Skip to content

AGENTS.md

Latest commit

 

History

History
35 lines (24 loc) · 1.85 KB

AGENTS.md

File metadata and controls

35 lines (24 loc) · 1.85 KB
Raw

Agent Instructions for Loop Goal Runner

This repository packages loop-goal-runner, a stateful Goal mode skill for Codex, Claude Code, and AI agents.

When To Use This Repository

Use this repository when an agent needs to run /goal, Goal mode, recurring work, stateful long-running execution, or "continue until done" tasks with durable memory and objective verification.

Core Contract

  • Read skills/loop-goal-runner/SKILL.md before changing the Codex skill.
  • Read claude-code/loop-goal-runner/SKILL.md before changing the Claude Code skill.
  • Keep the Codex and Claude Code copies behaviorally aligned unless a runtime-specific difference is documented.
  • Keep STATE.md goal-scoped in examples. Do not encourage unrelated goals to share state.
  • Never claim completion in docs or examples without a concrete gate such as tests, lint, typecheck, build, data validation, or an explicit manual verification step.
  • Treat credentials, external services, destructive actions, production deploys, auth changes, and payments as human approval boundaries.

Public Documentation Rules

  • The default README is English.
  • Keep a link to README.zh-CN.md near the top of README.md.
  • Keep a link back to README.md near the top of README.zh-CN.md.
  • Do not add real local machine paths, personal usernames, personal emails, tokens, cookies, OAuth credentials, browser profile paths, screenshots, logs, snapshots, or .env files.
  • Use placeholders such as <PROJECT_DIR>, <USER_HOME>, example.com, and your-token-here.
  • When citing public figures or external projects, phrase them as industry context or references, not endorsements.

Validation

Before publishing or handing this repository to another agent, run:

powershell -ExecutionPolicy Bypass -File .\scripts\validate-repo.ps1

Fix any reported privacy or packaging issue before release.