit)\n└── integration/         # Integration tests (auto-marked @pytest.mark.integration)\ndocs/                    # Documentation\nscripts/                 # Analysis tools (workflow metrics, A/B testing)\nplugins/                 # Exported plugin artefacts for distribution\nPLANNING.md              # Architecture, absolute rules\nTASK.md                  # Current tasks\nKNOWLEDGE.md             # Accumulated insights\n```\n\n### Claude Code Integration Points\n\nSuperClaude integrates with Claude Code through these mechanisms:\n- **Slash Commands**: 30 commands installed to `~/.claude/commands/sc/` (e.g., `/sc:pm`, `/sc:research`)\n- **Agents**: 20 agents installed to `~/.claude/agents/` (e.g., `@pm-agent`, `@system-architect`)\n- **Skills**: Installed to `~/.claude/skills/` (e.g., confidence-check)\n- **Hooks**: Session lifecycle hooks in `src/superclaude/hooks/`\n- **Settings**: Project settings in `.claude/settings.json`\n- **Pytest Plugin**: Auto-loaded via entry point, provides fixtures and markers\n- **MCP Servers**: 8+ servers configurable via `superclaude mcp`\n\n## 🔧 Development Workflow\n\n### Essential Commands\n\n```bash\n# Setup\nmake dev              # Install in editable mode with dev dependencies\nmake verify           # Verify installation (package, plugin, health)\n\n# Testing\nmake test             # Run full test suite\nuv run pytest tests/pm_agent/ -v              # Run specific directory\nuv run pytest tests/test_file.py -v           # Run specific file\nuv run pytest -m confidence_check             # Run by marker\nuv run pytest --cov=superclaude               # With coverage\n\n# Code Quality\nmake lint             # Run ruff linter\nmake format           # Format code with ruff\nmake doctor           # Health check diagnostics\n\n# MCP Servers\nsuperclaude mcp                              # Interactive install (gateway default)\nsuperclaude mcp --list                       # List available servers\nsuperclaude mcp --servers airis-mcp-gateway  # Install AIRIS Gateway (recommended)\nsuperclaude mcp --servers tavily context7    # Install individual servers\n\n# Plugin Packaging\nmake build-plugin            # Build plugin artefacts into dist/\nmake sync-plugin-repo        # Sync artefacts into ../SuperClaude_Plugin\n\n# Maintenance\nmake clean            # Remove build artifacts\n```\n\n## 📦 Core Architecture\n\n### Pytest Plugin (Auto-loaded)\n\nRegistered via `pyproject.toml` entry point, automatically available after installation.\n\n**Fixtures**: `confidence_checker`, `self_check_protocol`, `reflexion_pattern`, `token_budget`, `pm_context`\n\n**Auto-markers**:\n- Tests in `/unit/` → `@pytest.mark.unit`\n- Tests in `/integration/` → `@pytest.mark.integration`\n\n**Custom markers**: `@pytest.mark.confidence_check`, `@pytest.mark.self_check`, `@pytest.mark.reflexion`\n\n### PM Agent - Three Core Patterns\n\n**1. ConfidenceChecker** (src/superclaude/pm_agent/confidence.py)\n- Pre-execution confidence assessment: ≥90% required, 70-89% present alternatives, <70% ask questions\n- Prevents wrong-direction work, ROI: 25-250x token savings\n\n**2. SelfCheckProtocol** (src/superclaude/pm_agent/self_check.py)\n- Post-implementation evidence-based validation\n- No speculation - verify with tests/docs\n\n**3. ReflexionPattern** (src/superclaude/pm_agent/reflexion.py)\n- Error learning and prevention\n- Cross-session pattern matching\n\n### Parallel Execution\n\n**Wave → Checkpoint → Wave pattern** (src/superclaude/execution/parallel.py):\n- 3.5x faster than sequential execution\n- Automatic dependency analysis\n- Example: [Read files in parallel] → Analyze → [Edit files in parallel]\n\n### Slash Commands, Agents & Modes (v4.3.0)\n\n- Install via: `pipx install superclaude && superclaude install`\n- **30 Commands** installed to `~/.claude/commands/sc/` (e.g., `/sc:pm`, `/sc:research`, `/sc:implement`)\n- **20 Agents** installed to `~/.claude/agents/` (e.g., `@pm-agent`, `@system-architect`, `@deep-research`)\n- **7 Behavioral Modes**: Brainstorming, Business Panel, Deep Research, Introspection, Orchestration, Task Management, Token Efficiency\n- **Skills**: Installable to `~/.claude/skills/` (e.g., confidence-check)\n\n> **Note**: TypeScript plugin system planned for v5.0 ([#419](https://github.com/SuperClaude-Org/SuperClaude_Framework/issues/419))\n\n## 🧪 Testing with PM Agent\n\n### Example Test with Markers\n\n```python\n@pytest.mark.confidence_check\ndef test_feature(confidence_checker):\n    \"\"\"Pre-execution confidence check - skips if < 70%\"\"\"\n    context = {\"test_name\": \"test_feature\", \"has_official_docs\": True}\n    assert confidence_checker.assess(context) >= 0.7\n\n@pytest.mark.self_check\ndef test_implementation(self_check_protocol):\n    \"\"\"Post-implementation validation with evidence\"\"\"\n    implementation = {\"code\": \"...\", \"tests\": [...]