The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides predefined shell command templates in references/listening.md that use curl, jq, and xmllint to fetch and parse JSON/RSS data from well-known public services such as Reddit, Hacker News, and Bluesky. While these commands target legitimate public APIs, they involve the execution of subprocesses to handle external data.
[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection (Category 8) due to its core 'social listening' functionality which processes untrusted external content.
Ingestion points: Data enters the agent's context through public JSON endpoints (Reddit, HN, Bluesky) and via automated browser navigation (dev-browser) to LinkedIn and X feeds, as described in references/listening.md.
Boundary markers: There are no explicit instructions or delimiters defined to help the agent distinguish between its own system instructions and the potentially adversarial text content fetched from social media posts.
Capability inventory: The skill possesses capabilities including shell command execution (curl), browser automation (dev-browser), and file system access (reading and writing to .agents/listening-sources.md).
Sanitization: Content is parsed for specific fields using jq, but no further sanitization, escaping, or validation of the text content is performed before the agent processes and scores it.

social