The Agent Skills Directory

CREDENTIALS_UNSAFE (HIGH): The skill instructions in SKILL.md include a command that attempts to scrape an API key from a specific local path: /Users/lingzhi/Code/keys.md. This hardcoded path targets a specific user environment and risks exposure of sensitive information. Evidence: $(grep S2_API_Key /Users/lingzhi/Code/keys.md ...).
COMMAND_EXECUTION (MEDIUM): The skill uses shell command substitution ($(...)) to execute a pipeline of commands including grep, cut, and tr on the host system.
DATA_EXFILTRATION (LOW): The harvest_citations.py script reads local .tex and .bib files and transmits search queries derived from their content to the Semantic Scholar API.
PROMPT_INJECTION (LOW): The skill possesses an indirect prompt injection surface as it ingests untrusted LaTeX data and uses it in network operations. 1. Ingestion points: harvest_citations.py (via --tex and --bib arguments). 2. Boundary markers: Absent. 3. Capability inventory: Network requests via urllib.request. 4. Sanitization: Absent; content extracted from files is used directly in URL-encoded search queries.

citation-management