chapter-reviewer

name	chapter-reviewer
description	Review and critique a chapter for craft quality, voice consistency, and AI-tells. Use when: (1) User says "Kapitel reviewen", "review chapter", (2) After chapter-writer completes a draft. Memoir books → use `/storyforge:chapter-reviewer-memoir` instead.
model	claude-opus-4-7
user-invocable	true
argument-hint	<book-slug> <chapter-slug>

Chapter Reviewer

Step 0 — Resolve Book Category

Read book_category from the review brief (Step 1). Treat missing as fiction.

If book_category == "memoir" → invoke /storyforge:chapter-reviewer-memoir instead. Do not continue in this skill.

Prerequisites — MANDATORY LOADS

Step 1 — Load the review brief (single MCP call, replaces 6+ direct file reads)

Call MCP get_review_brief(book_slug, chapter_slug). This returns:

• book_category — must be fiction; if memoir → see Step 0
• chapter_timeline — intra-day time grid for this chapter
• previous_chapter_timeline — same for the preceding chapter
• canonical_timeline_entries — parsed plot/timeline.md events
• travel_matrix — parsed world/setting.md Travel Matrix rows
• canon_log_facts — parsed plot/canon-log.md facts
• tonal_rules — non-negotiable rules, litmus test, banned patterns from plot/tone.md
• active_rules — book CLAUDE.md ## Rules with severity
• active_callbacks — book CLAUDE.md ## Callback Register items
• errors — non-empty means some files missing; degrade gracefully, do not invent

Honor every populated field. Empty lists / null means "file missing — degrade gracefully."

Step 2 — Load author and craft context

• Author profile via MCP get_author(). Why: Voice consistency check needs the documented baseline. writing_discoveries.recurring_tics (Issue #151) lists cross-book tics — flag any hit as Major findings. style_principles and donts feed the same review pass.
• Author vocabulary from ~/.storyforge/authors/{slug}/vocabulary.md. Why: Banned-word scan and preferred-word check both run against this list.
• Craft references via MCP get_craft_reference():
  • dos-and-donts — general craft baseline for the Craft section (5 points).
  • anti-ai-patterns — AI-tell catalog for the Anti-AI section (5 points).
  • chapter-construction — hook/scene-sequel/ending criteria for the Structure section (5 points).
  • dialog-craft — subtext, voice differentiation, tag discipline — for points 9 and 15.
  • show-dont-tell — show/tell balance check for points 6 and 24.
  • simile-discipline — the two-question test for point 10b.
• Detect if Chapter 1: Check chapter slug (starts with 01- or 001-) or frontmatter chapter number. If Chapter 1: also load openings-and-endings craft reference.

Step 3 — Read the prose (direct file reads)

• Read the chapter draft: {project}/chapters/{chapter}/draft.md
• Read the chapter outline: {project}/chapters/{chapter}/README.md
• Read previous chapter draft for continuity context
• Optional: If {project}/research/manuscript-report.md exists, check whether any of THIS chapter's distinctive 5-7 word phrases appear in earlier chapters. Flag matches in the Continuity Report.

First Chapter Checklist — 13 Points (ONLY for Chapter 1)

If this is Chapter 1, run this checklist BEFORE the standard review. Rate each point: PASS / WARN / FAIL.

Set the Stage

1. Orient the reader — Time period and location established early without info dump.
2. Set the tone — Words, events, dialogue, and setting convey the story's mood.
3. Establish the genre — Reader knows what type of story they're entering.

Spotlight the Protagonist 4. Protagonist in their element — Shown doing what defines them best (not sitting, not waking up). 5. Protagonist wants something — A concrete desire or longing exists before the main plot hits. 6. Normal world revealed — Everyday life before disruption is visible. 7. Problem illustrated — Something is broken or unsatisfying in the protagonist's world. 8. Theme seeded — A subtle hint at the story's deeper questions exists. 9. Internal conflict established — Fear-versus-desire tension is present.

Give Readers a Reason to Stay 10. Killer first sentence — Opening line sparks curiosity and makes the reader ask a question. 11. Curiosity sparked — Unanswered questions or intriguing dialogue keep reader leaning forward. 12. Dread established — Coming challenges are foreshadowed. 13. First domino knocked — The initial event that sets the journey in motion occurs by chapter's end.

Load-bearing items (FAIL here = chapter needs revision before moving on): 4, 5, 9, 10, 13.

