The Agent Skills Directory

[COMMAND_EXECUTION]: The file references/listening.md provides specific shell command recipes using curl and jq to fetch and filter data from platforms like Reddit and Hacker News. These commands are intended for the agent to execute in the terminal environment.
[EXTERNAL_DOWNLOADS]: The skill fetches data from several well-known external sources through their public APIs and RSS feeds, including Reddit, Algolia (Hacker News), Bluesky, and YouTube. These are used to provide the 'social listening' functionality.
[PROMPT_INJECTION]: The 'Social Listening' workflow in references/listening.md involves ingesting untrusted data from external social platforms and instructing the agent to score it and draft comments based on that content. This creates a surface for indirect prompt injection, as there are no instructions for sanitizing this input or using boundary markers to prevent the agent from following instructions embedded within the external data.
Ingestion points: Data is fetched from the Reddit JSON API, Hacker News (Algolia) API, Bluesky API, and various RSS feeds.
Boundary markers: None provided in the instructions to distinguish external content from agent instructions.
Capability inventory: The agent is authorized to run bash commands (curl, jq) and use the dev-browser tool to interact with web pages.
Sanitization: No sanitization or content filtering logic is defined for the ingested data.

social