A collection of prompt templates for AI-assisted development workflows. Includes agents, commands, and skills for Claude Code and other AI coding assistants.
prompts/
├── agents/ # Agent role definitions (planning, testing, research, etc.)
├── commands/ # Workflow commands (/plan, /act, /chart, etc.)
├── skills/ # Domain-specific knowledge (testing, deployment, etc.)
└── legacy/ # Original claude-desktop and cline instructions
├── claude-desktop/
└── cline/
Role definitions for specialized tasks:
- planning-agent - Complex planning with codebase analysis
- testing-agent - E2E testing with Playwright (token-efficient)
- coding-agent - Implementation with isolated tracking
- analysis-agent - Deep multi-file investigation
- research-agent - Web searches for examples
- reference-agent - Library source code analysis
Workflow commands for development cycles:
| Command | Purpose |
|---|---|
/prepare |
Load context (memory bank + git + issues) |
/plan |
Create implementation plan with file tree |
/elaborate |
Get file-level implementation details |
/act |
Execute the plan systematically |
/chart |
Update tracking systems (close the loop) |
/research |
Research a topic with web search |
/navigate |
Codebase navigation helper |
Plus 40+ more commands for git, GitHub issues, deployment, etc.
Domain-specific knowledge that auto-activates based on task:
- planning - Feature planning with execution tasks
- testing - E2E testing workflow with deployment
- quiet-build - Filtered build output (reduces context noise)
- quiet-deploy - Filtered deploy output
- playwright-cli - Browser automation patterns
- slack-agents - Slack bot development
- vnc-tunnel - Remote GUI automation
- And more...
Reference prompts directly in your AI assistant:
# With pi coding agent
pi @~/code/prompts/commands/plan.md "add user authentication"
# Or copy to your preferred location
cp -r prompts/* ~/.prompts/The legacy/ directory contains the original Claude Desktop and Cline instruction sets that informed the evolution of these patterns. Preserved for reference.
MIT