# Critique Grimoire > Adversarial structural review of `output//grimoire.md` — the world grimoire. Audits the foundational worldbuilding document BEFORE `book-setup` or `plan-book` run, so that the per-book plan never inherits a flawed foundation. Use this when the user says "audita el grimorio" / "critique the grimoire" / "is the grimoire ready?" Invoke after writing or revising the grimoire, and before running `book-setup` for a new book in the series. - Skill: `abetatos/critique-grimoire` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add abetatos/critique-grimoire` - Raw SKILL.md: https://api.skillmd.com/api/skills/abetatos/critique-grimoire/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: abetatos (https://skillmd.com/u/abetatos) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/abetatos/critique-grimoire --- # critique-grimoire You are running the **critique-grimoire** skill. The grimoire is the foundational worldbuilding document for the whole series. Everything downstream (`setup.md`, `canon/*`, `plan/*`, chapters) ultimately quotes from it. If the grimoire has a structural gap, every per-book plan inherits it — and the writer will discover it mid-prose. This skill exists so that **the grimoire is bulletproof before any per-book work starts**. The author has explicitly chosen interactive critical planning over mid-writing pauses; this is where the criticism lives. ## When to invoke - The author has finished a draft of `output//grimoire.md` (or revised it after critique). - Before running `book-setup` for Book I of a series, or for any new book that introduces new worldbuilding. - The author says "audita el grimorio" / "critique the grimoire" / "está listo el grimorio para empezar?" ## Hard rules - **Adversarial bias is total.** A grimoire that passes with zero findings is suspicious; re-read. The cost of finding a problem here is one conversation. The cost of finding it during Libro II is a rewrite. - **Be specific.** "Magic is weak" is useless. "§4 declares two costs but both are physical — there is no emotional/social cost, so the magic has no political traction" is useful. - **Quote sources.** Every finding cites a line or a missing section. - **Tier ruthlessly:** - **MUST fix** — gaps that break everything downstream: missing source/cost/limits in magic, no antagonist with thesis, climax declared as passive instead of active, no clock / "why now", unresolved gating decisions, system-arc-shape (scaled vs inverted) not declared. - **SHOULD fix** — gaps that weaken the book: < 5 named places, < 3 historical events, subplots without ≥ 3 contact points with main, missing distinct theme on a subplot, characters without full want/need/lie/wound, missing thematic question on magic. - **CONSIDER** — taste-level (e.g., a faction declared but no clear conflict with the others; a place with no sensory anchor). - **Do not rewrite the grimoire.** Quote, name, point. Concrete direction yes; substitute text no. The author writes the grimoire. - **Cross-check, don't assume.** When a deep check looks ambiguous (e.g., "subplot has 2 contact points — is the 3rd implied?"), surface it as SHOULD with the ambiguity explicit, not as PASS. ## Steps ### 0. Dispatch the analysis to a fresh subagent (main thread only) If you can spawn subagents (you have the Agent tool), **do not audit in this conversation**. Dispatch steps 1-4 to the `book-critic` subagent (Agent tool, `subagent_type: book-critic`) with the prompt `grimoire --series-slug `. It audits in a **fresh context** — no leak from this session, no `/clear` needed to re-audit in clean. It writes `notes/critique-grimoire.md` and returns the verdict + counts + load-bearing findings. Read that file, then go **straight to step 5** (report) and step 6 (compose with critique-plan) — those stay here, in the main thread. > When the **book-critic subagent itself** runs this skill it has no Agent tool, > so it cannot dispatch — it simply executes steps 1-4 directly and stops before > step 5. (That is how recursion is structurally prevented.) ### 1. Run the deterministic audit ```bash python3 .claude/skills/critique-grimoire/scripts/audit_grimoire.py \ --series-slug \ --output output//notes/_audit-grimoire.md ``` The audit checks: - Section presence and "filled" state (header without content fails). - Magic depth: source / mechanic / costs ≥ 2 / hard limits ≥ 3 / thematic question. - System arc shape: three escalation tiers OR inverted-system explicitly declared. - Characters: ≥ 2 principals with full arc (want+need+lie+wound), antagonist with stated thesis. - Subplots: count, contact points, distinct theme. - Historical events: ≥ 3 enumerated. - Geography: ≥ 5 named places. - Structure: each book block has a Motor and an active-decision climax. - Open decisions: any gating decision still unresolved. - **Series-scoped spine floors** (scale with book count N, read from §12 / §14 / §14b): the grimoire only holds the *trans-series* spine, so the audit expects **≥ max(6, 3N) loaded guns**, **≥ max(5, 2N) master mysteries**, **≥ max(2, N) subplots**, **≥ max(1, N−1) decoys**, and **≥ max(1, N−1) trans-book threads** (guns that sow early / pay late) plus the same floor of bridging mysteries (intro early / confirm late). Shortfalls are SHOULD-fix — the qualitative pass below decides if a thin spine is load-bearing enough to block. - **Decoys**: deliberate misdirection (a false system, a hollow saviour, an institution