Developmental Edit

Perform professional developmental editing on a novel manuscript. Produce per-chapter edit reports, an incrementally built world-bible, a full editorial review, and a verification pass on revised chapters.

Scope: Structure, character, pacing, theme, continuity, and reader experience. This is not line editing, copy editing, or proofreading. Give sentence-level feedback only when it affects readability, clarity, or narrative function at a structural level.

Definition of done: the author can open the editorial review, see what is actually wrong and in what order to fix it, locate every piece of evidence, and — after revising — re-run the skill to learn which issues are resolved and which are not.

Read This Before Using the Skill

Developmental judgment is contestable, and this skill produces it at scale. Two things follow.

• Every finding is evidence, not a verdict. You hold the plan, the reasoning, and the locators; the author decides what the book is. Report divergences instead of resolving them, and let a contested call stay contested.
• Structure is where a machine is least reliable and most confident. Pacing, act balance, and whether an ending lands are book-level judgments read off chapter-level analysis. Present them as hypotheses with evidence, never as measurements.
• Never invent a structural problem to fill a template. A chapter that works gets a clean report and a Keep recommendation. That is a legitimate outcome.
• Do not overrule the author's declared intent. characters/*.md is intent; the manuscript is execution. A mismatch is a finding to report, not a defect to correct.
• The critical calls deserve a person. Say plainly that a structural revision at this scope benefits from the author, a trusted reader, or a human editor arbitrating the Critical list before a revision cycle is spent on it.

Inputs

• A folder path containing chapter files (see references/ingestion.md for formats, including Word and Scrivener exports)
• Optional flags or stated values: genre, audience, themes, comparable titles, title, author
• Optional author files (templates in assets/templates/):
  • characters/*.md — produced by character-interview; read them, do not duplicate the schema here
  • MANUSCRIPT.md — title, genre, audience, comparable titles, themes, and the author's questions (assets/templates/manuscript.md)
  • ORDER.md — explicit reading order (assets/templates/order.md)
  • chapters.json — the same order as a JSON array (assets/templates/chapters.json)
  • Chapter files may carry YAML properties — copy assets/templates/chapter-note.md; see references/ingestion.md
• Existing editorial-review/ folder from a prior run (continue, re-process, or verify)

Read characters/*.md first if it exists. Those files are the author's declared intent — arc, false belief, voice. This skill must never contradict them silently; a mismatch between the character file and the manuscript is a finding to report.

Output Contract

<project_folder>/editorial-review/
├── metadata.md
├── status.json
├── world-bible.md
├── editorial-review.md
└── chapters/
    ├── ch01.md
    ├── ch02.md
    └── archive/
        └── ch02.rev1.md

Never invent a second layout. Chapter reports always live in chapters/; superseded reports move to chapters/archive/.

• status.json is the authoritative machine state: which chapters were analyzed, from which source file, when, with what verdict. Resume from it, not from guesswork.
• If characters/ exists, the world-bible links to those files instead of duplicating them. Character files are intent; the world-bible is what the manuscript establishes.
• If the project has a legacy Editorial Review/ folder, use it as-is and say so. Folders are lowercase only for new projects.

Workflow

0. Pre-flight

Before analyzing anything:

1. Scan for chapter files. Handle nested folders, part files, front and back matter, and version suffixes per references/ingestion.md.
2. Sort naturally (ch1, ch2, ch10); detect Prologue, Epilogue, Interlude by name. If ORDER.md or chapters.json exists, validate it (files exist, no gaps, no unlisted files) and treat it as authoritative. If both exist, compare them: when they disagree, stop and ask which one is authoritative rather than picking one.
3. Count words and chapters. Present the manifest — a numbered table of file, word count, type — and state the plan and rough scale.
4. Confirm scope before starting. Ask about anything ambiguous: which draft if version suffixes are present, whether a partial draft should get a structural verdict, whether to include back matter.
5. If the folder is empty or unreadable, say so and stop.

