go

name	go
description	Reads and summarizes ONE specific paper. Triggers on "read this paper", "summarize X", or when the user gives a specific paper title/link.

go — paper reading & summary

Process

1. Fetch the paper. If given a link, use WebFetch to retrieve it. If only given a name/title, use WebSearch first to locate the canonical source (arXiv, venue page, OpenReview) before fetching.
2. Extract the core structure. Pull out: problem statement, method/approach, key results (with actual numbers, not vague claims), and stated limitations. Don't skip limitations — authors' own caveats are high-signal.
3. Note venue and year. Record whether this is a peer-reviewed publication or an arXiv-only preprint, and the year.
4. Write a structured summary note. Create a per-paper Markdown file under _papers/ with sections for Problem, Method, Results, Limitations. Cite the paper itself at the top using the Markdown link citation format:

   [Authors et al., Year — Title](link)
   

5. Flag claims needing verification. If the paper makes a strong or surprising claim, flag it inline (e.g. "⚠ needs check") rather than silently accepting it — defer the actual verification to citation_accuracy / source_verification, this skill only flags.

Anti-Rationalization

• "The abstract covers it" — the abstract oversells by design; always check the results section for the actual numbers before summarizing.
• "Limitations section is boilerplate" — authors often bury the real caveat there; read it, don't skip it.
• "I'll just paraphrase the conclusion" — a summary without concrete results/numbers is not a summary, it's a restatement of marketing copy.

Evidence

A summary is complete when: problem/method/results/limitations are all present, at least one concrete result number is quoted, venue and year are recorded, and the paper is cited with a working Markdown link.

Red Flags

• Summary with no numbers, only qualitative claims
• Limitations section omitted entirely
• Venue/year not recorded (can't tell preprint from published work)