oh-my-opencode-slim/src/hooks/loop-command/codemap.md

src/hooks/loop-command/

Responsibility

Implements the /loop runtime command as an automated execute-verify-retry loop. The hook extracts goal, successCriteria, and maxAttempts from the user's command text, then instructs the agent to iterate with on-disk per-run history under .opencode/loop-history/.

Design

Core Architecture

• createLoopCommandHook() — produces a command hook that registers /loop with registerCommandHook from command-hook-utils, and intercepts command.execute.before to rewrite the output.
• History directory — each invocation generates a unique .opencode/loop-history/loop-<timestamp>-<shortid>/ path; the agent writes history-NNN.md results into that directory.
• Safety defaults — if the user omits arguments, the hook emits help text instead of activating. Missing or unclear goal/successCriteria/maxAttempts causes a clarification push-back prompt.

Flow

/loop activation

1. User types `/loop <description>`
2. OpenCode invokes command.execute.before for 'loop'
3. Hook clears output and injects the structured loop prompt
4. Agent executes per-attempt: read history, dispatch @fixer, verify, write result
5. History persisted to `.opencode/loop-history/`
6. Loop stops on PASS or exits on FAIL after maxAttempts

Integration

Consumers

• Command Registration Utility (src/hooks/command-hook-utils.ts) — provides registerCommandHook
• OpenCode runtime — command.execute.before interception
• Agent workspace — .opencode/loop-history/ filesystem for loop history

Dependencies

• src/utils/internal-initiator.ts — createInternalAgentTextPart
• src/hooks/command-hook-utils.ts — command registration helper

Testing

• src/hooks/loop-command/index.test.ts