1. Discovery

1. Read characters/*.md if present — author intent for characters, arcs, and voice.
2. Discover genre, audience, and themes in priority order:
   • Existing editorial-review/metadata.md
   • MANUSCRIPT.md / user flags
   • Comparable titles — ask directly: "Name two books this sits next to on the shelf."
   • Inference from the first 2–3 chapters
3. Cross-validate stated vs. inferred values. If confidence is low or multiple genres are plausible, pause, present options with qualitative confidence (strong / moderate / weak signals — never invented percentages), and wait for confirmation. Until confirmed, use only universal developmental principles.
4. Load genre and audience heuristics from references/genre-audience-themes.md after confirmation.
5. Do not override a clearly stated genre or audience without reason.

2. Structural Pre-Pass

Read for shape before reading for detail:

1. Read each chapter's opening and closing pages, and skim for act boundaries, turning points, and POV/tense changes.
2. Build a structural map: act or part boundaries, the tension curve, the major turns, and where the manuscript's center of gravity sits.
3. Write the map into world-bible.md under Structural Map.
4. Ask the author which structural model they're working to, if any (three-act, seven-point, save-the-cat, none). Do not impose one.

Chapter reports produced before this map exists will judge pacing locally and get it globally wrong. Build the map first, then analyze chapter by chapter against it.

3. Setup

1. Create editorial-review/ and editorial-review/chapters/ if needed.
2. Write or update metadata.md (copy assets/templates/metadata.md). Set Last Updated.
3. If world-bible.md does not exist, copy assets/templates/world-bible.md and fill the header.
4. Create or load status.json (copy assets/templates/status.json).
5. Resume check: read status.json and compare against the source files. Detect chapters that are unanalyzed, analyzed, or changed since last analysis (source modified later than analysis). Offer: continue from the next unprocessed chapter, re-process specific chapters, verify revised chapters, or full re-run.
6. If the manuscript changed after editorial-review.md was generated, mark the review stale in status.json and tell the author.

4. Process Each Chapter

For every chapter in order:

1. Read the full text.
2. If the chapter exceeds ~12–15k words, analyze in scene or beat blocks but still write one report file. Split at scene breaks (* * *, ***, or a blank line around a centered marker); if none exist, split into equal blocks and note that in the report.
3. Load references/chapter-edit-criteria.md and the relevant genre/audience section of references/genre-audience-themes.md.
4. Analyze against the full criteria set, following references/report-standards.md for ratings, verdicts, locators, and priority labels.
5. Copy assets/templates/chapter-edit.md, fill it, and write to editorial-review/chapters/<source-basename>.md.
6. Update the world-bible using the merge protocol in references/world-bible-guide.md (net-new information only), with confidence tags and the character-knowledge matrix from references/continuity-heuristics.md.
7. Update status.json for this chapter.
8. Give the user a short progress note: main strength, verdict, critical issues, word count.

Quality self-check before writing each report:

• The report opens with a verdict line (purpose met / partial / not met) and a recommendation (keep / trim / merge / cut)
• Every Critical and Important item has a concrete, actionable suggestion and a locator (ch07 ¶12)
• Recommendations are grounded in this chapter (brief paraphrase or short quote, located)
• Ratings are applied consistently: Strong / Adequate / Needs work
• No pure line-editing advice unless it affects structure, pacing, clarity, or character
• Priority labels are consistent: 🔴 Critical / 🟡 Important / 🟢 Optional / ❓ Decision needed

5. Revision Verification

When chapters/ already contains a report for a chapter the author has since revised, do not silently re-analyze it. Load references/revision-verification.md and:

1. Read the prior report's Critical and Important items.
2. Check each against the revised text and mark it Resolved / Partially resolved / Not addressed / Regressed / No longer applicable, with a locator and one line of evidence.
3. Note any issue the revision introduced.
4. Write the verification into the updated report (a ## Revision Verification section) and archive the superseded report to chapters/archive/<basename>.rev<N>.md.
5. Update status.json.

This is the loop authors actually work in. A re-analysis from scratch destroys the record of what was asked for and whether it landed.

6. Final Editorial Review

After all chapters:

1. Load references/editorial-review-guide.md.
2. Cross-reference chapter reports for systemic patterns — including repetition: the same beat, argument, description, or emotional move recurring across chapters.
3. Produce the mandatory Theme Alignment Report (Stated / Detected / Tension).
4. Answer the author's own questions from MANUSCRIPT.md in a dedicated section.
5. Compile the Decisions Pending register from the world-bible.
6. Run a final continuity sweep using references/continuity-heuristics.md.
7. If any chapter was verified, include the Revision Verification Summary.
8. Copy assets/templates/editorial-review.md, fill it, and write editorial-review/editorial-review.md. Clear the stale flag in status.json.

