菜单
Targeted re-read of only the bundle fields flagged as uncertain by second-pass-grasp, using precise PDF queries rather than a full re-read. Use this immediately after second-pass-grasp whenever its uncertain_fields output is non-empty; skip straight to extract-structured-bundle when it's empty.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/yogsoth-ai/paper-reading --skill third-pass-verify命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Pick which candidate angle to write the article around, cross-checked against the verified bundle so the chosen angle is actually substantiated by the paper (not just appealing at the skim stage). Use this as the first step of drafting, after quality-assurance has passed, whenever you have candidate_angles from first-pass-skim and a verified bundle.
Writing strategy that turns a verified paper bundle into a finished public-audience article — selecting an angle, applying a pre-set style constraint during drafting (never a post-hoc rewrite), and checking the result for drift against the bundle. Use this after quality-assurance has passed on the bundle — this is the final strategy of the paper-reading pipeline, producing the WeChat article itself.
Deep-reading strategy that turns a single specified academic paper into a structured, source-anchored bundle (problem, method, key results, limitations) ready for fact-checking and article drafting. Use this whenever the user names one specific paper (by arXiv ID, URL, or title) and wants it read, summarized, explained, or turned into content — this is the entry point of the paper-reading pipeline.
基于 SOC 职业分类
正在显示 SKILL.md
| name | third-pass-verify |
| description | Targeted re-read of only the bundle fields flagged as uncertain by second-pass-grasp, using precise PDF queries rather than a full re-read. Use this immediately after second-pass-grasp whenever its uncertain_fields output is non-empty; skip straight to extract-structured-bundle when it's empty. |
| execution | subagent |
| prompt | ./prompt.md |
| input | paper_ref (string), draft_bundle (dict), uncertain_fields (list of strings) |
| output | verified_bundle (dict) |
| dependencies | {"sops":["spawn-agent"]} |
Keshav's third pass, scoped to only the fields second-pass-grasp flagged as uncertain. A no-op (pass-through) when nothing was flagged.
Subagent — spawned via subagent-spawning/spawn-agent skill.
Even a targeted re-read benefits from a fresh context focused only on the flagged fields, so the verification isn't anchored to the same reasoning that produced the original uncertain draft.
This sop calls answer_pdf_queries directly rather than importing literature-research, because it needs field-specific targeted queries rather than a general full-text read — see spec §5's MCP tool mapping table ("direct alphaxiv.answer_pdf_queries calls, targeted by specific claim/field").
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use |
|---|---|
| spawn-agent | Spawn a customized CC subagent with full MCP tool access. |