메뉴
Parallel multi-agent review for Laboratory changes. Prioritizes correctness, scientific validity, provenance, tests, and harness fit. Trigger: /code-review, /review, /critique.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Assess and improve codebase readiness for AI coding agents. Dispatches parallel subagents to evaluate style, testing, docs, architecture, CI, observability, security, and dev environment. Produces a scored report with prioritized remediation. Then executes the highest-impact fixes. Use when: "agent readiness", "is this codebase agent-ready", "readiness report", "make this codebase agent-friendly", "agent-ready assessment", "readiness audit", "prepare for agents". Trigger: /agent-readiness, /readiness.
Dialectical premise-and-alternatives audit for a plan, spec, or context packet. Four moves: premise challenge (is this the right problem?), mandatory structurally-distinct alternatives, cross-model outside voice, user-ratified convergence. Operationalizes the AGENTS.md "Diverge Before You Converge" doctrine at the plan-review stage. Use when: about to commit to a plan/spec/design, reviewing a ticket before shape, when "is this the right problem" would be useful, or any time a proposal smells like a symptom fix instead of a root-cause fix. Trigger: /ceo-review, /challenge, /premise-check.
Laboratory CI gate audit and verification workflow. Use when checking whether experiment, harness, analysis, docs, or Spellbook harness changes are safe to ship. Trigger: /ci, /gates.
Take one Laboratory backlog item to merge-ready. Composes /shape, /implement, /code-review, /ci, /refactor, and /qa as needed. Stops before merge. Trigger: /deliver.
Capture and share what changed — every app has a demo path; the format depends on the change and the audience. The spectrum runs from a release-notes blurb ("no demo needed, here's what moved") through a curl + JSON paste, a terminal recording, a screenshot, a GIF, up to a polished narrated launch video. Pick the path that fits the change shape (UI / API / CLI / library / agent tool / internal-only), the audience (PR reviewer / team Slack / external customer), and the time budget. Also scaffolds a project-local demo skill for repos with recurring polished-capture needs. Use when: "make a demo", "generate demo", "record walkthrough", "launch video", "PR evidence", "upload screenshots", "demo artifacts", "make a video", "demo this feature", "create a walkthrough", "show what changed", "paste the output", "release notes blurb", "scaffold demo", "generate demo skill". Trigger: /demo.
Investigate, audit, triage, and fix. Systematic debugging, incident lifecycle, domain auditing, and issue logging. Four-phase protocol: root cause → pattern analysis → hypothesis test → fix. Use for: any bug, test failure, production incident, error spikes, audit, triage, postmortem, "diagnose", "why is this broken", "debug this", "production down", "is production ok", "audit stripe", "log issues". Trigger: /diagnose.
| name | code-review |
| description | Parallel multi-agent review for Laboratory changes. Prioritizes correctness, scientific validity, provenance, tests, and harness fit. Trigger: /code-review, /review, /critique. |
| argument-hint | [branch|diff|files] |
Review the diff as a Laboratory artifact. Findings lead; summaries are secondary. Block on bugs, invalid scientific claims, missing tests, broken simulate/live separation, missing trace/provenance, and runtime paths that were not exercised.
Default scope is git diff <base>...HEAD, where base is master unless the
user names another base. Include uncommitted changes when the user asks for a
working-tree review.
Classify touched surfaces:
Use fresh reviewers for substantive reviews:
Use Thinktank or cross-harness review when the diff is broad, high-risk, or touches experiment conclusions. Provider output is evidence, not authority; the lead synthesizes and verifies.
--simulate behavior and require explicit
--live for API calls..agents roots and bridge links.ship: no blocking findings and required executable paths were exercised.conditional: only clearly bounded follow-up remains and the user can accept
the risk.dont-ship: any correctness, scientific-validity, provenance, safety, or
unverified-runtime blocker remains.Return findings first, ordered by severity with file/line references. Then list open questions, commands or artifacts reviewed, and the verdict.