SkillFoxx

qiaomu-goal-meta-skill

A meta-skill for Codex: a vague task becomes a ready "/goal" command with verification, boundaries and pause conditions.

  • 910 ★ on the repository
  • Last commit
  • MIT
Skill

Minimal access

Why this level

  • The skill drafts goal text and runs a local check script; it does not execute the goal or go online.
All reasons and checks

joeseesun/qiaomu-goal-meta-skillAuthor:Qiaomu

Install

In your terminal, with SkillFoxx CLI

npx skillfoxx add skills/qiaomu-goal-meta-skill

Detects the agents on your machine, shows the access level and pins the version.

Other ways to install

Access and checks

How we review

Minimal accessThe tool runs nothing and makes no network calls.

  • The skill drafts goal text and runs a local check script; it does not execute the goal or go online.

This is third-party code. Review the repository files before installing.

What it does

qiaomu-goal-meta-skill writes a `/goal` command for Codex from a phrase like "build an app" or "fix a bug". The skill does not ask you to fill out a form: it takes conservative defaults and immediately gives a recommended version you can copy. The goal text has an outcome, a way to verify, constraints, a write scope, iteration rules, a completion condition and pause conditions. After the command comes one sentence with the reason for the choice and numbered options like "1A 2B 3B 4A" when a decision truly matters. The English mirror keeps the fields `Verification`, `Constraints`, `Boundaries`, `Iteration policy`, `Stop when` and `Pause if`. The skill itself does not carry out the goal. Vague words such as "premium feel" it turns into checkable steps: screenshots, hierarchy, fonts and no more than three rounds of fixes.

Who it is for. Codex users who give the agent long tasks and lose time on clarifications because the request has no verification or boundaries.

Good fit when

  • The idea sounds like "make a GTA-like game" and you need a safe goal without third-party assets.
  • A bug-fix task should go with a check instead of a promise that everything works.
  • A release must not hit the main branch without your word, and that belongs in a pause condition.
  • The domain is unfamiliar: medicine, finance, compliance, and the first step must be studying the project.
  • A team goal must read in both Chinese and English.

Not a fit when

  • The task is a single phrase: a translation, a simple text edit or one shell line.
  • The result needs neither verification nor many steps, so a `/goal` command would be overkill.
  • You are not working in Codex: the goal format is written for its `/goal`.

Example request

I want to build an iOS teleprompter app. Write a `/goal` for it with verification and pause conditions.

Limitations

The skill only creates the text of a `/goal` command and by default aims at MVP level; full product strategy, monetisation and deployment need a separate request. Installing needs Node.js with `npx` and an environment that reads `~/.agents/skills`. If the output starts with `/目标` instead of `/goal`, ask it to keep the `/goal` prefix. The documentation is in Chinese and the English part is short. The `lint_goal_command.py` check catches placeholders and empty conditions but does not judge the meaning of the goal. The last commit was made in June 2026.

How to disable. Delete the `qiaomu-goal-meta-skill` folder from `~/.agents/skills`. Goals already written stay as plain text.

FAQ

How do I install the skill for Codex?

Run `npx skills add joeseesun/qiaomu-goal-meta-skill`, then check the file with `test -f ~/.agents/skills/qiaomu-goal-meta-skill/SKILL.md`. It works in Codex, Claude Code and other agents that support Agent Skills.

What comes out?

A "recommended version" block with the `/goal` command, one sentence on why it was chosen, numbered options and an English mirror. The command can be copied whole.

When does the skill stop the agent?

It writes into pause conditions access and payments, production data, destructive operations, legal, medical and financial decisions, third-party assets, publishing and an unclear project owner.

What does the skill check in a finished goal?

The script `lint_goal_command.py goal.txt` rejects the `/目标` prefix, `[Outcome]` and `TODO` placeholders, phrases like `make sure it works`, `edit anything` and `keep trying`, and verification without evidence.

What about an unfamiliar domain?

The skill builds a discovery-first goal: the agent first reads project documents, sample data and scripts, states working assumptions and only then delivers a minimal checkable slice.

Discussion

No questions or reviews yet.

Skills

A skill that builds a Codex /goal command from a five-section template so a long task does not drift and passes the audit

SkillMinimal access231

CodexPotter loops Codex with a clean context each round until the code matches the goal, as a replacement for the built-in /goal.

SkillBroad access636

A Codex plugin of five independent meta-skills and one tldraw skill: improves prompts, goals, research, plans and commits.

PluginExtended access227

Skill setting up Markdown file memory for Codex via AGENTS.md: profile, rules, lessons, errors and capability requests.

SkillMinimal access132

A skill that turns a vague request into a Goal Prompt with success criteria and stop conditions, plus a Loop Prompt if needed.

SkillMinimal access57

A meta-skill that turns a workflow into an agent skill: prior-art search, trigger evaluation and publication to GitHub

SkillBroad access391

README in short

The README explains why agents fail on weak goals, shows a finished `/goal` example for a game, covers the default principles and gives a troubleshooting table. An English section is included.

README badge

Original SKILL.md

---
name: qiaomu-goal-meta-skill
description: |
  Turn vague or complex Codex tasks into strong `/goal` commands with outcome, verification, constraints, boundaries, iteration policy, completion evidence, and pause/block conditions. Use when the user asks for Codex goal instructions, Goal 指令, 目标指令, `/goal` prompts, 中文 Goal 模板, plan-to-goal interviews, success criteria, verification commands, or bounded agent work definitions.
---

# Qiaomu Goal Meta Skill

把一个模糊任务，收敛成 Codex 可以持续执行、可以验证、知道何时停止和何时暂停的 `/goal` 指令。

Copyright (c) 向阳乔木
X: https://x.com/vista8
GitHub: https://github.com/joeseesun/

## Operating Mode

Run as a production-lite meta skill.

Default assumptions:

- The user wants a paste-ready Codex `/goal` command, not a general prompt.
- The executable slash command stays `/goal`. Do not output `/目标` as the command unless the user's environment explicitly documents that alias.
- The first goal block should be the best recommended executable version, not a half-filled template. Users often copy the first draft directly.
- For Chinese users, output Chinese content and Chinese field names by default while keeping the command prefix `/goal`.
- For Chinese users, include both `推荐执行版（中文，可直接复制）` and `Goal Draft (English-compatible)` unless the user asks for one language only.
- If the task is still vague but low-risk, choose the best conservative defaults and continue. Ask only when the answer changes cost, risk, ownership, or product direction.
- If the domain is unfamiliar or specialized, create a discovery-first goal that makes the agent inspect authoritative project/docs/runtime evidence before implementation instead of inventing domain rules.
- If the missing detail is low-risk, make an explicit assumption and continue.
- Do not start the work described by the goal unless the user explicitly asks. This skill creates the goal instruction.
- Prefer concrete verification commands and artifacts over vague confidence phrases.
- Prefer narrow write boundaries and explicit forbidden paths over broad permission.
- Treat `Stop when` and `Pause if` as part of the same completion/blocking contract.

## Workflow

1. Restate the task as an outcome, not an activity.
2. Classify the task using `references/default-goal-strategy.md`: familiar vs unknown domain, low vs high risk, new work vs existing project.
3. Choose best defaults for low-risk unknowns and write a one-sentence reason.
4. Identify missing information across the Goal contract:
   - success criteria
   - verification commands, artifacts, or evidence
   - constraints that must not change
   - allowed writes and forbidden paths
   - iteration policy
   - completion evidence

Report a problem