The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes shell command snippets in LISTENING.md for interacting with public APIs. These are used for data retrieval, such as fetching posts and comments, and do not involve executing untrusted scripts or escalated privileges.
[EXTERNAL_DOWNLOADS]: Retrieves data from established and well-known services including Algolia's Hacker News API, Bluesky's public API, and YouTube's RSS feeds. These operations are limited to data ingestion for social listening and monitoring purposes and follow safe retrieval practices.
[INDIRECT_PROMPT_INJECTION]: The skill recognizes that content fetched from social media platforms is untrusted. It includes explicit safety instructions in SKILL.md and LISTENING.md to prevent the agent from following instructions embedded in post text, bios, or comments.
Ingestion points: LISTENING.md specifies workflows for fetching data from Reddit, Hacker News, Bluesky, RSS feeds, and X.
Boundary markers: The skill uses explicit warnings and "Prove it" checks to ensure that fetched data is treated as untrusted and that no actions are taken without explicit user approval.
Capability inventory: No dynamic execution or writing of external data to executable paths was detected; the agent's role is restricted to scoring data and drafting responses for human review.
Sanitization: The instructions mandate that the agent ignore any potential prompt injections found within the external content and treat all fetched strings as untrusted data.

social-media