Planton361

life-os-codex-task-writer

Turn an approved Life OS GitHub Issue or bounded question into a concise Codex work contract for research, decision/design, experiment, delivery, evaluation or review without duplicating repository rules.

Planton361 0 Updated 1mo ago
GitHub

Install

npx skillscat add planton361/life-os/life-os-codex-task-writer

Install via the SkillsCat registry.

SKILL.md

Life OS Codex Task Writer

Use this skill to produce a directly executable Codex prompt. The approved GitHub Issue is the concrete delegated contract; repository sources provide product, architecture and safety context.

Sources

Always read:

  • AGENTS.md;
  • the approved GitHub Issue.

Then load only what the Issue needs:

  • referenced ROADMAP.md block and Capability Registry entries for product work;
  • PRODUCT.md / DESIGN.md for product/UI scope;
  • ARCHITECTURE.md, DATA_MODEL.md, SECURITY.md, ACCESSIBILITY.md when those boundaries are affected;
  • relevant project-local .agents/skills/*/SKILL.md;
  • relevant code, tests, history and evidence.

Historical workflow documents are evidence only and are not required context for normal tasks.

Work Types

Use exactly the work type appropriate to the Issue:

  • DISCOVER: clarify need, value, non-goals and uncertainty;
  • RESEARCH: answer a bounded question from sources/code;
  • DECIDE/DESIGN: compare viable options and persist an accepted decision when authorized;
  • EXPERIMENT: run a bounded, reproducible test with explicit budget/stop conditions;
  • DELIVER: implement one coherent observable result;
  • EVALUATE: interpret actual evidence and limits;
  • REVIEW: read-only focused review of a concrete revision.

Do not force every task through every work type.

Hard Rules

  • Prefer one coherent outcome over layer-only or docs-only microtasks.
  • Do not expand the approved Issue scope.
  • Do not delete/move files, touch protected paths, use remote DB/deployment/providers or broaden permissions without explicit authorization.
  • Preserve V5 and the established feature-local architecture.
  • One writer per branch; normal Delivery uses codex/<issue>-<slug>.
  • Research/Evaluation with no repository changes does not need an artificial branch/PR.
  • Same Issue repairs stay in the same session/branch/PR.
  • Include only the smallest useful skill set and validation.

Skill Mapping

UI / cross-domain Delivery:
- life-os-vertical-slice
- life-os-design-taste
- life-os-browser-proof
- life-os-surface-acceptance when a core surface is touched/closed
- life-os-completion-gate

Backend/Data Delivery:
- life-os-backend-action-slice
- life-os-browser-proof if user behavior is affected
- life-os-completion-gate

Review:
- life-os-completion-gate
- life-os-design-taste only for UI review

Research / Decision / Experiment / Evaluation:
- only a project skill that materially helps; otherwise no artificial skill requirement

Docs / Workflow:
- life-os-completion-gate

Prompt Shape

Issue:
Work Type:
Goal:
Roadmap Context:
Use Skills:
Read:
Scope / Non-goals:
Authorization:
Acceptance / Evidence:
Validation:
Return:

The return is concise:

Status: READY_FOR_REVIEW | PARTIAL | BLOCKED | DECISION_REQUIRED
Result:
References:
Revision:
Checks/Evidence:
Open:
Next action:

Do not request or reproduce a long session history.