}\n    passed, issues = self_check_protocol.validate(implementation)\n    assert passed, f\"Validation failed: {issues}\"\n\n@pytest.mark.reflexion\ndef test_error_learning(reflexion_pattern):\n    \"\"\"If test fails, reflexion records for future prevention\"\"\"\n    pass\n\n@pytest.mark.complexity(\"medium\")  # simple: 200, medium: 1000, complex: 2500\ndef test_with_budget(token_budget):\n    \"\"\"Token budget allocation\"\"\"\n    assert token_budget.limit == 1000\n```\n\n## 🌿 Git Workflow\n\n**Branch structure**: `master` (production) ← `integration` (testing) ← `feature/*`, `fix/*`, `docs/*`\n\n**Standard workflow**:\n1. Create branch from `integration`: `git checkout -b feature/your-feature`\n2. Develop with tests: `uv run pytest`\n3. Commit: `git commit -m \"feat: description\"` (conventional commits)\n4. Merge to `integration` → validate → merge to `master`\n\n**Current branch**: See git status in session start output\n\n### Parallel Development with Git Worktrees\n\n**CRITICAL**: When running multiple Claude Code sessions in parallel, use `git worktree` to avoid conflicts.\n\n```bash\n# Create worktree for integration branch\ncd ~/github/SuperClaude_Framework\ngit worktree add ../SuperClaude_Framework-integration integration\n\n# Create worktree for feature branch\ngit worktree add ../SuperClaude_Framework-feature feature/pm-agent\n```\n\n**Benefits**:\n- Run Claude Code sessions on different branches simultaneously\n- No branch switching conflicts\n- Independent working directories\n- Parallel development without state corruption\n\n**Usage**:\n- Session A: Open `~/github/SuperClaude_Framework/` (current branch)\n- Session B: Open `~/github/SuperClaude_Framework-integration/` (integration)\n- Session C: Open `~/github/SuperClaude_Framework-feature/` (feature branch)\n\n**Cleanup**:\n```bash\ngit worktree remove ../SuperClaude_Framework-integration\n```\n\n## 📝 Key Documentation Files\n\n**PLANNING.md** - Architecture, design principles, absolute rules\n**TASK.md** - Current tasks and priorities\n**KNOWLEDGE.md** - Accumulated insights and troubleshooting\n\nAdditional docs in `docs/user-guide/`, `docs/developer-guide/`, `docs/reference/`\n\n## 💡 Core Development Principles\n\n### 1. Evidence-Based Development\n**Never guess** - verify with official docs (Context7 MCP, WebFetch, WebSearch) before implementation.\n\n### 2. Confidence-First Implementation\nCheck confidence BEFORE starting: ≥90% proceed, 70-89% present alternatives, <70% ask questions.\n\n### 3. Parallel-First Execution\nUse **Wave → Checkpoint → Wave** pattern (3.5x faster). Example: `[Read files in parallel]` → Analyze → `[Edit files in parallel]`\n\n### 4. Token Efficiency\n- Simple (typo): 200 tokens\n- Medium (bug fix): 1,000 tokens\n- Complex (feature): 2,500 tokens\n- Confidence check ROI: spend 100-200 to save 5,000-50,000\n\n## 🔧 MCP Server Integration\n\n**Recommended**: Use **airis-mcp-gateway** for unified MCP management.\n\n```bash\nsuperclaude mcp  # Interactive install, gateway is default (requires Docker)\n```\n\n**Gateway Benefits**: 60+ tools, 98% token reduction, single SSE endpoint, Web UI\n\n**High Priority Servers** (included in gateway):\n- **Tavily**: Web search (Deep Research)\n- **Context7**: Official documentation (prevent hallucination)\n- **Sequential**: Token-efficient reasoning (30-50% reduction)\n- **Serena**: Session persistence\n- **Mindbase**: Cross-session learning\n\n**Optional**: Playwright (browser automation), Magic (UI components), Chrome DevTools (performance)\n\n**Usage**: TypeScript plugins and Python pytest plugin can call MCP servers. Always prefer MCP tools over speculation for documentation/research.\n\n## 🚀 Development & Installation\n\n### Current Installation Method (v4.3.0)\n\n**Standard Installation**:\n```bash\n# Option 1: pipx (recommended)\npipx install superclaude\nsuperclaude install\n\n# Option 2: Direct from repo\ngit clone https://github.com/SuperClaude-Org/SuperClaude_Framework.git\ncd SuperClaude_Framework\n./install.sh\n```\n\n**Development Mode**:\n```bash\n# Install in editable mode\nmake dev\n\n# Run tests\nmake test\n\n# Verify installation\nmake verify\n```\n\n### Plugin System (v5.0 - Not Yet Available)\n\nThe TypeScript plugin system (`.claude-plugin/`, marketplace) is planned for v5.0.\nSee `docs/plugin-reorg.md` for details.\n\n## 📊 Package Information\n\n**Package name**: `superclaude`\n**Version**: 4.3.0\n**Python**: >=3.10\n**Build system**: hatchling (PEP 517)\n\n**Entry points**:\n- CLI: `superclaude` command\n- Pytest plugin: Auto-loaded as `superclaude`\n\n**Dependencies**:\n- pytest>=7.0.0\n- click>=8.0.0\n- rich>=13.0.0\n\n## 🔌 Claude Code Native Features (for developers)\n\nSuperClaude extends Claude Code through its native extension points. When developing SuperClaude features, use these Claude Code capabilities:\n\n### Extension Points We Use\n- **Custom Commands** (`~/.claude/commands/sc/*.md`): 30 `/sc:*` commands\n- **Custom Agents** (`~/.claude/agents/*.md`): 20 domain-specialist agents\n- **Skills** (`~/.claude/skills/`): confidence-check skill\n- **Settings** (`.claude/settings.json`): Permission rules, hooks\n- **MCP Servers**: 8 pre-configured + AIRIS gateway\n- **Pytest Plugin**: Auto-loaded via entry point\n\n### Extension Points We Should Use More\n- **Hooks** (28 events): `SessionStart`, `Stop`, `PostToolUse`, `TaskCompleted` — ideal for PM Agent auto-restore, self-check validation, and reflexion triggers\n- **Skills System**: Commands should migrate to proper skills with YAML frontmatter for auto-triggering, tool restrictions, and effort overrides\n- **Plan Mode**: Could integrate with confidence checks (block implementation when < 70%)\n- **Settings Profiles**: Could provide recommended permission/hook configs per workflow\n- **Native Session Persistence**: `--continue`/`--resume` instead of custom memory files\n\nSee `docs/user-guide/claude-code-integration.md` for the full gap analysis.