Loop Design

Turn a rough goal or loop idea into clear instructions an agent can follow.

A goal is the smallest loop: keep acting until the outcome is proven or blocked. A loop is a goal with explicit triggers, gates, state, and stop rules.

Use this when the user wants Claude Code, Codex, a scheduled prompt, or an issue-tracker workflow to keep working across iterations. Do not use it for one-off edits, simple answers, or tasks with no meaningful verifier.

Workflow

1. Ground The Loop

Read the user's brief and any referenced issue, spec, plan, runbook, code, tests, docs, or tracker state.

If the loop depends on current Claude Code or Codex behavior, verify the current official guidance before making tool-specific claims.

Ask only when missing information changes the finish line, verifier, safety boundary, external side effect, or blocked condition.