The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements a social listening feature that processes untrusted data from external platforms, creating a surface for indirect prompt injection attacks.
Ingestion points: Data is retrieved from external APIs and RSS feeds as defined in references/listening.md.
Boundary markers: The skill does not provide specific instructions to the agent to treat fetched data as untrusted or to use delimiters to isolate it from primary instructions.
Capability inventory: The skill has the capability to run shell commands (curl, jq, xmllint) and use a browser to retrieve data, and generates drafts for social media responses.
Sanitization: No sanitization of the retrieved external content is specified before processing by the agent.
[COMMAND_EXECUTION]: The skill provides functional shell command recipes for fetching and processing web data.
Evidence: references/listening.md contains documented bash commands using curl, jq, and xmllint to fetch and parse JSON or XML responses from various social platforms.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to fetch data from several well-known external web services.
Evidence: Network connections are initiated to reddit.com, hn.algolia.com (Hacker News), and public.api.bsky.app (Bluesky) to support the listening functionality.

social