The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted bibliography data (RIS, BibTeX, CSL-JSON) provided by the user, which presents a surface for indirect prompt injection. 1. Ingestion points: The script scripts/format_bibliography.py reads user-provided text or files via the --input argument or STDIN. 2. Boundary markers: No explicit delimiters or instructions are provided to the agent to disregard embedded instructions within the processed citation data. 3. Capability inventory: The skill can write results to local files via the --output parameter and perform network operations using requests.get. 4. Sanitization: The script uses structural parsers (rispy, bibtexparser, citeproc-py), but there is no specific sanitization to prevent the agent from following instructions hidden within bibliography fields.
[EXTERNAL_DOWNLOADS]: The skill automatically fetches Citation Style Language (CSL) files from Zotero's public style repository. 1. The script scripts/format_bibliography.py performs network GET requests to https://www.zotero.org/styles/ and https://www.zotero.org/styles-files/styles.json. 2. This is a functional requirement for journal-specific formatting and targets Zotero, a well-known service in the academic community. The behavior is documented and expected for the skill's utility.

bib-formatter