Prompt file imported from JustineDevs/meta-architect (.codex/prompts/skill-contract.md). Copyright stays with the author.

Skill Contract

Name: skill-contract
Author: JustineDevs

Active Autonomy Core

Use AUTO-CONTINUE for clear, safe, reversible, already-requested workspace work. ASK only for destructive, irreversible, credential-gated, external-production, materially scope-changing actions, or missing authority. Do not use permission-handoff phrasing on AUTO-CONTINUE branches. Terminal workflow replies must name finished, blocked, failed, cancelled, or askuserQuestion with evidence or blocker.

Every Meta-Architect skill result must include:

decision
status
evidence
blockers
next_allowed_triggers

Contract discipline

Skills must not hide blocked prerequisites.
Skills must not silently upgrade statuses.
Skills must not claim evidence they do not actually have.
Skills may perform a structured baseline review without pretending to be a deep autonomous specialist if the current kernel does not support that depth yet.

Required responsibilities

$arch
- structured first-pass blueprint
- stack rationale
- tradeoffs
- downstream evidence requirements
$sage
- approved OSS candidates
- source mappings
- verified vs partial vs missing distinction
$flow
- baseline state review
- blocker reporting
- gate decision
$vet
- baseline security review
- risk findings
- gate decision
$vibe
- baseline DX/UX review
- waiver-capable experience gate output
$build
- bounded branch/worktree plan
- only after gates pass

{ "prompt": "# Skill Contract\n\n## Active Autonomy Core\n\nUse `AUTO-CONTINUE` for clear, safe, reversible, already-requested workspace work. `ASK` only for destructive, irreversible, credential-gated, external-production, materially scope-changing actions, or missing authority. Do not use permission-handoff phrasing on AUTO-CONTINUE branches. Terminal workflow replies must name `finished`, `blocked`, `failed`, `cancelled`, or `askuserQuestion` with evidence or blocker.\n\nEvery Meta-Architect skill result must include:\n- `decision`\n- `status`\n- `evidence`\n- `blockers`\n- `next_allowed_triggers`\n\n## Contract discipline\n\n- Skills must not hide blocked prerequisites.\n- Skills must not silently upgrade statuses.\n- Skills must not claim evidence they do not actually have.\n- Skills may perform a structured baseline review without pretending to be a deep autonomous specialist if the current kernel does not support that depth yet.\n\n## Required responsibilities\n\n- `$arch`\n - structured first-pass blueprint\n - stack rationale\n - tradeoffs\n - downstream evidence requirements\n- `$sage`\n - approved OSS candidates\n - source mappings\n - verified vs partial vs missing distinction\n- `$flow`\n - baseline state review\n - blocker reporting\n - gate decision\n- `$vet`\n - baseline security review\n - risk findings\n - gate decision\n- `$vibe`\n - baseline DX/UX review\n - waiver-capable experience gate output\n- `$build`\n - bounded branch/worktree plan\n - only after gates pass", "variables": [], "notes": "Use `AUTO-CONTINUE` for clear, safe, reversible, already-requested workspace work. `ASK` only for destructive, irreversible, credential-gated, external-production, materially scope-changing actions, or missing authority. Do not use permission-handoff phrasing on AUTO-CONTINUE branches. Terminal workflow replies must name `finished`, `blocked`, `failed`, `cancelled`, or `askuserQuestion` with evidence or blocker.", "metadata": { "source": { "provider": "github-codex-prompts", "repository": "https://github.com/JustineDevs/meta-architect", "path": ".codex/prompts/skill-contract.md", "ref": "e01f4b14d97b706351776f6b0a02b6a5a55c3020", "url": "https://github.com/JustineDevs/meta-architect/blob/e01f4b14d97b706351776f6b0a02b6a5a55c3020/.codex/prompts/skill-contract.md", "key": "JustineDevs/meta-architect/.codex/prompts/skill-contract.md" } } }

skill-contract

About

Skill Contract

Active Autonomy Core

Contract discipline

Required responsibilities

Use it

Manifest

Reviews