---

Review Checklist — 28 Points + 1 sub-point

Structure (5 points)

1. Opening hook — Does the first paragraph grab? Would you keep reading?
2. Scene-sequel flow — Does each scene have goal/conflict/outcome?
3. Chapter arc — Does something CHANGE from start to finish?
4. Ending — Does it compel the reader to turn the page?
5. Pacing — Does the chapter breathe? Action/reflection balance?

Craft (5 points + 1 sub-point)

6. Show don't tell — Are emotions shown through action/body, not named?
7. Sensory details — Are multiple senses engaged (not just visual)?
8. Specific details — Concrete nouns and precise verbs, not generic descriptions?
9. Dialog quality — Subtext present? Characters sound different? Minimal tags?
10. Conflict — Is there tension in every scene? No filler?

10b. Simile discipline (craft sub-point) — Apply the two-question test from simile-discipline.md to every like, as if, as [adj] as, the way [X], moved/felt/sounded like, and the kind of [noun] that [clause] construction. For each: does the vehicle literally resemble the tenor? Does it do work a concrete beat couldn't? Flag illogical or decorative similes, stacked similes (2+ per paragraph without each doing distinct work), and dead similes. Respect author-voice bias.

Voice (5 points)

11. Author consistency — Does this sound like the defined author profile?
12. Tone match — Does the tone match the profile's descriptors?
13. Vocabulary — Any banned words used? Preferred words present?
14. Sentence rhythm — Varied length? Matches author's style?
15. Dialog voice — Each character distinguishable without tags?

Continuity (5 points + 1 sub-point)

16. Canon consistency — Does the chapter contradict any fact in the Canon Log? Pay special attention to CHANGED facts.
17. Timeline accuracy — Do day/date references match plot/timeline.md?
18. Travel consistency — Do distances/travel times match the Travel Matrix?
19. Stale references — Does this chapter's slug appear in the revision_impact list of any **CHANGED** bullet in plot/canon-log.md? If so, verify the chapter uses the NEW version of every changed fact.
20. Character facts — Do character descriptions/behaviors match established facts? 20a. POV knowledge boundary — Does the narration attribute domain knowledge the POV character's profile says they don't have? Three remediation options: (a) move into dialog, (b) reframe as lay observation, (c) cut.

Plot Logic (5 points)

Load analyze_plot_logic(book_slug, scope="chapter", chapter_slug=...) once before scoring this section. The returned knowledge_index provides facts, promises, and chapter story-days.

20b. Information leak — Does the POV character reference any fact established in a later chapter, or in an earlier chapter where the POV was absent? Severity: high (FAIL) if demonstrably absent; WARN otherwise. 20c. Motivation chain — Does each significant decision follow from the character's established wants and knowledge? Flag contradictions without on-page justification. Severity: WARN. 20d. Causality direction — Run static causality_inversion findings for this chapter. FAIL on any. 20e. World rule consistency — Does the chapter break any rule from the canon log "World / Setting Facts" or world/rules.md? Look for negation patterns that the prose violates without an exception clause. Severity: high (FAIL). 20f. Chapter promise — If the chapter places a setup-element, is it logged in ## Promises? Suggest missing promises; on approval call register_chapter_promises. Severity: WARN.

Tonal Consistency (5 points) — only if plot/tone.md exists

21. Dominant mode — Does the chapter match the dominant mode for this chapter's position?
22. Warning signs — Does the chapter exhibit any warning-sign patterns for this stage?
23. Non-negotiable rules — Are ALL non-negotiable rules satisfied?
24. Litmus test — Answer every question from the Litmus Test section.
25. Banned patterns — Does the chapter use any book-specific banned prose patterns?

Intra-Day Timeline (3 points)

26. Time anchor — Does the chapter establish when it starts? Consistent with previous chapter?
27. Internal consistency — Do relative time references match the Chapter Timeline in README.md?
28. Cross-chapter consistency — Do references to earlier events match the previous chapter's timeline?

Anti-AI (5 points)

21. AI vocabulary — Any words from the banned list?
22. Structural uniformity — Paragraphs/sentences suspiciously uniform in length?
23. Generic descriptions — Any "bustling city", "warm smile", "piercing gaze" clichés?
24. Emotional telling — Any "he felt a wave of sadness" instead of showing?
25. Neat resolution — Does every scene wrap up too tidily?

