Post-Manuscript Benchmark Review

Use this skill as a post-draft quality gate. It reviews a complete or substantial academic draft against benchmark papers and produces the next-version revision plan.

Activation Gate

Proceed only when a complete initial manuscript, integrated thesis chapter, or substantial full draft exists.

Do not run this skill for topic selection, outline design, literature discovery before drafting, PDF/DOCX parsing alone, figure/table generation alone, citation verification alone, formatting, or de-AI polishing before a full draft exists. Route those tasks upstream.

Boundaries

Do:

• Judge a full draft against target standards and 3-10 benchmark papers.
• Identify manuscript gaps, claim risks, evidence needs, and next-version P0/P1/P2 actions.
• Analyze reviewer/editor/supervisor comments in the context of a complete draft.
• Produce a ten-dimension 100-point quality review and target-readiness judgment.
• When the user requests an editorial-style SCI/Q1 audit, delegate bounded checks to a standards-review agent for scope fit, novelty, and scientific logic before final judgment.
• Use page, paragraph, and sentence anchors whenever the source format exposes them.

Do not:

• Fabricate benchmark papers, DOI records, journal levels, acceptance probabilities, or benchmark-paper contents.
• Treat abstract-only or metadata-only benchmark information as full-paper evidence.
• Rewrite the manuscript as the primary task; route confirmed revision execution to $revision-control.
• Perform citation identity checks as final authority; route to $academic-research-verification.
• Maintain sentence-level revision state, object libraries, or pass/fail decisions; route to $revision-control.

Intake

Collect draft path/content, target paper type, target journal/school/quality tier, reference list, evidence register, figures/tables, results/data, code or analysis scripts, benchmark papers, reviewer/supervisor comments, and missing-material constraints.

Before reviewing, ask for missing materials that would materially improve review quality:

Required or high-value materials:
Missing materials:
Can proceed without them: yes/no
Fallback if unavailable:

If key inputs are unavailable, produce a provisional review plan and materials request instead of a final readiness judgment.

Workflow

1. Confirm the post-draft gate and review target.
2. Build or validate the 3-10 benchmark set with benchmark-selection-and-extraction.md.
3. Apply the ten-dimension 100-point review, four-criterion diagnostic lens, and claim triage in review-rubric.md.
4. If the user asks for editorial-style SCI/Q1 screening, run standards-review-agent.md and merge its scope-fit, novelty, and scientific-logic findings before the final judgment.
5. Use benchmark-report-schema.md to keep benchmark facts, access limits, manuscript claim records, gap severity, and next-version actions auditable.
6. Use reviewer-comment-response-workflow.md and reviewer-comment-action-plan-gate.md when concrete reviewer, editor, committee, or supervisor comments are provided for a complete draft.
7. Use main-text-and-rebuttal-claim-support-gate.md before approving factual, boundary, subjective, contribution, limitation, or rebuttal judgments.
8. Use cross-disciplinary-language-review-gate.md when formal review output must judge abstract wording, undefined local labels, operation-record prose, sentence purpose, citation proximity, proxy validity, or bilingual claim-strength consistency.
9. Use multi-agent-academic-workflow-gate.md before launching independent review checks for structure logic, evidence support, language gate issues, citation proximity, figure/format QA, reviewer-comment action planning, or editorial-style standards screening.
10. Use review-output-template.md for formal Chinese reports, readiness judgments, benchmark gap tables, standards-audit tables, and next-version plans.
11. Return readiness level, ten-dimension scores, evidence limits, benchmark gaps, standards-audit findings, P0/P1/P2 actions, material requests, rollback risks, and $revision-control handoff packets.

Quality Gate

The review is high quality only if it:

• names exact draft sections, figures, tables, claims, data, code, or evidence-register keys behind each critique;
• separates benchmark facts from inference;
• distinguishes fatal blockers from optional polish;
• gives actions executable in the next manuscript iteration;
• marks missing evidence, benchmark access limits, and material dependencies explicitly.
• turns two or more review comments into a triaged action plan before manuscript rewriting.
• applies the cross-disciplinary language review gate when wording quality, terminology clarity, sentence purpose, or support distance is part of the review.
• plans multi-agent parallel checks by default for substantial benchmark review, language audit, evidence-support review, figure/format QA, or reviewer-comment decomposition, then merges findings before final judgment.
• uses page, paragraph, and sentence anchors whenever the source format exposes them.
• does not directly edit the manuscript; it sends confirmed modification tasks to $revision-control.

Reference

Read benchmark-selection-and-extraction.md when selecting or extracting benchmark papers.

Read review-rubric.md when scoring, gap-analyzing, judging readiness, or triaging claims.

Read benchmark-report-schema.md when producing benchmark material records, manuscript claim records, benchmark gap tables, or next-version action tables.

Read reviewer-comment-response-workflow.md when user-provided review comments must be accepted, partly accepted, rejected with reasons, prioritized, and turned into a next-stage plan.

Read reviewer-comment-action-plan-gate.md when two or more review comments must become concrete modification actions.

Read main-text-and-rebuttal-claim-support-gate.md before treating benchmark, readiness, contribution, limitation, or rebuttal judgments as final prose.

Read cross-disciplinary-language-review-gate.md when reviewing formal body text or rebuttal prose for abstract, mechanical, internally defensive, undefined-term, citation-unclear, proxy-unsupported, or cross-language-inconsistent wording.

Read multi-agent-academic-workflow-gate.md when a complete draft review can be split into parallel checks. Use it for sub-agent scope, output format, confirmation gates, conflict handling, and main-agent merge rules.

Read review-output-template.md when producing a formal report, Chinese evaluation, SCI/Q1 readiness judgment, or benchmark gap table.

Read standards-review-agent.md when the user asks for editorial-style SCI/Q1 screening, desk-reject risk review, Applied Energy-level comparison, or page-by-page, paragraph-by-paragraph, and sentence-by-sentence audit.

Read revision-control-contract.md when turning review findings into a formal revision queue.