원클릭으로
book-refine
Incorporate feedback into book review policies, standards, and infrastructure
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Incorporate feedback into book review policies, standards, and infrastructure
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
Explicitly invoked only. Authoritative reference for the Ragu formal-verification project — Clean DSL semantics (Circuit monad, Expression, Operation, ProvableType, FormalCircuit) and Ragu-side concepts (extraction driver, formal instances, input/output serialization, assumptions, CI). Routes to the live in-tree FV docs; the docs are the spec, not opinion. Pair with /fv-review for opinion / pattern-matching guidance. Invoke when the user types `/fv-docs` or asks for the canonical definition of a Clean / Ragu-FV concept.
Explicitly invoked only. Lessons learned from porting Ragu Rust circuits to Clean Lean formal verification reimpls — when not to formalize an empty circuit, picking FormalCircuit vs FormalAssertion vs GeneralFormalCircuit, mirroring Rust delegation, length polymorphism, naming conventions, and reading proof failures (stuck goals, hypotheses you're forced to add) as under-constraint / undocumented-precondition bug signals. Distilled from PR review feedback (mitschabaude et al.) and refined over time. Do NOT auto-trigger on general formal verification, Lean, or Ragu questions; only invoke when the user explicitly types `/fv-review` or asks by name.
Explicitly invoked only. Loads context for the Verified-zkEVM `clean` codebase — an embedded Lean DSL for formally verified zk circuits (AIR/PLONK/R1CS) with Circuit monad, ProvableType/ProvableStruct, FormalCircuit (soundness/completeness), and gadget composition. Do NOT auto-trigger on general Lean, ZK, or circuit questions; only invoke when the user explicitly types `/clean-circuits` or asks by name to use this skill.
Review any part of the book using parallel specialist agents
Review code using parallel specialist agents
Audit documentation surface placement (book vs. rustdoc) and refine the policy
| name | book-refine |
| description | Incorporate feedback into book review policies, standards, and infrastructure |
| user-invocable | true |
Incorporate the user's feedback into the book review system. The feedback is in
$ARGUMENTS and/or in the preceding conversation context. This skill modifies
the book review infrastructure itself to make future reviews better.
You MUST use plan mode (EnterPlanMode) for this skill. Do all analysis and auditing first, present a plan for the user's approval, and only then execute.
Read ALL of these files before doing anything else:
.claude/review-shared/writing.md — Shared writing rules (used by book-review and code-review).claude/review-shared/math.md — Shared math notation rules (used by book-review and code-review).claude/book-review/standards.md — Master standards shared across all book reviewers.claude/book-review/*.md — Per-focus review policies.claude/skills/book-review/SKILL.md — Book review orchestration.claude/skills/book-refine/SKILL.md — This fileThe user's feedback could be anything. Common forms:
Don't just record the specific instance — extract the underlying principle. Examples:
| User says | Generalized rule |
|---|---|
| "I'd write 'the prover constructs' not 'a prover constructs'" | Use definite articles for named protocol roles (the prover, the verifier) |
| "This section should have come before the definition" | Motivating examples and intuition should precede formal definitions |
| "The math reviewer missed that alpha was reused" | Check that notation is consistent with prior chapters, not just within the current one |
| "Stop complaining about sentence fragments in bullet points" | Sentence fragments are acceptable in bullet points when the meaning is clear from the list context |
| "Don't rewrap it like that, do it like this" | A formatting correction — the user is showing a preferred rewrapping pattern. Add a before/after example to .claude/book-review/formatting-examples.md under the appropriate category, and refine the rule in formatting.md if the correction reveals a gap. |
| "You should check that code examples actually match the current API" | (New review dimension: code accuracy. Create .claude/book-review/code.md) |
If you're not sure how to generalize a specific correction into a principle, ask the user. Don't guess — a wrong generalization is worse than none.
Before deciding where to put anything, review the full set of policies and standards as a whole. Look for:
grammar.md that's really about document structure belongs in
structure.md. Flag and propose moves.Include any findings from this audit in your plan, even if they go beyond the user's immediate feedback. The goal is to keep the policy system coherent as it grows.
Enter plan mode (EnterPlanMode). Your plan should include:
.claude/review-shared/writing.md.claude/review-shared/math.md.claude/book-review/standards.md.claude/book-review/*.md.claude/book-review/{name}.md.claude/skills/book-*/SKILL.mdWhen in doubt between general and specific placement, prefer the more specific
location. A rule can always be promoted to .claude/book-review/standards.md later if it
turns out to be universal.
Wait for user approval before proceeding.
After the user approves the plan:
.claude/book-review/standards.md, add to the appropriate existing
section or create a new section if none fits.Tell the user: