@fermindi/pwn-cli 0.1.0

package/templates/workspace/.ai/agents/README.md

@@ -0,0 +1,204 @@
+# AI Agent Configuration
+
+This directory contains bootstrap instructions and configuration for different AI agents working on this project.
+
+## Overview
+
+Different AI agents (Claude, Cursor, Copilot, etc.) need specific instructions to work effectively with PWN framework and project context.
+
+Each agent has:
+- **`<agent>.md`** - Complete bootstrap instructions for that agent
+- **Auto-load triggers** defined in `patterns/index.md`
+- **Context loading order** for efficient session start
+- **Task execution mode** (batch vs. interactive)
+
+## Available Agents
+
+### Claude Code (`claude.md`)
+- **Status:** Primary agent
+- **Recommended for:** Full-stack development, complex tasks, refactoring
+- **Session model:** Interactive or batch (autonomous)
+- **Model:** Claude Opus 4.5 (recommended)
+
+### Cursor (Coming Soon)
+- **Status:** Planned
+- **Recommended for:** Single-file editing, rapid iteration
+- **Session model:** Interactive
+- **Model:** Latest Claude via Cursor
+
+### GitHub Copilot (Coming Soon)
+- **Status:** Planned
+- **Recommended for:** In-IDE completion, pair programming
+- **Session model:** Interactive
+- **Model:** Copilot Plus
+
+## Bootstrap Protocol
+
+Every agent follows this sequence on session start:
+
+1. **Load Session State**
+   - Read `/.ai/state.json` for current session info
+   - Identify developer and session mode
+
+2. **Load Shared Context**
+   - Read `memory/decisions.md` for architectural decisions
+   - Read `memory/patterns.md` for discovered patterns
+   - Read `memory/deadends.md` to avoid known failures
+
+3. **Load Task Context**
+   - Read `tasks/active.md` for current work
+   - Optionally read `tasks/backlog.md` for planning
+
+4. **Register Auto-Apply Triggers**
+   - Read `patterns/index.md`
+   - Register file watchers and import triggers
+   - Load relevant patterns as needed
+
+5. **Load Agent-Specific Instructions**
+   - Read `agents/<agent>.md` for specific bootstrap
+   - Configure agent settings from `state.json`
+   - Prepare for interactive or batch mode
+
+6. **Acknowledge Understanding**
+   - Display what context was loaded
+   - Show current task or available work
+   - Ready to accept commands
+
+## Agent-Specific Features
+
+### Claude Code
+- Full filesystem access
+- Git repository integration
+- Browser automation (Playwright)
+- Network requests
+- File editing and creation
+- Batch task execution
+- Pattern auto-application
+
+### Cursor
+- Single file focus
+- In-editor quick-start
+- Codebase navigation
+- Inline suggestions
+- Git integration (basic)
+
+### Copilot
+- In-IDE suggestions
+- Autocomplete
+- Ghost text
+- Chat interface
+- Limited filesystem access
+
+## Adding New Agents
+
+To add a new AI agent:
+
+1. Create `agents/<agent-name>.md` with complete bootstrap instructions
+2. Follow the template structure in `claude.md`
+3. Include agent-specific capabilities and limitations
+4. Define recommended use cases
+5. Add instructions for pattern auto-application
+6. Document any configuration needed in `state.json`
+7. Commit with message: `docs: add <agent-name> agent bootstrap`
+
+### Template Structure
+
+```markdown
+# [Agent Name] Bootstrap
+
+## Overview
+- What the agent is and does
+- Recommended use cases
+- Supported models
+
+## Session Start Protocol
+1. Step 1
+2. Step 2
+3. ...
+
+## Configuration
+- Settings in state.json
+- Environment variables
+- Installation requirements
+
+## Capabilities
+- What it can do
+- Limitations
+- Best practices
+
+## Examples
+- Example 1: Typical workflow
+- Example 2: Complex scenario
+
+## Troubleshooting
+- Common issues
+- Solutions
+```
+
+## Workflow Coordination
+
+Multiple agents can work on same project:
+
+- **Serial workflow:** Agent A finishes, Agent B continues
+  - Commit state before switching agents
+  - New agent reads latest context and tasks
+
+- **Parallel workflow:** Different agents on different tasks
+  - Each agent has separate active task
+  - Coordinate via `tasks/active.md`
+  - Merge regularly to avoid conflicts
+
+## State Sharing
+
+All agents share:
+
+```
+tasks/active.md          ← What's being worked on
+tasks/backlog.md         ← What's next
+memory/decisions.md      ← What was decided
+memory/patterns.md       ← What was learned
+memory/deadends.md       ← What failed
+```
+
+Each agent has personal state in:
+
+```
+.ai/state.json          ← Personal session state (git-ignored)
+.ai/config/             ← Agent-specific configs (git-ignored)
+```
+
+## Configuration Format
+
+Agent configuration in `state.json`:
+
+```json
+{
+  "agent": "claude",
+  "model": "claude-opus-4-5",
+  "session_mode": "interactive",
+  "auto_patterns": true,
+  "batch_config": { ... },
+  "developer": "username"
+}
+```
+
+## Tips for Agent Integration
+
+1. **Keep context compact** - Link to detailed docs, don't embed everything
+2. **Make decisions explicit** - Use `memory/decisions.md` for reasoning
+3. **Track failures** - Document dead ends to avoid repetition
+4. **Commit frequently** - Small commits are easier to understand
+5. **Update patterns** - Capture learned patterns for future use
+6. **Test quality gates** - Automate verification to catch issues early
+
+## Links
+
+- [Claude Code Bootstrap](claude.md) - Full instructions for Claude
+- [Patterns Auto-Apply System](../patterns/index.md) - How triggers work
+- [Batch Execution Workflow](../workflows/batch-task.md) - Autonomous task execution
+
+## Version
+
+**Last Updated:** (Set on template injection)
+**Supported Agents:** Claude, Cursor (planned), Copilot (planned)
+**Framework:** PWN 1.0.0