~/wiki / github / goal-maker-codex-skill-pm-loop

Goal Maker: Skill for Codex that turns vague long-term goals into a manageable PM cycle

◷ 4 min read 5/4/2026

Main chat

A chat for vibe coders: news, guides, live cases, marketplace, and finding executors.

$ cd section/ $ join vibe dev

Working with large, long-term tasks in Codex (or similar AI agents) is a classic challenge. The goal begins blurred, the agent quickly dives into the implementation, verification becomes obsolete, and the entire process loses direction.

Tolibear’s Goal Maker is a specialized Codex Skill that introduces a clear finite-state PM loop. It turns any large goal into a structured task board with Scout, Judge, Worker roles and mandatory reports.

What does Goal Maker decide?

  • Drift of targets - the agent stops wandering and constantly returns to charter.
  • Premature implementation – there is always a Scout and a Judge before coding.
  • Lack of evidence – each task completed leaves a compact receipt.
  • Obsolescence of verification - Board always contains the current state.

Codex works like a real project manager with a clear process.

Basic model and roles

Goal Maker is built around four key primitives:

  • Charter (goal.md) - description of the purpose, limitations, current tranche (stage) and stop rule.
  • Board (state.yaml) is the only source of truth about tasks, statuses and receipts.
  • Task – At any given time, only one task is active.
  • **Receipt - Compact proof of what was done, why the task was completed/blocked or escalated.

** Roles (agents):**

  • Scout - scout: analyzes the repository, specifications, finds candidates for tasks.
  • Judge: evaluates priorities, risks, scope, verifies completion.
  • *Worker: Performs one clearly defined task.
  • PM is the main stream of /goal, which owns the board, selects the next task and leads the process.

How it works

  1. Install Skill:

    bash
    npx goal-maker
  2. In Codex, call:

    code
    $goal-maker

Skill creates a docs/goals/<slug>/ folder with goal.md and state.yaml.

  1. Start the main cycle:
    code
    /goal Follow docs/goals/<slug>/goal.md

Codex also operates in a cycle: Blurred Purpose → Discovery (Scout) → Board → One Active Task → Receipt → Board Update → Repeat

Target folder structure

code
docs/goals/<slug>/
Goal.md # Charter
─state.yaml # Taskboard + receipts
ки ─ Notes/ # Additional major notes

Most of the results are stored directly in state.yaml. Large reports are made in notes/.

Advantages

  • Great for multi-hour and multi-day purposes (the author mentioned a record of 71 hours).
  • Discipline of the agent: does not allow you to write code for a large purpose.
  • Transparency – you can always see what’s done and why.
  • It is easy to check the status of the board through the built-in script.
  • Version 0.2.x with improved receipts (v1 not supported).

Who's good for Goal Maker?

  • Developers who use Codex for serious refactoring, project improvement, or long-term features.
  • Those who want to turn AI from a “fast coder” to a real junior/senior developer running PM.
  • Teams and solo developers working with large codebases.

Conclusion

Goal Maker is a powerful tool for those who are tired of chaotic long sessions with AI. Instead of writing me a big feature, you get a structured, controlled process with intelligence, risk assessment, and proof of work done.

Reference to repository: https://github.com/tolibear/goal-maker

$ cd ../ ← back to GitHub