Specialized AI agents that orchestrate your entire feature development workflow - from PRD to production.
Your AI engineering team that ships features, not just code. From PRD to production, these specialized agents handle tech shaping, implementation, testing, review, optimization, and documentation—autonomously executing the full development workflow while you focus on what to build, not how to build it.
The quality of your output is directly proportional to the quality of your input. Give agents clear requirements, context, and goals—they'll handle the rest.
- Install: Claude Code + Anthropic API key
- Run:
./scripts/install.sh
Tracks checksums so re-running safely updates changed files without clobbering local edits. See ./scripts/install.sh --help for --status, --uninstall, --dry-run, and --verbose.
| Agent | Purpose | Model | Features |
|---|---|---|---|
| planner | Design implementation plans and architecture decisions | Opus | read-only |
| engineer | Implement features following existing codebase patterns | Opus | |
| debugger | Systematic bug investigation and verified fixes | Opus | |
| tester | Write unit tests for new functionality | Sonnet | |
| reviewer | Code review with confidence-based filtering (≥80 threshold) | Opus | read-only |
| security | Security audit — OWASP Top 10, auth flows, dependency risks | Opus | read-only |
| researcher | Explore codebases, compare technologies, gather information | Opus | |
| gap-finder | Verify implementations match specs, find missing requirements | Opus | read-only |
| optimizer | Practical code improvements and refactoring | Sonnet | |
| documenter | Create and maintain documentation, diagrams, and guides | Sonnet | |
| context-auditor | Audit markdown docs for token efficiency and redundancy | Sonnet | read-only |
All agents use project-scoped memory (memory: local) to learn codebase patterns across sessions without cross-project contamination.
Bug fix:
debugger → engineer → tester → reviewer
New feature:
planner → engineer → tester → gap-finder → reviewer
Security review:
security → reviewer
Documentation:
documenter (standalone or after feature work)
Each agent works autonomously and returns results. Claude Code decides which agent to invoke next.
| Command | Description |
|---|---|
/contribution-report |
GitHub contribution summaries for performance reviews |
/pr-description |
Generate PR title and description from branch changes |
/upgrade-dep |
Upgrade a dependency and fix breaking changes |
| Script | Description |
|---|---|
scripts/install.sh |
Install/update/uninstall agents, commands, and scripts with checksum-based conflict detection |
scripts/count-tokens.sh |
Token counting with Anthropic API (caches results, falls back to estimation) |