plan
Codex compatibility note:
- Invoke
$skill-name: load the registered.agents/skills/<name>/SKILL.mdand execute it with Codex tools; no separateSkilltool or Claude mode switch is needed. When a workflow step or another skill names a skill, read that file in full yourself (every line, with no line limit; in parts only when output is cut off): a skill whose implicit invocation is off is never loaded for you. Author.claude/**; mirrors are generated.- Track steps before execution and keep progress/evidence current. Explicit skill requests execute their protocol. Use the native question tool for required questions.
- Map foreign tool names to available host capabilities. A name mismatch needs no approval; an actually unavailable required capability stops with its step/evidence and a user question before adaptation.
- Subagent authorization: when a skill is user-invoked or AI-detected and its protocol requires subagents, that skill activation authorizes use of the required
spawn_agentsubagent(s) for that task.- Preserve protocol order unless the user approves a deviation. Workflows use
$start-workflow: gates stay fixed; other steps may flex with logged reasons and evidence. [BLOCKING] Mode routing — detect FIRST. Select explicit--mode=review|validate|execute|ci|cro; otherwise create a plan. Read the selected reference in full before acting (see Mode Dispatch). Every mode works directly without a workflow./plan-review,/plan-validateand/plan-executedo not resolve.
Work tracking: Keep the project's tracked work current in the same change, as its root instructions (
CLAUDE.mdorAGENTS.md) say. When the project uses the portable tracker (taskTracking.modeisobserveorlinked), read the linked work integration guide at capture, start, saved-work, verification, handoff and close-out checkpoints. Inspect exact selected owners and declared spec/task/subtask/plan concerns before planning, incorporate applicable unresolved concerns into task-derived quality gates, and reread after an actual artifact save. Execute mode retains actual linked producer/run/occurrence and the primary saving owner's one checkpoint. Continue untracked; report partial/unavailable scope and saved/pending secondary results. Plan prose and passing checks never authorize readiness, acceptance or publication.
Quick Summary
Review modes: For review/audit work, standalone defaults to
--review-only(--report-onlyalias);--fix-loopenables review → validate → authorized fix → fresh re-review, default cap 3.--fix-loop --loop-owner=callerreturns a read-only pass to the caller that owns fixes/re-review. Create triage, review, validation and re-review todo tasks first; apply the carriedreview-policycontract for LOW deferral and user-approved bounded extensions. Non-review modes and terminal findings validation keep their own dispatch.
Goal: Produce a concise, evidence-backed implementation plan that fixes direction and proof while leaving code-level discovery and mechanics to the executing agent.
Summary: Select the mode first: create a plan, review once, interview for validation, execute, or add CI/CRO intake. For creation: resolve evidence → settle decisions → derive gates → write → self-check → $why-review self-review (one round; up to three for a high-risk full-profile plan) → validate with the user (standalone, or in a workflow that lists no validate step) → hand back (a standalone run ends with the one Standalone Close question).