7. Deliverable Summary

• List all outputs created or updated
• Highlight top Critical issues and overall readiness
• Present the genre / audience / theme summary (stated vs. inferred, fit, alignment)
• Report the Decisions Pending count and the verification results, if any
• Offer to dive deeper into any chapter, verify a revision, or interview the character files further

Report Standards (summary)

Full rules in references/report-standards.md.

• Verdict first. Every chapter report opens with purpose met?, top priority, and keep/trim/merge/cut.
• Ratings per section: Strong / Adequate / Needs work — so reports are comparable and the final review can aggregate them.
• Locators on every quote. ch07 ¶12, or ch07 "opening four words…" when paragraphs are unnumbered.
• Priority includes effort. Scope (line / scene / chapter / structural) and rough effort (light / moderate / heavy).
• ❓ Decision needed is a first-class marker. Not every finding is a defect; some are questions about author intent.

Key References (load on demand)

Templates (copy, then fill)

Outputs — these belong in editorial-review/.

Author input files — these belong in the manuscript folder, alongside the chapter files. All are optional.

Edge Cases (summary)

• Empty folder → notify and stop
• Unreadable files → warn, skip, list them
• .docx or other binary format → extract to text first (references/ingestion.md); if extraction is unreliable, warn and degrade, never abort silently
• Nested folders or part files → follow the structure; do not flatten
• Version suffixes (ch02_v2, ch02-final) → ask which draft is live; never analyze both
• Partial draft → analyze what exists; qualify any structural verdict as provisional
• Front / back matter → classify, do not treat as chapters; include only if the author asks
• Numbering gaps → flag and continue
• Near-duplicate content → warn of possible copy-paste error
• Report filename collision (two sources, same basename) → disambiguate and say so
• Chapter < 300–500 words → flag; still produce a lightweight report
• Chapter > ~12–15k words → scene/beat analysis; one report
• Prologue / Epilogue / Interlude → process as chapters; mark special status
• Multiple POVs or dual timelines → track in world-bible; check POV/tense consistency
• Author notes inside chapters → ignore content between <!-- author-note --> and <!-- /author-note -->; also ignore [TK], TODO, and [[double-bracket]] markers
• Existing editorial-review/ folder → resume from status.json; never silently overwrite metadata or reports
• Existing folder under a legacy name or casing (Editorial Review/) → use the existing folder as-is and say so; do not rename or duplicate it
• Encoding issues → note unreadable segments and continue
• Non-English or mixed-language text → note language; apply developmental criteria to structure and character

Priority Markers

• 🔴 Critical — Must address before submission; major structural or continuity risk
• 🟡 Important — Significantly improves quality
• 🟢 Optional — Polish / nice-to-have
• ❓ Decision needed — Unresolvable without knowing author intent; pose it as a question

Working With the Other Skills

• character-interview — if characters/*.md exists, those files are the author's declared intent for arc, false belief, and voice. Read them, link to them, and report divergences rather than resolving them silently. If a major character has no file and their arc is unclear, say so and suggest an interview.
• line-edit — sentence- and paragraph-level work. Do not line-edit a chapter that carries open Critical structural notes; those scenes may be rewritten. Structural revision comes first, polish after.
• copyedit — the third pass, after line editing. If your chapter reports carry open Critical notes, that skill will refuse to copyedit an unstable text, so clear them first. The world-bible and chapter reports are useful evidence when it queries a continuity discrepancy.
• proofread — the final pass, after the copyedit and after formatting. It refuses to run until the copyedit's query log is closed and the text is frozen, because proofreading text that is still changing produces errata for text that no longer exists.

What This Skill Does Not Do

• Rewrite the manuscript
• Perform line editing or proofreading as the primary service
• Auto-correct continuity errors (flag only)
• Invent genre or audience when the author has already stated them clearly
• Invent confidence percentages, page numbers, or evidence locators