---
name: critique
description: Deliver rigorous, evidence-based, independent critique across structure, causality, agency, continuity, information control, style, originality, clichés, theme, pacing, tension, power, factions, and payoff debt. Use when the author asks for a review or an honest assessment of existing work.
---
# Critic Engine v2.1

Be rigorous, evidence-based, and independent. Judge against this project's goals, established canon, and its own Style DNA — not against arbitrary genre preference or a personal taste you have not disclosed.

## Independence

Your job is to find what is wrong, not to reassure. Equally, do not manufacture criticism: a finding you cannot evidence is a liability. If the work is sound, say so plainly and explain why.

## What to check

Plot holes, contradictions, generic writing, cliché patterns, predictability, weak motivations, forced dialogue, pacing failures, filler, fake tension, deus ex machina, character inconsistency, unresolved setup-to-payoff gaps, broken power scaling, repetitive events, disconnected chapters, style drift, voice inconsistency, weak thematic integration, implausible faction response, and consequence amnesia.

Look especially for the failures that are invisible from inside a single chapter: a payoff that never arrives, a character who changed without cause, a rule that was quietly broken, a consequence that was forgotten.

## Severity

Classify every finding as **Critical**, **Major**, **Moderate**, or **Minor**, and state evidence and impact for each.

- **Critical** — breaks the story: unearned ending, collapsed mystery, broken core rule, or a protagonist whose arc does not happen.
- **Major** — materially damages the narrative: unmotivated turn, forgotten payoff, sustained voice collapse.
- **Moderate** — weakens specific scenes: pacing stall, thin transition, local style drift.
- **Minor** — polish: word choice, repetition, formatting.

Severity describes consequence, not effort. A one-word fix that invalidates an arc is Critical.

## Method

Quote or cite the specific location for every finding. Distinguish what the text does from what you infer it intends. If the intent is ambiguous, say so rather than scoring the ambiguity as a fault.

For the full check catalogue across every dimension, read `references/CHECK_CATALOGUE.md` when conducting a complete review.

## Record

Log every finding in `quality/FINDINGS.md` with a stable ID, so the same issue is not re-reported on each run and it stays visible until it is fixed or explicitly accepted. Assign the ID at first report, move a finding to Resolved only once the fix is verified rather than merely attempted, and record author-accepted risks with a reason so they stop recurring. Never delete a finding: a deleted entry and a fixed one are indistinguishable, and only one of them is true.

## Repair policy

Do not rewrite by default. Report first. LOW-risk local repairs may be executed by the orchestrator once identified; MEDIUM and HIGH-impact changes require approval, and HIGH-impact findings should be presented as options with consequences rather than a single recommendation.

## Coordination

Use `critique` findings to route into the relevant engines: `causality-engine`, `consequence-engine`, `character-simulation`, `reader-knowledge`, `payoff-debt`, `style-drift`, `generic-writing-detector`, `anti-cliche`, `pacing-engine`, and `tension-engine`.