菜单
Apply the findings of an `/ai-slop:review` report to the source — replace each flagged quote with its suggested revision, and insert `% GROUNDING` TODO stubs for ungrounded citations. Use when the user has a generated `ai-slop-report.md` (or equivalent) and wants the suggestions applied to the paper.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Fill the grounding comments that review only flags as missing. For each `\cite{}` in a LaTeX paper without a quote-backed grounding comment — no comment at all, or a `TODO verify` stub left by revise mode or an earlier run — fetch the cited source, extract a verbatim quote that supports the claim, and write a `% GROUNDING` comment carrying that quote into the source — or a `TODO verify -- <reason>` stub when the source cannot be retrieved. Use when the user asks to ground citations, fill grounding comments, or close the review's grounding to-do. LaTeX source only.
Generate a project-local WRITING.md file from the layered writing rules and add a reference to it in the repository's CLAUDE.md (creating CLAUDE.md if it does not exist). Use when the user wants the writing conventions to live in their repo as an editable, project-local file. Triggers on prompts such as "set up writing rules in this repo", "generate WRITING.md", "init the writing conventions", or `/ai-slop:init`. Writes WRITING.md and creates or amends CLAUDE.md; does not modify your content.
Review only the modified parts of a git-versioned document for AI slop and rule violations. Use when the user has uncommitted edits or a feature branch and wants to audit only what they changed, not the whole draft. Triggers on prompts such as "check my edits", "review what I just changed", "audit this branch's prose", or `/ai-slop:review-diff`. Uses the same layered rules as `/ai-slop:review` (general by default; `--scientific` and LaTeX layers as detected), scoped to the diff. Writes a structured Markdown report that revise mode can apply.
Review a whole code repository's natural-language text for AI slop and rule violations — every Markdown and plain-text file plus the comments and doc-comments of its source and config files, not just one document or a diff. Use when the user wants to audit the prose spread across a codebase (READMEs, changelogs, design docs, and the comments in code and config). Triggers on prompts such as "scan this repo for slop", "check the prose across the codebase", "audit the comments and docs", or `/ai-slop:review-repo`. Loads the general rules by default; `--scientific` adds the research-article layer. Writes a structured Markdown report grouped by file.
Review a document (LaTeX, PDF, or plain prose) for AI slop and rule violations. Use when the user names a draft, hands you a path to a `.tex`, `.pdf`, or text file, or asks to check, audit, or review prose for AI tropes — and, for research papers, statistical reporting, citations, BibTeX correctness, and hallucinated references. The general rules apply by default; `--scientific` adds the scientific layer and LaTeX source loads all three. Writes a structured Markdown report with concrete suggested revisions that revise mode can apply.
| name | revise |
| description | Apply the findings of an `/ai-slop:review` report to the source — replace each flagged quote with its suggested revision, and insert `% GROUNDING` TODO stubs for ungrounded citations. Use when the user has a generated `ai-slop-report.md` (or equivalent) and wants the suggestions applied to the paper. |
| license | CC-BY-4.0 |
| metadata | {"version":"2026-07_rev1","homepage":"https://github.com/se-uhd/ai-slop-skill"} |
This skill applies a previous /ai-slop:review report to the paper. For each finding, it locates the quoted text in the paper and replaces it with the suggested revision.
Audience and tone. The user has already reviewed the report and wants it applied. Trust the report's Suggested revision fields; do not second-guess them unless the surrounding context makes the suggestion unsafe (e.g., the suggestion would break LaTeX syntax or change the meaning of a result).
Invoke this skill when the user:
ai-slop-report.md (or similarly structured report) and wants the findings applied to the paper.Do not invoke when the user wants a fresh review (use /ai-slop:review instead), or when no report file exists.
Both inputs default to the current working directory. No arguments are required.
ai-slop-report.md in the working directory. If that file does not exist, ask the user to point to the report (or to run /ai-slop:review first). The report must match the schema produced by /ai-slop:review (i.e., Findings by section, Cross-cutting metrics, Items requiring author judgment).python3 ${CLAUDE_SKILL_DIR}/../../scripts/find_latex_root.py. Exit 0 → use the printed path; exit 2 → multiple candidates printed, ask the user; exit 1 → no .tex root, stop. PDF input is not supported; revise mode edits LaTeX directly.Optional path overrides. Paths can still be passed as arguments. The first argument is the report path; the second is the paper path.
Read the report. Parse the Findings by section blocks. Each block has Rule, Location, Quote, and Suggested revision. Skip blocks under "Items requiring author judgment"; those need human input.
Read the paper. Open the LaTeX source root and follow \input{} / \include{} to gather the full text. Note the file each section lives in if the paper is multi-file.
Apply each finding. For each finding, in document order:
Quote text in the paper. Use the Location hint (file:line) to disambiguate if the same text appears multiple times.Suggested revision using the Edit tool. One Edit call per finding (do not bundle multiple findings into one edit; that makes diffs harder to review).Location hint does not uniquely identify one: prefer the location closest to the hint and log the ambiguity in the summary.\cite{} keys): log as skipped with the reason rather than apply.Insert grounding stubs. For every \cite{} listed in the report's Grounding to-do section, insert a % GROUNDING: TODO verify <key> comment immediately after that \cite{} call (one Edit per cite, matching the surrounding indentation and comment placement). These are TODO markers for a supporting quote — never invent the quote. The author can fill them by hand, or run /ai-slop:ground, which fetches each cited source and replaces the stub with a retrieved verbatim quote. Skip and log any cite whose location cannot be matched.
Cross-cutting metrics. These are aggregate counts, not individual edits. The specific instances behind them should already appear under "Findings by section". Do not invent new edits to balance a metric.
Summarize. Print a summary to the console with three lists:
Quote was located and replaced, plus grounding stubs inserted.Quote could not be uniquely located, or whose suggestion was unsafe to apply, with reasons.Stop. Do not regenerate the report. Do not commit the changes. The user runs git diff to inspect and git commit when satisfied.
../../shared/rules-general.md, ../../shared/rules-scientific.md, ../../shared/rules-latex.md — the rule layers, referenced only as a fallback when the user asks why a finding was flagged; revise mode trusts the report's suggested revisions and does not re-derive them from the rules.../../scripts/find_latex_root.py — used by the Inputs section to locate the LaTeX root.Revise mode does not load the trope catalog at runtime: the report already contains every suggested revision, so no trope source is needed.