Output Format

Report target: 800–1200 words total. Critical findings first, then Recommended, Minor only if a pattern emerges (3+ instances). Skip empty sections — if there are zero AI-tells, the AI-Tell Report is one line ("Flagged words: none. Variance high. Generic descriptions: 0."), not a full table.

## Chapter Review: {Chapter Title}

### First Chapter Report (only if Chapter 1)
| # | Requirement | Result | Notes |
|---|---|---|---|
| 1 | Orient the reader | PASS/WARN/FAIL | |
| 2 | Set the tone | PASS/WARN/FAIL | |
| 3 | Establish genre | PASS/WARN/FAIL | |
| 4 | Protagonist in element ⚠️ | PASS/WARN/FAIL | |
| 5 | Protagonist wants something ⚠️ | PASS/WARN/FAIL | |
| 6 | Normal world revealed | PASS/WARN/FAIL | |
| 7 | Problem illustrated | PASS/WARN/FAIL | |
| 8 | Theme seeded | PASS/WARN/FAIL | |
| 9 | Internal conflict ⚠️ | PASS/WARN/FAIL | |
| 10 | Killer first sentence ⚠️ | PASS/WARN/FAIL | |
| 11 | Curiosity sparked | PASS/WARN/FAIL | |
| 12 | Dread established | PASS/WARN/FAIL | |
| 13 | First domino knocked ⚠️ | PASS/WARN/FAIL | |

⚠️ = load-bearing, FAIL requires revision before proceeding.

---

### Score: [X]/25 (core) + [X]/5 (tonal, if tone.md exists) + [X]/3 (timeline)

### Strengths
- *What works well*
- *Specific quotes that shine*

### Issues

#### Critical (Must Fix)
- [Issue] — [Location] — [Suggested fix]

#### Recommended (Should Fix)
- [Issue] — [Location] — [Suggested fix]

#### Minor (Nice to Have)
- [Issue] — [Location] — [Suggested fix]

### Continuity Report
- Canon conflicts: [count] — [details]
- Timeline conflicts: [count] — [details]
- Travel Matrix conflicts: [count] — [details]
- Stale references (from revisions): [count] — [details]

### Plot Logic Report
- Information leaks (high): [count] — [details with evidence]
- Information leaks (warn): [count] — [details]
- Motivation breaks: [count] — [details]
- Causality inversions: [count] — [details]
- World-rule violations: [count] — [details]
- Promise log: [N placed, M logged, K missing]

### Tonal Report (if tone.md exists)
- Dominant mode match: [yes/no — expected: X, actual: Y]
- Warning signs triggered: [list or "none"]
- Non-negotiable rules: [PASS/FAIL per rule]
- Litmus test: [pass/fail per question]
- Banned patterns found: [list or "none"]

### Chapter Timeline Report
- Time anchor: [established / missing / inconsistent with previous chapter]
- Internal time conflicts: [count] — [details]
- Cross-chapter time conflicts: [count] — [details]

### AI-Tell Report
- Flagged words: [list]
- Sentence length variance: [high/medium/low]
- Generic descriptions found: [count]

### Simile Report
- Total simile markers found: [count]
- Illogical / decorative (cut or revise): [list with quote + location]
- Stacked (2+ per paragraph without distinct work): [list]
- Dead similes: [list]
- Passing similes: [count]
- Author-voice register applied: [sparse / simile-heavy / as documented in profile]

### Verdict
[PASS / NEEDS REVISION / MAJOR REVISION]

VERDICT: PASS | WARN | FAIL

### Suggested Next Step
[/storyforge:chapter-writer to revise | move to next chapter]

Verdict mapping:

• PASS ↔ no Critical issues, no AI-tell banlist hits, all load-bearing first-chapter rows clear.
• WARN ↔ Recommended issues exist but nothing blocks moving on after a single targeted pass.
• FAIL ↔ at least one Critical issue, any AI-tell banlist hit, or any load-bearing first-chapter row at FAIL.

Rules

• Be BRUTALLY honest.
• Pair every critical finding with a specific strength — the report is signal, not just complaint.
• Quote specific lines when flagging issues.
• Suggest concrete rewrites, not just "make this better."
• The AI-tell check runs as a hard gate — if banned words appear, flag the chapter as NEEDS REVISION regardless of other scores.
• When the user flags an issue: VERIFY before accepting. Re-read the passage, check context from earlier chapters, and push back if the user misunderstood (especially English nuances).