\n"},"files":{"AGENTS.md":"# Repository Guidelines\n\n## Project Structure & Module Organization\n- `src/superclaude/` holds the Python package and pytest plugin entrypoints.\n- `tests/` contains Python integration/unit suites; markers map to features in `pyproject.toml`.\n- `pm/`, `research/`, and `index/` house TypeScript agents with standalone `package.json`.\n- `skills/` holds runtime skills (e.g., `confidence-check`); `commands/` documents scripted Claude commands.\n- `docs/` provides reference packs; start with `docs/developer-guide` for workflow expectations.\n\n## Build, Test, and Development Commands\n- `make install` installs the framework editable via `uv pip install -e \".[dev]\"`.\n- `make test` runs `uv run pytest` across `tests/`.\n- `make doctor` or `make verify` check CLI wiring and plugin health.\n- `make lint` and `make format` delegate to Ruff; run after significant edits.\n- TypeScript agents: inside `pm/`, run `npm install` once, then `npm test` or `npm run build`; repeat for `research/` and `index/`.\n\n## Coding Style & Naming Conventions\n- Python: 4-space indentation, Black line length 88, Ruff `E,F,I,N,W`; prefer snake_case for modules/functions and PascalCase for classes.\n- Keep pytest markers explicit (`@pytest.mark.unit`, etc.) and match file names `test_*.py`.\n- TypeScript: rely on project `tsconfig.json`; keep filenames kebab-case and exported classes PascalCase; align with existing PM agent modules.\n- Reserve docstrings or inline comments for non-obvious orchestration; let clear naming do the heavy lifting.\n\n## Testing Guidelines\n- Default to `make test`; add `uv run pytest -m unit` to scope runs during development.\n- When changes touch CLI or plugin startup, extend integration coverage in `tests/test_pytest_plugin.py`.\n- Respect coverage focus on `src/superclaude` (`tool.coverage.run`); adjust configuration instead of skipping logic.\n- For TypeScript agents, add Jest specs under `__tests__/*.test.ts` and keep coverage thresholds satisfied via `npm run test:coverage`.\n\n## Commit & Pull Request Guidelines\n- Follow Conventional Commits (`feat:`, `fix:`, `refactor:`) as seen in `git log`; keep present-tense summaries under ~72 chars.\n- Group related file updates per commit to simplify bisects and release notes.\n- Before opening a PR, run `make lint`, `make format`, and `make test`; include summaries of verification steps in the PR description.\n- Reference linked issues (`Closes #123`) and, for agent workflow changes, add brief reproduction notes; screenshots only when docs change.\n- Tag reviewers listed in `CODEOWNERS` when touching owned directories.\n\n## Plugin Deployment Tips\n- Use `make install-plugin` to mirror the development plugin into `~/.claude/plugins/pm-agent`; prefer `make reinstall-plugin` after local iterations.\n- Validate plugin detection with `make test-plugin` before sharing artifact links or release notes.\n","CLAUDE.md":"# CLAUDE.md\n\nThis file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.\n\n## 🐍 Python Environment Rules\n\n**CRITICAL**: This project uses **UV** for all Python operations. Never use `python -m`, `pip install`, or `python script.py` directly.\n\n### Required Commands\n\n```bash\n# All Python operations must use UV\nuv run pytest                    # Run tests\nuv run pytest tests/pm_agent/   # Run specific tests\nuv pip install package           # Install dependencies\nuv run python script.py          # Execute scripts\n```\n\n## 📂 Project Structure\n\n**Current v4.3.0 Architecture**: Python package with 30 commands, 20 agents, 7 modes\n\n```\n# Claude Code Configuration (v4.3.0)\n# Installed via `superclaude install` to user's home directory\n~/.claude/\n├── settings.json\n├── commands/sc/         # 30 slash commands (/sc:research, /sc:implement, etc.)\n│   ├── pm.md\n│   ├── research.md\n│   ├── implement.md\n│   └── ... (30 total)\n├── agents/              # 20 domain-specialist agents (@pm-agent, @system-architect, etc.)\n│   ├── pm-agent.md\n│   ├── system-architect.md\n│   └── ... (20 total)\n└── skills/              # Skills (confidence-check, etc.)\n\n# Python Package\nsrc/superclaude/\n├── __init__.py          # Public API: ConfidenceChecker, SelfCheckProtocol, ReflexionPattern\n├── pytest_plugin.py     # Auto-loaded pytest integration (5 fixtures, 9 markers)\n├── pm_agent/            # confidence.py, self_check.py, reflexion.py, token_budget.py\n├── execution/           # parallel.py, reflection.py, self_correction.py\n├── cli/                 # main.py, doctor.py, install_commands.py, install_mcp.py, install_skill.py\n├── commands/            # 30 slash command definitions (.md files)\n├── agents/              # 20 agent definitions (.md files)\n├── modes/               # 7 behavioral modes (.md files)\n├── skills/              # Installable skills (confidence-check, etc.)