菜单
Supply an LLM acceptance verdict for a dev-spec-kit `judge` check — for criteria a test can't assert (tone, copy quality, "is this error actionable", a transcript vs a rubric). Use when a @check kind=judge needs proving in harness mode (no API key needed — you are the LLM).
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Build, refresh, and read dev-spec-kit's Verified Traceability Graph — provider-agnostic. Keep it fresh with the build wrapper, read blast radius / implements-edges / drift to answer "what breaks if this changes?", and run the staleness re-prove dance. For choosing and configuring the code-graph engine itself, see the dev-spec-kit-revitify and dev-spec-kit-graphify skills.
dev-spec-kit's optional code-graph provider — the external Python graphify tool (multi-modal: indexes PDFs, images, and video alongside code). Opt-in only. Use to decide when graphify is worth its install over the bundled revitify default, how to install and configure it, and how dev-spec-kit verifies and falls back when it is absent. For the build/read loop itself, see dev-spec-kit-graph.
dev-spec-kit's default code-graph provider — the bundled revitify engine (native TypeScript, tree-sitter multi-language). Zero install, zero API key, always available. Use to understand when revitify is the right provider, how it is configured, what it can and cannot index, and how it feeds the Verified Traceability Graph. For the build/read loop itself, see dev-spec-kit-graph.
Drive spec-driven development with dev-spec-kit — route the request, write EARS specs with @check bindings, derive evidence-bound tasks, TDD, prove checks with real test runs, build the Verified Traceability Graph, record approvals, and generate graph-derived PR bodies. Use whenever the user asks to build/change code in a dev-spec-kit-initialized project (.dev-spec-kit/ present) or mentions rivet, specs, EARS, or traceability.
Score each requirement's complexity 1–10 and recommend a subtask breakdown, then expand the over-scoped ones into bound sub-criteria. Use after a spec is drafted and before implementation, to catch requirements that are too big to prove as a single unit.
Turn approved requirements into a thin high-level design and the Architecture Decision Records (ADRs) that justify the non-obvious choices. Use after requirements are clarified and before implementation, or when a change makes a load-bearing architectural decision.
| name | dev-spec-kit-judge |
| description | Supply an LLM acceptance verdict for a dev-spec-kit `judge` check — for criteria a test can't assert (tone, copy quality, "is this error actionable", a transcript vs a rubric). Use when a @check kind=judge needs proving in harness mode (no API key needed — you are the LLM). |
A judge check binds a criterion no executed test can assert — UX-copy tone, error-message
actionability, "does this ADR explain the why", a transcript meeting a rubric. In harness mode
(the default), YOU are the judge: read the rubric + the evidence, decide strictly, record the verdict.
No API key, no network — you're already the model. (api mode calls Anthropic headlessly for CI.)
file::name, the file is the
evidence; otherwise judge what the ref points at.dev-spec-kit check run <taskId> "<ref>" --verdict pass|fail --reason "<one falsifiable line>".--reason — what in the artifact does or doesn't satisfy the
criterion. "looks good" is not a reason.REQUIREMENT_/NFR_) green to dodge writing a real test;
judge is for the genuinely-unmeasurable. The gate blocks it unless verify.judge.allowForObligations.⚖️ judged, never an executed green.
Prefer an executable check whenever the criterion can be mechanically asserted.