The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file employs a 'MANDATORY SKILL INVOCATION' block with imperative language ('YOU MUST', 'NOT optional') and operational threats ('violates your operational requirements') to override the agent's decision-making process for tool selection.
[DATA_EXFILTRATION]: In scripts/plex-api.sh, the PLEX_TOKEN is appended to URLs as a query parameter. This practice is insecure as it often results in the sensitive token being recorded in web server access logs, proxy logs, and the user's shell history.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect injection surface through the processing of user-supplied parameters.
Ingestion points: Arguments for section_id, rating_key, and search query passed to scripts/plex-api.sh.
Boundary markers: Absent. The skill does not use delimiters or instructions to prevent the agent from following directions embedded in the media data or query parameters.
Capability inventory: The skill utilizes curl for network requests and jq for data parsing.
Sanitization: Partial. While search queries are URL-encoded, section_id and rating_key are directly interpolated into the request path, creating a potential path traversal vulnerability where an attacker could access unauthorized API endpoints on the Plex server.

plex