Creates a new Codex skill for this project. Guides the user through designing the skill's purpose, trigger conditions, structure, and content. Use when the user wants to add a new slash command or automated workflow.
Acts on review panel findings by dispatching concurrent GPT 5.5 workers with medium reasoning effort for small-blast-radius items (one per file), then a GPT 5.5 worker with high reasoning effort for remaining issues with knock-on effects. All agents read relevant context files. Use after /review-panel produces findings.
Orchestrates execution of a plan by spawning agents to work on tasks according to the plan's sequencing instructions. Reads a plan from context/plans/ready/, moves it to in-progress, and coordinates task execution across phases. Use when a reviewed plan is ready for implementation.
Multi-agent review of a draft spec in `context/plans/drafts/`. Spawns two parallel reviewers — a broad reviewer and a codebase-anchor reviewer that fact-checks every named identifier against source. Auto-applies mechanical fixes via a GPT 5.5 worker with medium reasoning effort unless --no-auto-apply is set. Recommends apply / re-review / promote. Use after a draft session, or when a human wants to validate before promoting to ready/.
Runs a multi-agent review panel: parallel code reviewers plus a dedicated comment drift checker. Aggregates findings with deduplication and severity merging. Runs in a forked context so the active agent's context window stays clean and reviewers have no bias from prior work. Use mid-session after implementing a feature, or before opening a pull request.
Acts on review panel findings by dispatching concurrent Sonnet agents for small-blast-radius items (one per file), then an Opus agent for remaining issues with knock-on effects. All agents read relevant context files. Use after /review-panel produces findings.
Drafts feature or epic specs for PostRetro. A session may produce zero, one, or several plans depending on scope. Use when starting new planning work. Does not promote to ready/ — that is a separate step after review.
Multi-agent review of a draft spec in `context/plans/drafts/`. Spawns two parallel reviewers — a broad reviewer and a codebase-anchor reviewer that fact-checks every named identifier against source. Auto-applies mechanical fixes via a Sonnet sub-agent unless --no-auto-apply is set. Recommends apply / re-review / promote. Use after a draft session, or when a human wants to validate before promoting to ready/.