The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted bibliographic data from user-supplied files. While the verification logic is script-based, the agent ultimately evaluates the output, creating a potential surface for indirect influence via maliciously crafted citation metadata.
Ingestion points: Processes *.bib citation files and plain-text thesis/abstract files (--thesis-file).
Boundary markers: The skill provides a structured triage guide (references/triage-guide.md) and a specific fix workflow to constrain agent interpretation of results.
Capability inventory: The scripts/check_bibtex.py script performs network lookups, writes to a local cache directory, and the agent executes this script via shell commands.
Sanitization: The verification script normalizes text using norm_text (stripping LaTeX commands and math) prior to matching, though the final metadata is presented to the agent for triage.
[COMMAND_EXECUTION]: The skill executes a local Python utility (scripts/check_bibtex.py) to perform citation validation. This is a functional requirement for the skill's purpose.
[EXTERNAL_DOWNLOADS]: The verification script fetches metadata and resolution information from well-known scholarly services including Crossref, DBLP, arXiv, DataCite, and Semantic Scholar. These connections are intended for bibliographic verification and target established research infrastructure.

verify-citations