\n├── hooks/               # Claude Code hook definitions\n├── mcp/                 # MCP server configurations (10 servers)\n└── core/                # Core utilities\n\n# Project Files\ntests/                   # Python test suite (136 tests)\n├── unit/                # Unit tests (auto-marked @pytest.mark.unit)\n└── integration/         # Integration tests (auto-marked @pytest.mark.integration)\ndocs/                    # Documentation\nscripts/                 # Analysis tools (workflow metrics, A/B testing)\nplugins/                 # Exported plugin artefacts for distribution\nPLANNING.md              # Architecture, absolute rules\nTASK.md                  # Current tasks\nKNOWLEDGE.md             # Accumulated insights\n```\n\n### Claude Code Integration Points\n\nSuperClaude integrates with Claude Code through these mechanisms:\n- **Slash Commands**: 30 commands installed to `~/.claude/commands/sc/` (e.g., `/sc:pm`, `/sc:research`)\n- **Agents**: 20 agents installed to `~/.claude/agents/` (e.g., `@pm-agent`, `@system-architect`)\n- **Skills**: Installed to `~/.claude/skills/` (e.g., confidence-check)\n- **Hooks**: Session lifecycle hooks in `src/superclaude/hooks/`\n- **Settings**: Project settings in `.claude/settings.json`\n- **Pytest Plugin**: Auto-loaded via entry point, provides fixtures and markers\n- **MCP Servers**: 8+ servers configurable via `superclaude mcp`\n\n## 🔧 Development Workflow\n\n### Essential Commands\n\n```bash\n# Setup\nmake dev              # Install in editable mode with dev dependencies\nmake verify           # Verify installation (package, plugin, health)\n\n# Testing\nmake test             # Run full test suite\nuv run pytest tests/pm_agent/ -v              # Run specific directory\nuv run pytest tests/test_file.py -v           # Run specific file\nuv run pytest -m confidence_check             # Run by marker\nuv run pytest --cov=superclaude               # With coverage\n\n# Code Quality\nmake lint             # Run ruff linter\nmake format           # Format code with ruff\nmake doctor           # Health check diagnostics\n\n# MCP Servers\nsuperclaude mcp                              # Interactive install (gateway default)\nsuperclaude mcp --list                       # List available servers\nsuperclaude mcp --servers airis-mcp-gateway  # Install AIRIS Gateway (recommended)\nsuperclaude mcp --servers tavily context7    # Install individual servers\n\n# Plugin Packaging\nmake build-plugin            # Build plugin artefacts into dist/\nmake sync-plugin-repo        # Sync artefacts into ../SuperClaude_Plugin\n\n# Maintenance\nmake clean            # Remove build artifacts\n```\n\n## 📦 Core Architecture\n\n### Pytest Plugin (Auto-loaded)\n\nRegistered via `pyproject.toml` entry point, automatically available after installation.\n\n**Fixtures**: `confidence_checker`, `self_check_protocol`, `reflexion_pattern`, `token_budget`, `pm_context`\n\n**Auto-markers**:\n- Tests in `/unit/` → `@pytest.mark.unit`\n- Tests in `/integration/` → `@pytest.mark.integration`\n\n**Custom markers**: `@pytest.mark.confidence_check`, `@pytest.mark.self_check`, `@pytest.mark.reflexion`\n\n### PM Agent - Three Core Patterns\n\n**1. ConfidenceChecker** (src/superclaude/pm_agent/confidence.py)\n- Pre-execution confidence assessment: ≥90% required, 70-89% present alternatives, <70% ask questions\n- Prevents wrong-direction work, ROI: 25-250x token savings\n\n**2. SelfCheckProtocol** (src/superclaude/pm_agent/self_check.py)\n- Post-implementation evidence-based validation\n- No speculation - verify with tests/docs\n\n**3. ReflexionPattern** (src/superclaude/pm_agent/reflexion.py)\n- Error learning and prevention\n- Cross-session pattern matching\n\n### Parallel Execution\n\n**Wave → Checkpoint → Wave pattern** (src/superclaude/execution/parallel.py):\n- 3.5x faster than sequential execution\n- Automatic dependency analysis\n- Example: [Read files in parallel] → Analyze → [Edit files in parallel]\n\n### Slash Commands, Agents & Modes (v4.3.0)\n\n- Install via: `pipx install superclaude && superclaude install`\n- **30 Commands** installed to `~/.claude/commands/sc/` (e.g., `/sc:pm`, `/sc:research`, `/sc:implement`)\n- **20 Agents** installed to `~/.claude/agents/` (e.g., `@pm-agent`, `@system-architect`, `@deep-research`)\n- **7 Behavioral Modes**: Brainstorming, Business Panel, Deep Research, Introspection, Orchestration, Task Management, Token Efficiency\n- **Skills**: Installable to `~/.claude/skills/` (e.g., confidence-check)\n\n> **Note**: TypeScript plugin system planned for v5.0 ([#419](https://github.com/SuperClaude-Org/SuperClaude_Framework/issues/419))\n\n## 🧪 Testing with PM Agent\n\n### Example Test with Markers\n\n```python\n@pytest.mark.confidence_check\ndef test_feature(confidence_checker):\n    \"\"\"Pre-execution confidence check - skips if < 70%\"\"\"\n    context = {\"test_name\": \"test_feature\", \"has_official_docs\": True}\n    assert confidence_checker.assess(context) >= 0.7\n\n@pytest.mark.self_check\ndef test_implementation(self_check_protocol):\n    \"\"\"Post-implementation validation with evidence\"\"\"\n    implementation = {\"code\": \"...\", \"tests\": [...]}\n    passed, issues = self_check_protocol.validate(implementation)\n    assert passed, f\"Validation failed: {issues}\"\n\n@pytest.mark.reflexion\ndef test_error_learning(reflexion_pattern):\n    \"\"\"If test fails, reflexion records for future prevention\"\"\"\n    pass\n\n@pytest.mark.complexity(\"medium\")  # simple: 200, medium: 1000, complex: 2500\ndef test_with_budget(token_budget):\n    \"\"\"Token budget allocation\"\"\"\n    assert token_budget.limit == 1000\n```\n\n## 🌿 Git Workflow\n\n**Branch structure**: `master` (production) ← `integration` (testing) ← `feature/*`, `fix/*`, `docs/*`\n\n**Standard workflow**:\n1. Create branch from `integration`: `git checkout -b feature/your-feature`\n2. Develop with tests: `uv run pytest`\n3. Commit: `git commit -m \"feat: description\"` (conventional commits)\n4. Merge to `integration` → validate → merge to `master`\n\n**Current branch**: See git status in session start output\n\n### Parallel Development with Git Worktrees\n\n**CRITICAL**: When running multiple Claude Code sessions in parallel, use `git worktree` to avoid conflicts.