The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches and processes untrusted content from external social media platforms (Reddit, Hacker News, Bluesky) to summarize trends and draft engagement responses. This workflow creates an attack surface for indirect prompt injection, where malicious instructions hidden in social media posts could attempt to manipulate the agent's behavior.
Ingestion points: references/listening.md provides recipes for retrieving data from external APIs.
Boundary markers: The skill does not instruct the agent to use specific delimiters or ignore potentially malicious commands within the external data during processing.
Capability inventory: The skill is designed to generate social media drafts and interact with platforms using the dev-browser tool.
Sanitization: There are no explicit instructions for sanitizing or escaping the content retrieved from external sources before it is analyzed by the agent.
[EXTERNAL_DOWNLOADS]: The skill includes instructions to fetch data from well-known and reputable public services for social listening purposes.
Evidence: references/listening.md provides curl recipes for accessing JSON and RSS feeds from reddit.com, hn.algolia.com (Hacker News), public.api.bsky.app, and youtube.com.

social