The Agent Skills Directory

[COMMAND_EXECUTION] (MEDIUM): The skill instructs the agent to execute curl commands in bash using interpolated values from user input (<VIDEO_ID>) and API responses (<url>). This pattern lacks explicit sanitization or validation instructions, creating a potential risk for command injection if the agent processes malformed or malicious strings.
[EXTERNAL_DOWNLOADS] (LOW): The skill makes multiple network requests to fetch video metadata and transcript files from loom.com and other URLs provided by the API. This involves downloading and processing untrusted external content.
[PROMPT_INJECTION] (LOW): The skill is vulnerable to indirect prompt injection. Ingestion points: video transcripts downloaded from external URLs (Step 4). Boundary markers: None provided in the presentation logic (Step 5). Capability inventory: Shell execution capabilities (bash, curl). Sanitization: None. The agent is directed to present the full transcript text, which could contain hidden instructions designed to hijack the agent's session or influence its behavior.

loom-transcript