review

Review

Two-stage review pipeline for narrative chapters:

1. Comment — evaluates the chapter through reader personas from <brief>/personas/, scores engagement/clarity/immersion/satisfaction on /20, and produces a structured 3-section output routing to doctor or rewrite.
2. Doctor — applies technical corrections (French grammar, typography, terminology, formatting, content enrichments) guided by the output-style; terminology reference comes from <brief>/summary.md.

Run comment first to identify what needs fixing, then doctor to apply patchable corrections. If ≥2 personas are capped at ≤11/20 by structural issues, route to write --feedback instead.

comment and doctor are the two stages of a convergence loop: comment → triage → fix (doctor / write --feedback / input revision) → re-comment → compare scores. The loop stops at PLATEAU (score delta Δ < 1.0 between iterations). The full routing — triage routes, the Δ/PLATEAU criterion, the score-history artifact, and when input revision (persona:train / tone-finder:improve) fires — is specified in ${CLAUDE_PLUGIN_ROOT}/references/review-loop.md.

Reads the brief model: ${CLAUDE_PLUGIN_ROOT}/references/brief-model.md. The <brief>/ is read-only; review never reads outside <brief>/ and <output>/.

Available actions

#	Action	Role	Input
01	comment	Persona-based chapter evaluation with scoring and triage routing	<output>/chapters/chapter-<NN>.md --brief <brief> [persona-ids…] [--all]
02	doctor	Technical corrections: grammar, typography, terminology, enrichments	<output>/chapters/chapter-<NN>.md --brief <brief> [--remarks "notes" or file] [--dry-run]

Default flow

01 → 02 (standard pipeline). 02 alone if corrections are known. After a fix (02 or write --feedback), return to 01 (re-comment / re-scoring) and compare to the prior score: Δ ≥ 1.0 → loop again; Δ < 1.0 → PLATEAU, stop. Hard guard: max 5 iterations per chapter (cf. review-loop.md).

Trigger-to-action mapping:

• "review chapter", "evaluate chapter", "persona feedback", "comment on chapter", "reader analysis" → comment
• "doctor chapter", "fix chapter", "correct chapter", "apply corrections", "technical review" → doctor

Transversal rules

• comment: embody each persona fully; run mandatory cross-reference check (stats vs summary.md) before scoring; run redundancy check (chapters 02+); complete craft checklist per chapter type; run devil's advocate (3 weaknesses per persona after scoring).
• doctor: preserve original meaning and narrative voice; never use personas (that is comment's role); log ALL corrections to <output>/review/chapter-<NN>-changelog.md.
• Corrections priority: 🔴 Critical (grammar, orthography, terminology) → 🟡 Important (typography, Markdown format) → 🟢 Optional (style, flow).
• Doctor applies ALL priority levels by default; the priorities appear in the report, not as filters.
• Routing from comment to rewrite: if ≥2 personas scored ≤11/20 on structural must-haves → write --feedback. Full triage table (incl. systemic-pattern → tone-finder:improve over ≥3 chapters, and same-persona-capped → persona:train over ≥3 chapters) in ${CLAUDE_PLUGIN_ROOT}/references/review-loop.md.
• comment maintains the score-history artifact <output>/review/chapter-<NN>-scores.md (one row per iteration: consensus, Δ, verdict, route) and never declares PLATEAU while Δ ≥ 1.0.

External data

• ${CLAUDE_PLUGIN_ROOT}/references/brief-model.md — the brief → output working-dir contract.
• ${CLAUDE_PLUGIN_ROOT}/references/review-loop.md — convergence loop: triage routes, Δ/PLATEAU criterion, score-history artifact, input-revision triggers.
• <brief>/summary.md — autonomous brief; sole source of lore, terminology, and context.
• <brief>/personas/ — YAML persona files consumed by comment.
• <brief>/output-styles/ — writing conventions consumed by comment and doctor.
• <output>/chapters/chapter-<NN>.md — chapter(s) to review.
• <output>/review/chapter-<NN>-<persona>.md — persona feedback produced by comment, consumed by write --feedback.