The Agent Skills Directory

[DATA_EXFILTRATION]: The skill performs outbound HTTP requests via curl and Python's urllib.request to non-whitelisted domains (dblp.org and doi.org) to fetch and validate paper citation data.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection vulnerability surface because it processes external, untrusted content.
Ingestion points: Reads text from PAPER_PLAN.md, NARRATIVE_REPORT.md, and external files or URLs provided via the — style-ref argument.
Boundary markers: Absent. The skill does not instruct the agent to enclose untrusted content within strict boundary tags or include specific warnings to disregard nested instructions.
Capability inventory: The skill maintains powerful capabilities including arbitrary shell execution via Bash(*) and file modification tools (Write, Edit).
Sanitization: Absent. There is no pre-processing or sanitization phase to strip potentially malicious syntax or instructions embedded within user-supplied text or draft repositories.

paper-write