The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes a social listening workflow that ingests content from external sources, which introduces an attack surface for indirect prompt injection where malicious instructions in social media posts could attempt to influence the agent.
Ingestion points: Untrusted content is fetched from Reddit, Hacker News (via Algolia), and Bluesky using curl recipes and dev-browser scans described in references/listening.md.
Boundary markers: The skill does not explicitly define delimiters or "ignore embedded instructions" warnings for the agent to use when processing the fetched content.
Capability inventory: The skill is restricted to network data retrieval and browser navigation; it lacks capabilities for arbitrary file writes, privileged command execution, or dynamic code evaluation.
Sanitization: No explicit sanitization, validation, or filtering of the ingested external content is specified in the instructions.
[EXTERNAL_DOWNLOADS]: The references/listening.md file provides curl commands to fetch JSON data from public endpoints of well-known platforms including Reddit, Algolia, and Bluesky. These are standard, legitimate services used for their intended purpose within the social listening functionality.

social