fronting for the real parasite) counted across the gun / mystery tables and the factions / characters / two-layer prose. A trilogy that lies to the reader should name them on purpose. If `grimoire.md` does not exist, the audit refuses and tells the user to start from `references/grimoire-template.md`. ### 2. Read the grimoire in full Read `output//grimoire.md` and the audit report. Also read these craft references for what "good" looks like: - `references/magic-design-checklist.md` - `references/seed-craft.md` - `references/fantasy-beats.md` - `references/hard-magic-laws.md` ### 3. Run the qualitative pass The audit catches mechanical gaps. You catch the substance. Go through each of these categories — every check produces either a finding or an explicit "clean". #### 3a. Premise integrity - Is the one-sentence idea actually unique, or is it a recombination of well-worn elements? Push back if generic. - Do the **two layers** declared in §2 add up to a story with retroactive payoff, or do they restate the surface in code words? - Is the tonal register declared and coherent (gothic / epic / romantasy / progression / etc.)? #### 3b. Magic system as engine - Source, costs, limits — are they present **and specific** (not "energy", not "physical exhaustion")? - Is the magic system politically integrated — does it stratify the world economically and morally? - Is the thematic question forced by the magic, or is it a separate layer pasted on top? - If the system is **scaled**: are the three tiers visible at the right times in the structure (common in act 1, skilled in act 2, apex at climax)? - If the system is **inverted**: is the erosion mechanic concrete (what does the protagonist lose, on what schedule)? Is the climax the end of their resources, not the peak of their power? #### 3c. Cast distinctness — *and enough cast to survive the whole series* A full-arc count alone is a trap: the audit will happily report "2 full arcs" while one is a `[PROPUESTA]` placeholder and the other is consumed in Book I. Counting bullets is not the same as having a cast that carries N books. So check, in this order: - **Fixed vs proposed.** For every arc-bearing figure (in §9 **and** in the subplots section — the de-facto deuteragonist often lives there), is the arc **`[FIJO]`** or still **`[PROPUESTA]`/draft**? An arc that carries the back half but is tagged proposed is **not a commitment** — flag every one by name (SHOULD). The trilogy cannot be audited against an un-fixed arc; the per-book plan inherits the gap. - **Survival into the back half.** Which fixed arcs are *spent* in Book I (the figure dies / is lost / is consumed / is grisado)? Subtract them. For an N-book series you want **≥ max(2, N−1) fixed arcs that survive into the later books.** One living designed arc entering Book II is the single most likely mid-Book-II rewrite — call it out explicitly as the trilogy's thinnest spine (SHOULD, escalate toward REVISE if it's the only finding-cluster). - **Devices vs people.** Is any back-half emotional pivot (the love interest, the bond whose loss is an all-is-lost) a one-line *function* with no wound/want/lie of its own? A pivot that carries a climax must be a person, not a device — SHOULD give it an arc. - **Distinctness.** Do their wounds/wants/lies differ? If two figures carry the same lie or decision-shape, the book has one arc, not two. - **Antagonist thesis.** A worldview that, in different light, would be correct? If stock evil, that's MUST fix. - **A second POV/deuteragonist** beyond protagonist and antagonist? Epic fantasy without one feels claustrophobic — and a *committed* one (not a mere "se permiten otros POV" permission punted to `plan-book`). #### 3d. Subplots earning their place - ≥ 1 declared (≥ 2 if total target > 250k). - Each one with ≥ 3 contact points with the main plot? Quote them. - Distinct theme from the main? If the subplot argues the same thesis as the main, it's decoration. - Resolution before or during the climax of the main? Subplots that resolve after drain energy. #### 3e. World economy - ≥ 5 named places, each with a sensory anchor and a function. - Each named faction has a dramatic function (someone uses them, fears them, fights them). Decorative castes / orders / institutions are SHOULD-cut. - Geography supports travel logistics — distances declared or derivable? #### 3f. Historical weight - ≥ 3 past events that reverberate in the present. - Each has a stated wound it left. ("Mil años de criba" needs a concrete wound: who still bears it?) - Are mythical accounts paired with **the real truth** (writer-only), so the reveal mechanism is honest? #### 3g. Trilogy / structure rigor - Each book block declared with a distinct **motor**. - No diminishing-returns reveals across books (Book II's twist must be different in kind from Book I's, not a louder echo). - Each climax uses an **active-decision verb** (decide / elige / rechaza / acepta / niega / entrega). Passive verbs ("abre", "cae", "sucede") at the climax are MUST fix — the protagonist must act, not be acted upon, at the moment the theme is paid. - All-is-lost moments named (or at least the dramatic shape sketched) for each book. #### 3h. The clock — "why now?" - Is the clock explicit? The series happens **in this moment** because something causal forces it. - Does the clock connect to the protagonist's existence (best case — protagonist and crisis are the same event) or is it independent? - Without