\n\n```bash\n# Create worktree for integration branch\ncd ~/github/SuperClaude_Framework\ngit worktree add ../SuperClaude_Framework-integration integration\n\n# Create worktree for feature branch\ngit worktree add ../SuperClaude_Framework-feature feature/pm-agent\n```\n\n**Benefits**:\n- Run Claude Code sessions on different branches simultaneously\n- No branch switching conflicts\n- Independent working directories\n- Parallel development without state corruption\n\n**Usage**:\n- Session A: Open `~/github/SuperClaude_Framework/` (current branch)\n- Session B: Open `~/github/SuperClaude_Framework-integration/` (integration)\n- Session C: Open `~/github/SuperClaude_Framework-feature/` (feature branch)\n\n**Cleanup**:\n```bash\ngit worktree remove ../SuperClaude_Framework-integration\n```\n\n## 📝 Key Documentation Files\n\n**PLANNING.md** - Architecture, design principles, absolute rules\n**TASK.md** - Current tasks and priorities\n**KNOWLEDGE.md** - Accumulated insights and troubleshooting\n\nAdditional docs in `docs/user-guide/`, `docs/developer-guide/`, `docs/reference/`\n\n## 💡 Core Development Principles\n\n### 1. Evidence-Based Development\n**Never guess** - verify with official docs (Context7 MCP, WebFetch, WebSearch) before implementation.\n\n### 2. Confidence-First Implementation\nCheck confidence BEFORE starting: ≥90% proceed, 70-89% present alternatives, <70% ask questions.\n\n### 3. Parallel-First Execution\nUse **Wave → Checkpoint → Wave** pattern (3.5x faster). Example: `[Read files in parallel]` → Analyze → `[Edit files in parallel]`\n\n### 4. Token Efficiency\n- Simple (typo): 200 tokens\n- Medium (bug fix): 1,000 tokens\n- Complex (feature): 2,500 tokens\n- Confidence check ROI: spend 100-200 to save 5,000-50,000\n\n## 🔧 MCP Server Integration\n\n**Recommended**: Use **airis-mcp-gateway** for unified MCP management.\n\n```bash\nsuperclaude mcp  # Interactive install, gateway is default (requires Docker)\n```\n\n**Gateway Benefits**: 60+ tools, 98% token reduction, single SSE endpoint, Web UI\n\n**High Priority Servers** (included in gateway):\n- **Tavily**: Web search (Deep Research)\n- **Context7**: Official documentation (prevent hallucination)\n- **Sequential**: Token-efficient reasoning (30-50% reduction)\n- **Serena**: Session persistence\n- **Mindbase**: Cross-session learning\n\n**Optional**: Playwright (browser automation), Magic (UI components), Chrome DevTools (performance)\n\n**Usage**: TypeScript plugins and Python pytest plugin can call MCP servers. Always prefer MCP tools over speculation for documentation/research.\n\n## 🚀 Development & Installation\n\n### Current Installation Method (v4.3.0)\n\n**Standard Installation**:\n```bash\n# Option 1: pipx (recommended)\npipx install superclaude\nsuperclaude install\n\n# Option 2: Direct from repo\ngit clone https://github.com/SuperClaude-Org/SuperClaude_Framework.git\ncd SuperClaude_Framework\n./install.sh\n```\n\n**Development Mode**:\n```bash\n# Install in editable mode\nmake dev\n\n# Run tests\nmake test\n\n# Verify installation\nmake verify\n```\n\n### Plugin System (v5.0 - Not Yet Available)\n\nThe TypeScript plugin system (`.claude-plugin/`, marketplace) is planned for v5.0.\nSee `docs/plugin-reorg.md` for details.\n\n## 📊 Package Information\n\n**Package name**: `superclaude`\n**Version**: 4.3.0\n**Python**: >=3.10\n**Build system**: hatchling (PEP 517)\n\n**Entry points**:\n- CLI: `superclaude` command\n- Pytest plugin: Auto-loaded as `superclaude`\n\n**Dependencies**:\n- pytest>=7.0.0\n- click>=8.0.0\n- rich>=13.0.0\n\n## 🔌 Claude Code Native Features (for developers)\n\nSuperClaude extends Claude Code through its native extension points. When developing SuperClaude features, use these Claude Code capabilities:\n\n### Extension Points We Use\n- **Custom Commands** (`~/.claude/commands/sc/*.md`): 30 `/sc:*` commands\n- **Custom Agents** (`~/.claude/agents/*.md`): 20 domain-specialist agents\n- **Skills** (`~/.claude/skills/`): confidence-check skill\n- **Settings** (`.claude/settings.json`): Permission rules, hooks\n- **MCP Servers**: 8 pre-configured + AIRIS gateway\n- **Pytest Plugin**: Auto-loaded via entry point\n\n### Extension Points We Should Use More\n- **Hooks** (28 events): `SessionStart`, `Stop`, `PostToolUse`, `TaskCompleted` — ideal for PM Agent auto-restore, self-check validation, and reflexion triggers\n- **Skills System**: Commands should migrate to proper skills with YAML frontmatter for auto-triggering, tool restrictions, and effort overrides\n- **Plan Mode**: Could integrate with confidence checks (block implementation when < 70%)\n- **Settings Profiles**: Could provide recommended permission/hook configs per workflow\n- **Native Session Persistence**: `--continue`/`--resume` instead of custom memory files\n\nSee `docs/user-guide/claude-code-integration.md` for the full gap analysis.