Update to Novelist

{
  "description": "Review novel outlines, drafted chapters, full manuscripts, and prior revision findings as an independent analytical editor. Use when the author explicitly requests critique, developmental review, work examination, line editing, proofreading, whole-manuscript review, or regression verification, and when an explicitly fully autonomous workflow covers the complete novel lifecycle from initial planning through publication. Produce an intent-aware, evidence-backed report with Overall Assessment, Work Examination, and Line Editing or Proofreading sections while remaining read-only by default. Do not invoke this skill automatically for ordinary planning, drafting, revision, continuity validation, EPUB export, build, regeneration, packaging, validation, or publication requests.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 218
    },
    {
      "relative_path": "references/report-contract.md",
      "size_in_bytes": 4205
    },
    {
      "relative_path": "references/review-method.md",
      "size_in_bytes": 6616
    },
    {
      "relative_path": "references/review-modes.md",
      "size_in_bytes": 5761
    },
    {
      "relative_path": "scripts/inventory_review.py",
      "size_in_bytes": 17324
    },
    {
      "relative_path": "tests/test_inventory_review.py",
      "size_in_bytes": 19450
    },
    {
      "relative_path": "tests/test_skill_contract.py",
      "size_in_bytes": 4679
    }
  ],
  "name": "analytical-review",
  "skill_md_contents": "---\nname: analytical-review\ndescription: Review novel outlines, drafted chapters, full manuscripts, and prior revision findings as an independent analytical editor. Use when the author explicitly requests critique, developmental review, work examination, line editing, proofreading, whole-manuscript review, or regression verification, and when an explicitly fully autonomous workflow covers the complete novel lifecycle from initial planning through publication. Produce an intent-aware, evidence-backed report with Overall Assessment, Work Examination, and Line Editing or Proofreading sections while remaining read-only by default. Do not invoke this skill automatically for ordinary planning, drafting, revision, continuity validation, EPUB export, build, regeneration, packaging, validation, or publication requests.\n---\n\n# Analytical Review\n\nBefore running a bundled command, resolve `<skill-dir>` to the absolute\ndirectory containing this `SKILL.md`. Keep `--project-root` pointed at the\nactive novel workspace, not at the skill or plugin installation.\n\n## Operating Contract\n\nAdopt an independent reviewer role. Diagnose the work without changing it.\nDo not edit outlines, chapters, canon, the story ledger, or generated output\nduring a review-only request.\n\nIf the author requests both review and revision, finish and present the\ndiagnosis before applying selected findings. Use `novel-story-telling` for the\nrevision phase and preserve finding IDs across the handoff.\n\nWrite all internal instructions, temporary schemas, and script inputs in\nEnglish. Return the review in this order of precedence:\n\n1. the language explicitly requested by the author;\n2. `project.md` `language`;\n3. `style/000.style-guide.md` `language`;\n4. the dominant language of the reviewed source.\n\nLocalize report headings naturally while preserving the semantic section order\ndefined in [report-contract.md](references/report-contract.md). For Korean\noutput, use the canonical labels `작품 총평`, `작품 검토`, and `문면 교열`.\n\nTreat the current manuscript and review response as private runtime material.\nUse relative source anchors and only the shortest excerpt needed for evidence.\nDo not copy manuscript prose or review output into the project, plugin,\nexamples, fixtures, or generated package. Save a report only when the author\nexplicitly requests a file and confirms its destination.\n\n## 1. Confirm Authorization, Mode, And Scope\n\nRead [review-modes.md](references/review-modes.md). Select exactly one mode:\n\n- `outline`: planned story direction before prose exists;\n- `chapter`: one or more explicitly named Drafts;\n- `manuscript`: every numbered publishable chapter plus cross-chapter synthesis;\n- `regression`: bounded verification of earlier findings after revision.\n\nInvoke this skill only when:\n\n- the author explicitly requests review, critique, line editing, proofreading,\n  developmental feedback, or verification of prior findings; or\n- the author explicitly delegates the complete autonomous lifecycle from\n  initial planning through final publication.\n\nDo not infer review authorization from an ordinary planning, drafting,\nrevision, validation, export, build, package, or publish request. `Review and\npublish` authorizes review but not automatic revision. `Proofread and fix`\nauthorizes both phases; keep them distinct.\n\nConfirm the requested source scope. For a chapter request, identify exact\nchapter numbers or relative paths. For regression mode, require the prior\nfinding IDs, dispositions, and changed units.\n\n## 2. Orient To The Work\n\nRead:\n\n- `AGENTS.md`;\n- `project.md`;\n- `style/000.style-guide.md`;\n- `plot/000.master-plot.md`;\n- `outlines/000.master-outline.md`;\n- the sources in the declared review scope;\n- only the linked character, world, material, MacGuffin, and plot files needed\n  to establish intent or continuity.\n\nDo not load every source file by default. Use author-side sources to understand\nintent, rules, and hidden truth. Use only publishable `## Draft` text to claim\nthat a reader knows, sees, or can infer something.\n\nResolve conflicting evidence by the project authority order: explicit author\ndirection > `final` source > `revision` source > latest manuscript fact >\n`draft` source > outline or seed. Keep unresolved conflicts visible.\n\n## 3. Inventory The Review Scope\n\nUse the bundled content-free inventory helper before reading findings into a\nwhole-work conclusion.\n\nFor outline review:\n\n```bash\npython3 \"<skill-dir>/scripts/inventory_review.py\" \\\n  --project-root . \\\n  --mode outline\n```\n\nAdd one or more `--target outlines/<file>.md` or `--target plot/<file>.md`\noptions when the author selected a narrower plan.