Track

Track is a lightweight, human-guided orchestration skill for complex work. It keeps the goal stable while tasks, dependencies, tracks, and working sessions change around it.

Version: 0.1.10 Codex skill: $track

Open the English landing page · 한국어 랜딩 페이지 · Read the skill instructions

What Track does

• preserves the original goal while the execution plan evolves;
• separates sequential dependencies from work that can safely run in parallel;
• applies a small over-engineering gate before adding work or expanding scope;
• treats tracks as stable workstreams and sessions as replaceable resources;
• accepts progress only after inspecting outputs and verification evidence;
• keeps one development map current instead of producing disconnected plans.

Use Track for multi-stage product work, research programs, migrations, or any project where several workstreams must meet at shared gates. A simple one-step task usually does not need it.

Track example: ship an onboarding beta

Ask Codex:

Use $track to coordinate the onboarding beta. Reuse the product work that already exists, split independent work, keep one development map, and stop at a verified beta release.

Track shows independent work as parallel lanes, then makes each handoff and shared gate visible on the same timeline:

sequenceDiagram
    participant G as Goal / scope
    participant P as Product
    participant M as Metrics
    participant S as Support
    participant I as Integration gate

    G->>G: Accept minimal beta scope
    par Product flow
        G->>P: Start product work
        P-->>I: Submit flow evidence
    and Success metrics
        G->>M: Reuse completed metrics
        M-->>I: Submit metrics evidence
    and Support readiness
        G->>S: Start support preparation
        S-->>I: Submit readiness evidence
    end
    Note over P,S: Independent tracks move together
    I->>I: Inspect all required evidence
    alt Evidence passes
        I-->>G: Unlock beta release
    else Issue found
        I-->>P: Return affected work only
        P-->>I: Resubmit verified evidence
    end

Loading

The goal stays fixed in the first lane. Three tracks progress independently and hand their evidence to the integration gate, but the beta remains locked until all required evidence arrives. If a problem appears, Track returns only the affected work instead of restarting the whole plan. Sessions may be reassigned without changing these workstream lanes.

Mermaid is the default visualization. Track uses an ASCII fallback only when Mermaid cannot render or plain text is explicitly required.

Workflow

1. Establish the goal, constraints, definition of done, and observed state.
2. Reuse existing decisions and completed artifacts before creating new work.
3. Decompose the goal into inspectable outcomes and draw their dependencies.
4. Pass each new task, session, or scope expansion through the over-engineering gate.
5. Assign bounded work with an output, acceptance condition, and stopping point.
6. Inspect evidence, update the living map, and unlock dependent work.
7. Finish when the acceptance conditions are met, or revise the same map when the goal changes.

Install

Place this repository in your Codex skills directory:

mkdir -p ~/.codex/skills
cp -R /path/to/track ~/.codex/skills/track

Then start a Codex task with a request such as:

Use $track to turn this project into a living development map and coordinate it to completion.

Repository contents

track/
├── SKILL.md               # operational instructions for Codex
├── agents/openai.yaml     # skill-list metadata
├── README.md              # public orientation and example
├── VERSION                # current public version
└── index.html             # standalone landing page

Boundaries

Track is a coordination method, not a project-management backend. It does not create idle sessions for every component, automate staffing, or expand the project to solve hypothetical future problems. It keeps deferred ideas visible without presenting them as active commitments.