The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests untrusted data from external sources.
Ingestion points: Web search results and social media posts fetched via the X.AI API (SKILL.md).
Boundary markers: The prompt attempts to constrain the output to a numbered list with specific fields, but does not explicitly warn the model to ignore instructions embedded in the search results.
Capability inventory: The skill can write to the file system (memory/logs/) and send system notifications (./notify).
Sanitization: No explicit escaping or filtering of external content is performed before processing.
[COMMAND_EXECUTION]: The skill executes shell commands to interact with external APIs and local tools.
Evidence: Uses ./secretcurl to send JSON payloads to https://api.x.ai/v1/responses for X/Twitter searches.
Evidence: Uses the gh CLI tool to fetch repository metrics (stars, forks) from GitHub.

mention-radar