llms.txt

# Loop Goal Runner

Loop Goal Runner is a stateful Goal mode skill for Codex, Claude Code, and AI agents. Use it for `/goal`, long-running tasks, recurring agent work, "continue until done", or any workflow that needs durable `STATE.md` memory, objective verification gates, and explicit stop rules.

Primary entry points:

- `README.md`: English default README with setup, advantages, industry context, similar project comparison, license, and support.
- `README.zh-CN.md`: Chinese README.
- `AGENTS.md`: Public agent instructions for maintaining this repository.
- `skills/loop-goal-runner/SKILL.md`: Codex skill.
- `claude-code/loop-goal-runner/SKILL.md`: Claude Code skill.
- `skills/loop-goal-runner/references/state-template.md`: Goal-scoped `STATE.md` template.
- `skills/loop-goal-runner/references/agents-snippet.md`: Optional trigger reliability snippet.
- `docs/AI_AGENT_GUIDE.md`: Detailed operational guide for AI agents.
- `scripts/validate-repo.ps1`: Repository validation and privacy scan.

Key contract:

1. Shape a measurable goal.
2. Read or create a goal-scoped `STATE.md`.
3. Run one smallest useful iteration.
4. Verify with a fresh objective gate.
5. Update state with actions, verification, blockers, lessons, and next action.
6. Stop on success, repeated blocker, budget limit, permissions, credentials, risky production work, or human judgment.