Agent Loops

Run protocol

How to use these loops

The current product is deliberately safe: it does not write files into your repository from the browser. The main action is Copy Markdown, which gives your AI agent a complete execution protocol.

Copy Markdown
Copies a complete protocol with what the loop is, required context, goal, check command, exit condition, steps, guardrails, source, and expected final response.
Copy Prompt
Copies a shorter kickoff prompt when you want a lightweight instruction instead of the full Markdown protocol.
Hook bundles later
Some loops are hook-bundle-ready, but the MVP keeps installation manual until D1 review and source safety are implemented.
Safety boundary
Always review destructive commands, secret exposure, dependency install scripts, and deployment actions before allowing an agent to run them.

Recommended loop

  1. Open a loop detail page.
  2. Read when to use, when not to use, required tools, exit condition, and guardrails.
  3. Click Copy Markdown.
  4. Paste into Cursor, Claude Code, Codex, Gemini CLI, OpenCode, or your preferred coding agent.
  5. Give any missing repository-specific context, such as the real test command or deployment URL.
  6. Require the agent to report evidence before it claims completion.