ViggyV/agent-help

Agent Help

I'm using the agent-help skill to explain the agent pipeline system.

Quick Reference

Available Commands

| Command | Agent | What It Does | |---------|-------|-------------| | /code | Coding Agent | Full-stack builds, quality audits, prototypes, debugging | | /code full-stack | Coding Agent | brainstorm → plan → parallel build → review → finish | | /code quality | Coding Agent | code-review → bugs → security → tests → perf → verify | | /code prototype | Coding Agent | api → db → FastAPI → React → Docker | | /code debug | Coding Agent | debug → root-cause → fix → test → verify | | /docs | Docs Agent | API docs, project scaffolding, content, releases | | /docs api | Docs Agent | documenter → code-docs → examples → OpenAPI | | /docs project | Docs Agent | CLAUDE.md → ARCHITECTURE → DEVLOG → DOMAIN | | /docs content | Docs Agent | research → parallel(Medium, LinkedIn, newsletter) → polish | | /docs release | Docs Agent | PRs → commits → changelog → release-notes | | /startup | Startup Agent | MVPs, go-to-market, pitch decks, analytics | | /startup mvp | Startup Agent | brainstorm → Hormozi → plan → parallel build → deploy | | /startup gtm | Startup Agent | research → pitch → parallel(5 channels) → email | | /startup pitch | Startup Agent | Hormozi → coach → PPTX → outreach email | | /startup analytics | Startup Agent | metrics → pipeline → D3.js → PPTX report |

How It Works

1. Say /code, /docs, or /startup (optionally with a pipeline name)
2. The super-agent asks what you want to build and which mode to use
3. It loads the pipeline JSON and invokes the pipeline-orchestrator
4. The orchestrator runs each stage, passing outputs between them
5. Gates pause for approval, run code review, or require verification
6. On completion, you get a summary of all outputs and next steps

Modes

• Interactive (default): Pauses at every gate for your approval
• Autonomous: Skips approval gates, still enforces quality and verification

Creating Custom Pipelines

Read the full guide: agents/GUIDE.md

1. Create a JSON file in agents/<agent>/references/pipelines/<name>.json
2. Define stages with skill, skill-chain, or parallel types
3. Add gates (approval, quality, verification) between stages
4. Declare outputs and inputs for data passing
5. Update the agent's SKILL.md routing table

Troubleshooting

• Pipeline won't start: Check SKILL.md exists, JSON is valid, restart session
• Stage fails: Check skill path exists, verify input files are on disk
• Parallel conflict: Ensure branches write to different output paths
• Gate blocks: Fix Critical issues from quality gate, provide evidence for verification

For the complete guide with architecture diagrams, JSON schema, examples, and custom pipeline tutorial, read agents/GUIDE.md.