\n"},"items":[{"name":"AGENTS.md","path":"AGENTS.md","title":"AGENTS.md","content":"# Repository Guidelines\n\n## Project Structure & Module Organization\n- `src/superclaude/` holds the Python package and pytest plugin entrypoints.\n- `tests/` contains Python integration/unit suites; markers map to features in `pyproject.toml`.\n- `pm/`, `research/`, and `index/` house TypeScript agents with standalone `package.json`.\n- `skills/` holds runtime skills (e.g., `confidence-check`); `commands/` documents scripted Claude commands.\n- `docs/` provides reference packs; start with `docs/developer-guide` for workflow expectations.\n\n## Build, Test, and Development Commands\n- `make install` installs the framework editable via `uv pip install -e \".[dev]\"`.\n- `make test` runs `uv run pytest` across `tests/`.\n- `make doctor` or `make verify` check CLI wiring and plugin health.\n- `make lint` and `make format` delegate to Ruff; run after significant edits.\n- TypeScript agents: inside `pm/`, run `npm install` once, then `npm test` or `npm run build`; repeat for `research/` and `index/`.\n\n## Coding Style & Naming Conventions\n- Python: 4-space indentation, Black line length 88, Ruff `E,F,I,N,W`; prefer snake_case for modules/functions and PascalCase for classes.\n- Keep pytest markers explicit (`@pytest.mark.unit`, etc.) and match file names `test_*.py`.\n- TypeScript: rely on project `tsconfig.json`; keep filenames kebab-case and exported classes PascalCase; align with existing PM agent modules.\n- Reserve docstrings or inline comments for non-obvious orchestration; let clear naming do the heavy lifting.\n\n## Testing Guidelines\n- Default to `make test`; add `uv run pytest -m unit` to scope runs during development.\n- When changes touch CLI or plugin startup, extend integration coverage in `tests/test_pytest_plugin.py`.\n- Respect coverage focus on `src/superclaude` (`tool.coverage.run`); adjust configuration instead of skipping logic.\n- For TypeScript agents, add Jest specs under `__tests__/*.test.ts` and keep coverage thresholds satisfied via `npm run test:coverage`.\n\n## Commit & Pull Request Guidelines\n- Follow Conventional Commits (`feat:`, `fix:`, `refactor:`) as seen in `git log`; keep present-tense summaries under ~72 chars.\n- Group related file updates per commit to simplify bisects and release notes.\n- Before opening a PR, run `make lint`, `make format`, and `make test`; include summaries of verification steps in the PR description.\n- Reference linked issues (`Closes #123`) and, for agent workflow changes, add brief reproduction notes; screenshots only when docs change.\n- Tag reviewers listed in `CODEOWNERS` when touching owned directories.\n\n## Plugin Deployment Tips\n- Use `make install-plugin` to mirror the development plugin into `~/.claude/plugins/pm-agent`; prefer `make reinstall-plugin` after local iterations.\n- Validate plugin detection with `make test-plugin` before sharing artifact links or release notes.\n","category":"root","tokens":715},{"name":"CLAUDE.md","path":"CLAUDE.md","title":"CLAUDE.md","content":"# CLAUDE.md\n\nThis file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.\n\n## 🐍 Python Environment Rules\n\n**CRITICAL**: This project uses **UV** for all Python operations. Never use `python -m`, `pip install`, or `python script.py` directly.\n\n### Required Commands\n\n```bash\n# All Python operations must use UV\nuv run pytest                    # Run tests\nuv run pytest tests/pm_agent/   # Run specific tests\nuv pip install package           # Install dependencies\nuv run python script.py          # Execute scripts\n```\n\n## 📂 Project Structure\n\n**Current v4.3.0 Architecture**: Python package with 30 commands, 20 agents, 7 modes\n\n```\n# Claude Code Configuration (v4.3.0)\n# Installed via `superclaude install` to user's home directory\n~/.claude/\n├── settings.json\n├── commands/sc/         # 30 slash commands (/sc:research, /sc:implement, etc.)\n│   ├── pm.md\n│   ├── research.md\n│   ├── implement.md\n│   └── ... (30 total)\n├── agents/              # 20 domain-specialist agents (@pm-agent, @system-architect, etc.)\n│   ├── pm-agent.md\n│   ├── system-architect.md\n│   └── ... (20 total)\n└── skills/              # Skills (confidence-check, etc.)\n\n# Python Package\nsrc/superclaude/\n├── __init__.py          # Public API: ConfidenceChecker, SelfCheckProtocol, ReflexionPattern\n├── pytest_plugin.py     # Auto-loaded pytest integration (5 fixtures, 9 markers)\n├── pm_agent/            # confidence.py, self_check.py, reflexion.py, token_budget.py\n├── execution/           # parallel.py, reflection.py, self_correction.py\n├── cli/                 # main.py, doctor.py, install_commands.py, install_mcp.py, install_skill.py\n├── commands/            # 30 slash command definitions (.md files)\n├── agents/              # 20 agent definitions (.md files)\n├── modes/               # 7 behavioral modes (.md files)\n├── skills/              # Installable skills (confidence-check, etc.)