a clock, the world feels timeless and the urgency is vibes-only. #### 3i. Loaded guns (§14) and master mysteries (§14b) - Distinct worldbuilding elements declared — rituals (the Lectura), pseudo-systems (the Blanco Falso), implicit physical laws (the complementario) — are they marked as siembras obligatorias with a payoff plan? - Are there any **declared elements with no payoff plan**? Those are decoration. SHOULD cut or SHOULD plant. - **§14 completeness (the audit checks this):** each loaded gun has a sow-book AND a payoff-book, and — by the grimoire's own rule — appears at least once in Book I. A row missing a book, or that never lands in Book I, is `MUST fix`. An empty §14 table is `MUST fix`: the per-book plan would have no concrete obligations to realize. - **§14b master mysteries:** the grimoire must enumerate, as a table, the hidden truths the trilogy reveals, each with a real-truth, an intro-book and a confirm-book. These are what `critique-plan` later holds each book's `shadow.md` to (every mystery introduced in a book needs a `SHADOW-TRUTH` carrying it). An empty/missing §14b table or rows missing intro-book are `MUST fix` — without them the plan's shadow coverage cannot be checked. #### 3j. Open decisions discipline - Every open decision marked with **Gating: sí / no**? - All gating decisions resolved? If not, MUST fix. - Non-gating decisions named so they don't get forgotten? #### 3k. Cross-book inheritance (trilogy only) — the attrition ledger A trilogy that kills its tutor in Book I and grises the love interest in Book II is **spending cast**; if that spend is never costed, the back half quietly runs out of people. The grimoire (or `series-state.md`) should carry an explicit ledger — **carries / dies / enters per book**: - What does the reader **carry** from Book I into Book II that they didn't have at the start? Is it stated? - What **dies** between books (characters, factions, places)? Is it named — and is the resulting gap *refilled* (who enters to carry the arc the dead figure was carrying)? - What new figures **enter** in Books II/III, and are their arcs fixed by the time they have to bear weight? - What promises does Book I make that Book III must pay? If this ledger is absent and the grimoire is killing/consuming arc-bearing figures across books (check against 3c), that is a **SHOULD fix**, not a nicety: it's the bookkeeping that proves the cast survives to the end. Note when `series-state.md` is still the empty template. ### 4. Write the critique Write to: ``` output//notes/critique-grimoire.md ``` This audit runs fresh every time: **do not read the previous `notes/critique-grimoire.md`** before forming your findings, and **overwrite** it — never merge. The only audit you trust is the one you generate this run. Structure (write the findings; leave the verdict to the script): ```markdown # Grimoire critique — **Verdict:** **Summary:** one-paragraph diagnosis. ## MUST fix - **[issue-type]** — concrete finding with quoted text or missing section pointer → suggested direction. - ... ## SHOULD fix - ... ## CONSIDER - ... ## What works - 4-6 brief notes on what the grimoire does well. The author needs the signal too, especially since this skill is harsh. ## Cross-references to plan - If a per-book plan already exists (`book-NN/`), note any inconsistencies between grimoire and book setup. ``` Tag each MUST with a bracketed issue-type. Use these **REJECT-tier** slugs for structural breaks needing the foundation rebuilt: `[magic-missing-pillar]` (source/cost/limits), `[no-antagonist-thesis]`, `[climax-passive]`, `[no-clock]`, `[system-arc-undeclared]`, `[empty-14-table]`. Any other MUST (a §14/§14b row to complete, a gating decision to resolve, a stale cross-ref) is REVISE-tier. **The verdict is computed, not judged.** After writing the findings, run: ```bash python3 scripts/compute_verdict.py \ --critique-file output//notes/critique-grimoire.md --target grimoire ``` Write the printed verdict into `**Verdict:**`. Thresholds it applies: **PASS** = zero MUST and ≤6 SHOULD; **REJECT** = any REJECT-tier MUST; **REVISE** otherwise. When in doubt between REVISE- and REJECT-tier, pick the REVISE-tier tag. ### 5. Report to user Print: - Verdict and counts by tier. - The 3-5 most load-bearing findings (one line each, citing source). - Path to the critique file. - Suggested next step: - **REJECT** → revise grimoire, then re-run `critique-grimoire`. - **REVISE** → walk through the MUST/SHOULD findings interactively; author updates grimoire; re-run. - **PASS** → `book-setup` for Book I (or for the next book in series), with grimoire as fixed reference. ### 6. Compose with `critique-plan` If a per-book plan already exists for Book N (`output//book-NN/`), `critique-plan` should be re-run AFTER `critique-grimoire` PASS. The per-book audit can then assume the grimoire is sound and focus on the book-level plan against the grimoire's contract. ## What this skill does NOT do - Does not modify `grimoire.md`. The author writes the grimoire. - Does not generate prose, plan files, or canon entries. - Does not invent worldbuilding to fill gaps. It surfaces gaps and proposes directions; the author decides. - Does not auto-parse the grimoire into `setup.md` (no `ingest-grimoire` — intentional design decision). ## Files this skill writes - `output//notes/_audit-grimoire.md` — deterministic audit. - `output//notes/critique-grimoire.md` — qualitative critique.