editorial-review

name	editorial-review
description	Document-level STRUCTURAL editor for long deliverables. Reviews a draft's argument architecture -- section ordering, claim-to-evidence linkage, hierarchy, redundancy, buried lede, missing sections -- and proposes CUT / MERGE / MOVE / CONDENSE / SPLIT / ADD / PROMOTE / PRESERVE operations with word-savings estimates, BEFORE any sentence is touched. Content is sacrosanct: it reorganizes, it never argues with the ideas. Phase 2 hands all sentence-level work (rhythm, specificity, vocabulary) to the always-on humanization.md prose rule -- it never duplicates that. Use when the user says "editorial pass", "structural review", "review the structure of this", "tighten this document", "restructure this draft", or wants a long proposal / brief / report's argument arc checked. Do NOT use for: sentence-level prose / "make this sound human" (that is humanization.md), typo or grammar fixes (sanitize-text + humanization-check), fact-checking (/validate), artifact grading (/evaluate), short chat replies, or atomic /zk notes.
argument-hint	[file:<path> | pasted text]
allowed-tools	Read, Edit, Write, Bash(python3:*), Bash(python:*)
metadata	{"author":"Misha Hanin","email":"[e-mail]","version":"1.0"}
x-heading-orchestration	{"parallel_safe":false,"shared_state":[],"triggers":["editorial review","structural review","review the structure","editorial pass","tighten this document","restructure this draft"]}
x-heading-capability	{"what":"A document-level structural editor for long deliverables. It maps a draft's argument architecture and proposes CUT / MERGE / MOVE / CONDENSE / SPLIT / ADD / PROMOTE / PRESERVE operations with word-savings estimates, then applies only the approved ones. It reorganizes; it never rewrites sentences or argues with the ideas.\n","how":"Run /editorial-review file:<path> (or paste the text). Phase 1 presents findings behind an approval gate; Phase 2 hands all sentence-level prose work to the humanization.md rule and runs the sanitize + humanization-check gates.\n","when":"Use on a long proposal, brief, or report whose argument arc needs checking. For sentence-level \"make this sound human\" work use the humanization rule; for fact-checking use /validate; for artifact grading use /evaluate.\n"}
x-heading-routing	{"category":"Operations","label":"/editorial-review [file:<path>]","triggers":["editorial pass","structural review","review the structure of this","tighten this document","restructure this draft"],"exclusions":["Sentence-level prose / \"make this human\" -> humanization.md","typo or grammar fix -> sanitize-text + humanization-check","fact-check -> /validate","artifact grade -> /evaluate","atomic note -> /zk. Document-structure only","hands all prose work to humanization.md."],"compound":"No","router":"auto"}

Editorial Review (structural pass)

Critique and tighten a long deliverable's argument architecture — then hand the prose off to the humanization layer. This skill operates on structure only; it never rewrites sentences and never challenges the ideas.

The full method (document arcs, operation vocabulary, defect checklist, the hard boundary with humanization.md) lives in reference/editorial-review.md. Load it; do not restate it here.

---

Phase 0: Load context

1. Read reference/editorial-review.md (the document-model arcs, the operation vocabulary, the defect checklist, the boundary table).
2. Read .claude/rules/humanization.md — confirm the boundary: this skill stops at the paragraph; everything below it (rhythm, specificity, banned vocabulary, the calibration gate) belongs to that rule.
3. Read the target document in full (the file:<path> argument, or the pasted text). Never review a draft you have only skimmed.

Phase 1: Structural pass

1. Select one document model from reference/editorial-review.md (proposal arc, intel-brief arc, investor-narrative arc, argument/decision-note arc, or a generic model). State the choice and its primary rule in one sentence: "This deliverable should follow the {model}; its primary rule is {rule}."
2. Map the structure — list each major section with an approximate word count. This forces measurement before recommendation.
3. Walk the defect checklist (orphan section, claim-without-evidence, evidence-without-claim, buried lede, flat hierarchy, redundant paragraphs, missing scaffolding, no document-level stance, scope violation).
4. Emit findings, each tagged with exactly one operation (CUT / MERGE / MOVE / CONDENSE / SPLIT / ADD / PROMOTE / DEMOTE / PRESERVE / QUESTION), a one-sentence rationale, and a word-impact estimate. PRESERVE protects comprehension aids from over-cutting (summaries and examples are reinforcement, not redundancy). Close with the summary block: total recommendations, estimated reduction (and % of original), meets-length-target verdict, comprehension trade-offs.
5. Optional — pull a critique method. For a hard restructure, you may run python "${CLAUDE_PLUGIN_ROOT}"/scripts/elicit.py show "Critique and Refine" or "Red Team vs Blue Team" from the shared elicitation catalog and apply it to the argument arc. Skip it when the checklist already bites.
6. APPROVAL GATE. Present the findings. Apply NO structural edit until the CEO approves (per .claude/rules/voice.md: no structural changes without approval). "No substantive changes recommended — the structure is sound" is a valid, explicit completion. On approval, apply only the approved operations, editing structure (move/cut/merge/condense), never wording.

Phase 2: Hand off to the prose layer

1. State explicitly: "Structural pass complete. Sentence-level rhythm, specificity, and vocabulary are now governed by .claude/rules/humanization.md (two-pass voice edit)." Do not perform that work here.
2. Run the standard prose gates on the resulting document:
   • python "${CLAUDE_PLUGIN_ROOT}"/scripts/sanitize-text.py <path> --scan
   • python "${CLAUDE_PLUGIN_ROOT}"/scripts/humanization-check.py <path>
3. Report the validation line: Structural ops applied: N. Hidden characters: clean. Humanisation audit: <result>.

---

Voice rules

• Hyphens, not em-dashes in the operation labels and prose; ODUN.ONE, DPI+, Tribe per .claude/rules/terminology.md.
• The skill proposes; the CEO decides. Frame every finding as a recommendation, never an executed change.
• Sentence case in any headings the skill itself emits.

NEVER

• NEVER do sentence-level prose work — rhythm, specificity density, banned-vocabulary swaps, or the calibration gate. That is humanization.md's job; duplicating it is a defect.
• NEVER challenge or rewrite the ideas. Content is sacrosanct; reorganize only.
• NEVER apply a structural edit before the CEO approves the findings.
• NEVER trigger a prose rewrite on already-human (sub-15%-AI) text — the structural pass operates on the argument layer only and leaves human prose byte-for-byte intact.
• NEVER run on an atomic /zk note or a short chat reply — the structural pass is a near no-op there.
• NEVER declare done without the Phase 2 sanitize + humanization-check gates and the validation line.