The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several shell commands to facilitate literature search and metadata validation. This includes the scholar CLI tool for querying databases and the scripts/check_provenance.py script for linting BibTeX files. These commands are integral to the skill's primary function.
[EXTERNAL_DOWNLOADS]: The skill facilitates the retrieval of academic metadata and full-text PDFs from well-known scholarly services such as Semantic Scholar, OpenAlex, DBLP, arXiv, and Crossref. It also uses WebFetch to read content from arbitrary publisher domains for verification purposes.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external academic search results, abstracts, and PDF files. While a malicious document could theoretically contain instructions to influence the agent, the skill implements a rigorous verification protocol to mitigate this risk.
Ingestion points: Data is ingested via scholar search, scholar rq, scholar pdf open, and WebFetch operations.
Boundary markers: The skill uses structured provenance comment blocks and BibTeX entry formats to separate citation metadata from other content.
Capability inventory: The agent has the ability to execute shell commands (scholar, scripts/check_provenance.py), perform file I/O on .bib and .md files, and conduct network operations via the search tools.
Sanitization: The skill requires a manual verification step where the agent must confirm that a verbatim quote from the source entails the claim, guided by the references/verification-checklist.md.

backing-claims