\n├── hooks/               # Claude Code hook definitions\n├── mcp/                 # MCP server configurations (10 servers)\n└── core/                # Core utilities\n\n# Project Files\ntests/                   # Python test suite (136 tests)\n├── unit/                # Unit tests (auto-marked @pytest.mark.unit)\n└── integration/         # Integration tests (auto-marked @pytest.mark.integration)\ndocs/                    # Documentation\nscripts/                 # Analysis tools (workflow metrics, A/B testing)\nplugins/                 # Exported plugin artefacts for distribution\nPLANNING.md              # Architecture, absolute rules\nTASK.md                  # Current tasks\nKNOWLEDGE.md             # Accumulated insights\n```\n\n### Claude Code Integration Points\n\nSuperClaude integrates with Claude Code through these mechanisms:\n- **Slash Commands**: 30 commands installed to `~/.claude/commands/sc/` (e.g., `/sc:pm`, `/sc:research`)\n- **Agents**: 20 agents installed to `~/.claude/agents/` (e.g., `@pm-agent`, `@system-architect`)\n- **Skills**: Installed to `~/.claude/skills/` (e.g., confidence-check)\n- **Hooks**: Session lifecycle hooks in `src/superclaude/hooks/`\n- **Settings**: Project settings in `.claude/settings.json`\n- **Pytest Plugin**: Auto-loaded via entry point, provides fixtures and markers\n- **MCP Servers**: 8+ servers configurable via `superclaude mcp`\n\n## 🔧 Development Workflow\n\n### Essential Commands\n\n```bash\n# Setup\nmake dev              # Install in editable mode with dev dependencies\nmake verify           # Verify installation (package, plugin, health)\n\n# Testing\nmake test             # Run full test suite\nuv run pytest tests/pm_agent/ -v              # Run specific directory\nuv run pytest tests/test_file.py -v           # Run specific file\nuv run pytest -m confidence_check             # Run by marker\nuv run pytest --cov=superclaude               # With coverage\n\n# Code Quality\nmake lint             # Run ruff linter\nmake format           # Format code with ruff\nmake doctor           # Health check diagnostics\n\n# MCP Servers\nsuperclaude mcp                              # Interactive install (gateway default)\nsuperclaude mcp --list                       # List available servers\nsuperclaude mcp --servers airis-mcp-gateway  # Install AIRIS Gateway (recommended)\nsuperclaude mcp --servers tavily context7    # Install individual servers\n\n# Plugin Packaging\nmake build-plugin            # Build plugin artefacts into dist/\nmake sync-plugin-repo        # Sync artefacts into ../SuperClaude_Plugin\n\n# Maintenance\nmake clean            # Remove build artifacts\n```\n\n## 📦 Core Architecture\n\n### Pytest Plugin (Auto-loaded)\n\nRegistered via `pyproject.toml` entry point, automatically available after installation.\n\n**Fixtures**: `confidence_checker`, `self_check_protocol`, `reflexion_pattern`, `token_budget`, `pm_context`\n\n**Auto-markers**:\n- Tests in `/unit/` → `@pytest.mark.unit`\n- Tests in `/integration/` → `@pytest.mark.integration`\n\n**Custom markers**: `@pytest.mark.confidence_check`, `@pytest.mark.self_check`, `@pytest.mark.reflexion`\n\n### PM Agent - Three Core Patterns\n\n**1. ConfidenceChecker** (src/superclaude/pm_agent/confidence.py)\n- Pre-execution confidence assessment: ≥90% required, 70-89% present alternatives, <70% ask questions\n- Prevents wrong-direction work, ROI: 25-250x token savings\n\n**2. SelfCheckProtocol** (src/superclaude/pm_agent/self_check.py)\n- Post-implementation evidence-based validation\n- No speculation - verify with tests/docs\n\n**3. ReflexionPattern** (src/superclaude/pm_agent/reflexion.py)\n- Error learning and prevention\n- Cross-session pattern matching\n\n### Parallel Execution\n\n**Wave → Checkpoint → Wave pattern** (src/superclaude/execution/parallel.py):\n- 3.5x faster than sequential execution\n- Automatic dependency analysis\n- Example: [Read files in parallel] → Analyze → [Edit files in parallel]\n\n### Slash Commands, Agents & Modes (v4.3.0)\n\n- Install via: `pipx install superclaude && superclaude install`\n- **30 Commands** installed to `~/.claude/commands/sc/` (e.g., `/sc:pm`, `/sc:research`, `/sc:implement`)\n- **20 Agents** installed to `~/.claude/agents/` (e.g., `@pm-agent`, `@system-architect`, `@deep-research`)\n- **7 Behavioral Modes**: Brainstorming, Business Panel, Deep Research, Introspection, Orchestration, Task Management, Token Efficiency\n- **Skills**: Installable to `~/.claude/skills/` (e.g., confidence-check)\n\n> **Note**: TypeScript plugin system planned for v5.0 ([#419](https://github.com/SuperClaude-Org/SuperClaude_Framework/issues/419))\n\n## 🧪 Testing with PM Agent\n\n### Example Test with Markers\n\n```python\n@pytest.mark.confidence_check\ndef test_feature(confidence_checker):\n    \"\"\"Pre-execution confidence check - skips if < 70%\"\"\"\n    context = {\"test_name\": \"test_feature\", \"has_official_docs\": True}\n    assert confidence_checker.assess(context) >= 0.7\n\n@pytest.mark.self_check\ndef test_implementation(self_check_protocol):\n    \"\"\"Post-implementation validation with evidence\"\"\"\n    implementation = {\"code\": \"...\", \"tests\": [...]}\n    passed, issues = self_check_protocol.validate(implementation)\n    assert passed, f\"Validation failed: {issues}\"\n\n@pytest.mark.reflexion\ndef test_error_learning(reflexion_pattern):\n    \"\"\"If test fails, reflexion records for future prevention\"\"\"\n    pass\n\n@pytest.mark.complexity(\"medium\")  # simple: 200, medium: 1000, complex: 2500\ndef test_with_budget(token_budget):\n    \"\"\"Token budget allocation\"\"\"\n    assert token_budget.limit == 1000\n```\n\n## 🌿 Git Workflow\n\n**Branch structure**: `master` (production) ← `integration` (testing) ← `feature/*`, `fix/*`, `docs/*`\n\n**Standard workflow**:\n1. Create branch from `integration`: `git checkout -b feature/your-feature`\n2. Develop with tests: `uv run pytest`\n3. Commit: `git commit -m \"feat: description\"` (conventional commits)\n4. Merge to `integration` → validate → merge to `master`\n\n**Current branch**: See git status in session start output\n\n### Parallel Development with Git Worktrees\n\n**CRITICAL**: When running multiple Claude Code sessions in parallel, use `git worktree` to avoid conflicts.