\n\nFor chapter review:\n\n```bash\npython3 \"<skill-dir>/scripts/inventory_review.py\" \\\n  --project-root . \\\n  --mode chapter \\\n  --chapter <number>\n```\n\nFor manuscript review:\n\n```bash\npython3 \"<skill-dir>/scripts/inventory_review.py\" \\\n  --project-root . \\\n  --mode manuscript \\\n  --max-batch-tokens 12000\n```\n\nThe command prints relative paths, metadata, line ranges, hashes, estimates,\nand batches without copying prose. Prefer this non-persistent output. If an\nexternal file is explicitly needed, pass a fresh, non-existing `--output`\npath; the helper creates it exclusively with private permissions and refuses\ncollisions. Stop short of a complete-coverage claim when `coverage_complete`\nis false. Do not omit a valid numbered chapter because it appears unlikely to\nproduce feedback.\n\n## 4. Build The Review Charter\n\nRead [review-method.md](references/review-method.md). Before producing findings,\nrecord:\n\n- mode, scope, form, genre, point of view, style, and completion stage;\n- work promise, intended reader experience, and target-reader assumptions;\n- project constraints;\n- protected techniques and the boundary of each protection;\n- work-specific review axes;\n- limitations and unresolved premises;\n- any external research required to avoid a material false positive.\n\nDerive axes from this work rather than forcing a universal checklist. Protect\nan intentional device from generic correction while still testing whether each\ninstance performs its intended function.\n\nUse external research only when it can materially change a finding. Prefer\nprimary sources and separate external evidence from manuscript evidence. Do not\nbrowse merely to decorate the report.\n\n## 5. Review Bounded Units\n\nRead each eligible inventory unit in order. For a full manuscript, use the\ninventory batches, but keep one result per chapter or stable bounded span.\nTrack:\n\n- reviewed unit and Draft line range;\n- evidence-backed strengths;\n- work findings and line findings;\n- open questions and unresolved premises;\n- candidate cross-unit patterns;\n- active, revised, or withdrawn review axes.\n\nMark a unit reviewed even when it has no material finding. Complete all unit\npasses before synthesizing whole-work tendencies.\n\nTest patterns with deterministic search when useful. A global tendency requires\nmultiple independent evidence locations. Keep a single local issue local.\nWithdraw or revise a hypothesis when the complete available work disproves it.\n\nIn outline mode, evaluate planned causality, unit function, motivation,\nreader-model progression, setup/payoff, constraints, and direction. Do not\nclaim defects in unwritten dialogue, diction, rhythm, or scene prose.\n\nIn regression mode, verify prior material findings and affected dependencies.\nLimit new findings to revision regressions or directly related blockers. Do not\nreopen deferred or rejected findings without materially changed evidence, and\ndo not begin another unrestricted review.\n\n## 6. Write The Report\n\nRead [report-contract.md](references/report-contract.md). Lead with:\n\n1. Overall Assessment;\n2. Work Examination;\n3. Line Editing / Proofreading.\n\nLocalize those headings in the output language. For Korean, use `작품 총평`,\n`작품 검토`, and `문면 교열`. Keep their semantic order.\n\nDistinguish observation, interpretation, confidence, reader effect, and\nsuggested direction. Use stable `W1` and `L1` style IDs. Present minimal\nrevision options rather than a replacement manuscript.\n\nDo not force a finding count. State when no material finding exists. Put\nstructural dependencies before dependent line edits in the revision order.\n\nFor manuscript mode, append a compact coverage table and compare its reviewed\nunit set with the inventory. Do not claim whole-manuscript completion unless\nevery eligible unit appears and the inventory has no blocking issue.\n\nThe optional method appendix may summarize the charter, sources, limitations,\nand withdrawn axes. Do not expose hidden chain-of-thought.\n\n## 7. Handle The Handoff\n\nFor a review-only request, stop after the report and wait for the author to\nselect findings or request revision.\n\nFor an explicit review-and-publish request that does not authorize revision,\npresent the report, leave all literary findings unapplied, and then hand the\nunchanged manuscript to `publish-novel` for its existing structural and\ncontinuity preflight plus deterministic packaging. Literary findings do not\nbecome build failures.\n\nFor an explicitly authorized review-and-fix request:\n\n1. preserve finding IDs and author dispositions;\n2. apply only selected changes through `novel-story-telling`;\n3. update canonical sources and the story ledger as required;\n4. run one focused regression review unless the author requests broader work.\n\nFor a fully autonomous start-to-publication request, follow the bounded\nhandoff in [review-modes.md](references/review-modes.md): manuscript review,\n`apply`/`defer`/`reject` disposition, story revision, one focused regression\npass, strict continuity validation, then deterministic `publish-novel`\npackaging.\n\nNever inject this loop into an ordinary publication request. Deferred or\nrejected literary findings are handoff state, not EPUB validation errors.\n\n## Completion Gate\n\nDo not finish until:\n\n- the mode and source scope are explicit;\n- the charter records protected techniques, axes, limitations, and uncertainty;\n- reader-effect claims use reader-visible Draft evidence;\n- every unit in scope has been inspected;\n- full-manuscript coverage matches the deterministic inventory;\n- strengths and recurring tendencies have evidence;\n- outline review defers unwritten prose judgments;\n- observations, interpretations, and revision options remain distinguishable;\n- the report contains the three localized semantic sections in order;\n- structural dependencies precede dependent line edits;\n- review-only work left source files unchanged;\n- no manuscript or review output was persisted without explicit authorization.\n"
}