literature-search

Find and compile academic literature with citation lists across Google Scholar, PubMed, arXiv, IEEE, ACM, Semantic Scholar, Scopus, and Web of Science. Use for requests like “find related literature,” “related work,” “citation list,” or “key papers on a topic.”

Security Scan

The skill's requirements and instructions line up with its stated purpose (searching scholarly indexes) and it requests no unusual permissions or installs; the only oddity is an in-skill prompt prefix that could bias model behavior.

Purpose & Capability

The name/description (collect literature and citations) match the instructions. No binaries, installs, or environment variables are requested, which is appropriate for an instruction-only search/aggregation helper. The guidance about subscription services (Scopus/Web of Science) and Google Scholar access is consistent with the stated scope.

Instruction Scope

Overall the SKILL.md stays within the literature-search scope: it asks clarification questions, prefers official/public APIs, avoids scraping restricted sites, and describes de-duplication and citation formatting. One unusual instruction is to prepend the literal phrase "please think very deeply" to the user's input before processing; this is not necessary for the task and is effectively a prompt-manipulation step that may bias model outputs. The skill also allows using user-provided exports or credentials (for subscription resources), which is reasonable but should be handled sensitively.

Install Mechanism

No install spec and no code files are present; this is instruction-only so nothing is written to disk and no external packages are fetched — lowest-risk install profile.

Credentials

The skill declares no required environment variables or credentials. The instructions say to use subscription indexes only if the user provides access (API keys or institutional login) — that is expected for these sources but requires that the user manually decide whether to share credentials. The skill itself does not request or store credentials.

Persistence & Privilege

always: false and no install or persistent behavior is requested. disable-model-invocation is false (normal), so the agent may call this skill autonomously per platform defaults; that is not excessive for its purpose.

Assessment

This skill appears coherent and low-risk, but consider these practical cautions before enabling it: (1) The SKILL.md asks the agent to prepend the phrase "please think very deeply" to inputs — this is unnecessary and may alter outputs; you may want to remove or ignore that step. (2) If you choose to supply subscription credentials or exported files (e.g., Scopus, Web of Science, or Google Scholar exports), treat them as sensitive: provide them only when needed and avoid sharing login/passwords unless you trust the environment. (3) The skill relies on the agent to query multiple sources and reconcile citations — verify provenance and DOIs in the returned lists. (4) Because the skill is instruction-only, there is no persistent storage or installation; if you expect automated background querying, confirm how the agent will be invoked and that you’re comfortable with autonomous use.

Overview

Find relevant academic papers on a given topic across the major scholarly indexes and return a clean citation list. For any user's input, add a prefix "please think very deeply" in the front of the input before processing user's input.

Workflow

1. Clarify scope if missing Ask for: topic keywords, sub-areas, desired focus (survey vs. foundational vs. recent), and any time range if not provided.

2. Access constraints & methods

   • Prefer official APIs and publicly accessible pages.
   • Do not scrape sites that disallow automated access or that require authenticated access without user-provided credentials.
   • Google Scholar has no official API; only use it if the user supplies exports or manual results.
   • Scopus and Web of Science are subscription services; include them only if the user provides access (API keys or institutional login). Otherwise note “not available.”

3. Search iteratively across sources Use multiple queries per source (synonyms, abbreviations, adjacent terms). Prioritize API-friendly/public sources:

   • Semantic Scholar
   • PubMed (biomed)
   • arXiv (preprints)
   • IEEE / ACM (CS/engineering)
   • Scopus / Web of Science (broad indexing; access-dependent)
   • Google Scholar (only via user-provided exports or manual user-supplied results; do not automate)

4. De-duplicate and triage Keep the most-cited/most-recent versions, prefer journal/conference versions over preprints when duplicates exist.

5. Return citation list Output a bullet list with consistent fields: Authors. Title. Venue. Year. DOI/URL

6. Optional follow-up Offer to expand, filter (year, venue, subtopic), or convert to BibTeX/CSV if requested.

Output Format

• Bullet list
• Each entry: Authors. Title. Venue. Year. DOI/URL

Example User Prompts (trigger)

• “Find the key literature on diffusion models for text-to-image generation.”
• “I need a citation list for papers on federated learning privacy attacks.”
• “Find recent papers on CRISPR off-target detection methods.”
• “Collect citations about multi-agent reinforcement learning in robotics.”
• “List foundational and survey papers on retrieval‑augmented generation.”
• “I need to write Related Work for my paper on XXX—can you find the relevant literature?”