\n\n```bash\n# Create worktree for integration branch\ncd ~/github/SuperClaude_Framework\ngit worktree add ../SuperClaude_Framework-integration integration\n\n# Create worktree for feature branch\ngit worktree add ../SuperClaude_Framework-feature feature/pm-agent\n```\n\n**Benefits**:\n- Run Claude Code sessions on different branches simultaneously\n- No branch switching conflicts\n- Independent working directories\n- Parallel development without state corruption\n\n**Usage**:\n- Session A: Open `~/github/SuperClaude_Framework/` (current branch)\n- Session B: Open `~/github/SuperClaude_Framework-integration/` (integration)\n- Session C: Open `~/github/SuperClaude_Framework-feature/` (feature branch)\n\n**Cleanup**:\n```bash\ngit worktree remove ../SuperClaude_Framework-integration\n```\n\n## 📝 Key Documentation Files\n\n**PLANNING.md** - Architecture, design principles, absolute rules\n**TASK.md** - Current tasks and priorities\n**KNOWLEDGE.md** - Accumulated insights and troubleshooting\n\nAdditional docs in `docs/user-guide/`, `docs/developer-guide/`, `docs/reference/`\n\n## 💡 Core Development Principles\n\n### 1. Evidence-Based Development\n**Never guess** - verify with official docs (Context7 MCP, WebFetch, WebSearch) before implementation.\n\n### 2. Confidence-First Implementation\nCheck confidence BEFORE starting: ≥90% proceed, 70-89% present alternatives, <70% ask questions.\n\n### 3. Parallel-First Execution\nUse **Wave → Checkpoint → Wave** pattern (3.5x faster). Example: `[Read files in parallel]` → Analyze → `[Edit files in parallel]`\n\n### 4. Token Efficiency\n- Simple (typo): 200 tokens\n- Medium (bug fix): 1,000 tokens\n- Complex (feature): 2,500 tokens\n- Confidence check ROI: spend 100-200 to save 5,000-50,000\n\n## 🔧 MCP Server Integration\n\n**Recommended**: Use **airis-mcp-gateway** for unified MCP management.\n\n```bash\nsuperclaude mcp  # Interactive install, gateway is default (requires Docker)\n```\n\n**Gateway Benefits**: 60+ tools, 98% token reduction, single SSE endpoint, Web UI\n\n**High Priority Servers** (included in gateway):\n- **Tavily**: Web search (Deep Research)\n- **Context7**: Official documentation (prevent hallucination)\n- **Sequential**: Token-efficient reasoning (30-50% reduction)\n- **Serena**: Session persistence\n- **Mindbase**: Cross-session learning\n\n**Optional**: Playwright (browser automation), Magic (UI components), Chrome DevTools (performance)\n\n**Usage**: TypeScript plugins and Python pytest plugin can call MCP servers. Always prefer MCP tools over speculation for documentation/research.\n\n## 🚀 Development & Installation\n\n### Current Installation Method (v4.3.0)\n\n**Standard Installation**:\n```bash\n# Option 1: pipx (recommended)\npipx install superclaude\nsuperclaude install\n\n# Option 2: Direct from repo\ngit clone https://github.com/SuperClaude-Org/SuperClaude_Framework.git\ncd SuperClaude_Framework\n./install.sh\n```\n\n**Development Mode**:\n```bash\n# Install in editable mode\nmake dev\n\n# Run tests\nmake test\n\n# Verify installation\nmake verify\n```\n\n### Plugin System (v5.0 - Not Yet Available)\n\nThe TypeScript plugin system (`.claude-plugin/`, marketplace) is planned for v5.0.\nSee `docs/plugin-reorg.md` for details.\n\n## 📊 Package Information\n\n**Package name**: `superclaude`\n**Version**: 4.3.0\n**Python**: >=3.10\n**Build system**: hatchling (PEP 517)\n\n**Entry points**:\n- CLI: `superclaude` command\n- Pytest plugin: Auto-loaded as `superclaude`\n\n**Dependencies**:\n- pytest>=7.0.0\n- click>=8.0.0\n- rich>=13.0.0\n\n## 🔌 Claude Code Native Features (for developers)\n\nSuperClaude extends Claude Code through its native extension points. When developing SuperClaude features, use these Claude Code capabilities:\n\n### Extension Points We Use\n- **Custom Commands** (`~/.claude/commands/sc/*.md`): 30 `/sc:*` commands\n- **Custom Agents** (`~/.claude/agents/*.md`): 20 domain-specialist agents\n- **Skills** (`~/.claude/skills/`): confidence-check skill\n- **Settings** (`.claude/settings.json`): Permission rules, hooks\n- **MCP Servers**: 8 pre-configured + AIRIS gateway\n- **Pytest Plugin**: Auto-loaded via entry point\n\n### Extension Points We Should Use More\n- **Hooks** (28 events): `SessionStart`, `Stop`, `PostToolUse`, `TaskCompleted` — ideal for PM Agent auto-restore, self-check validation, and reflexion triggers\n- **Skills System**: Commands should migrate to proper skills with YAML frontmatter for auto-triggering, tool restrictions, and effort overrides\n- **Plan Mode**: Could integrate with confidence checks (block implementation when < 70%)\n- **Settings Profiles**: Could provide recommended permission/hook configs per workflow\n- **Native Session Persistence**: `--continue`/`--resume` instead of custom memory files\n\nSee `docs/user-guide/claude-code-integration.md` for the full gap analysis.\n","category":"root","tokens":3190}]}