article-workflow-section-review

Article Workflow Section Review

这是流程类 Skill，默认仅在用户显式调用 $article-workflow-section-review 时运行。

This is Phase 1 of the article-workflow-* skill series, corresponding to "Section-by-Section Narrative Review" in the article optimization workflow.

Goal: Review article materials section by section for narrative structure issues, providing actionable diagnostics for subsequent editing. The AI produces editorial diagnostics and default resolution suggestions — it does not revise body text, polish, or make key trade-off decisions on behalf of the author. The author makes judgment calls; trade-offs that would affect the direction of subsequent edits must have clearly marked author response slots, and can be collected in advance via dialog when necessary.

Trigger Scenarios

Use this skill when the user asks to:

• Execute article optimization workflow Phase 1
• Perform section-by-section narrative review on .article-workflow/00-cleaned-sources/
• Generate .article-workflow/01-section-reviews/
• Check sections for logical jumps, information density, reader interest, or material selection
• Organize review results into actionable editing references for subsequent agents or the author

Inputs and Outputs

Default inputs:

• .article-workflow/brief.md
• .article-workflow/00-cleaned-sources/*.md

Default outputs:

• .article-workflow/01-section-reviews/<original-filename>.review.md: per-section review
• .article-workflow/01-section-reviews/index.md: full-text issue index and global decision entry point

If the user provides an article directory, first look for .article-workflow/ under that directory. If brief.md is missing, ask whether to proceed; without a brief, only basic narrative diagnostics can be performed — do not infer the article's main thread, target readers, or material selection criteria on your own.

Core Principles

• Review, do not revise.
• Diagnose narrative structure; do not polish language.
• Every identified issue must have a default resolution path; do not only point out problems.
• Items requiring the author's decision must be explicitly marked and include a 📝 Author Response: slot.
• When the author has not responded, subsequent phases may follow the "default resolution path"; author responses override default paths.
• For blocking decisions marked as "Requires Author Confirmation: Yes" that would affect section splitting, material retention/removal, factual boundaries, or the author's viewpoint, Phase 1 may provide a default path, but subsequent editing phases must not execute such blocking decisions without author confirmation.
• Review files must serve subsequent editing; do not write generic editorial opinions.
• Do not impose standardized writing frameworks to reshape the author's article; only point out how existing materials can better serve the brief.
• Preserve the author's personal experience, article tone, and necessary judgments; deletion or compression suggestions must not override author intent by default.
• After each round of output is complete, a sub-agent must be launched for goal-alignment review; the sub-agent only raises review opinions, and the main agent decides whether to adopt them.
• If a sub-agent is unavailable, a self-check covering the same review focus areas must be performed, and the final response must note the limitation of not completing a sub-agent review.