The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is configured to ingest and process information from external, untrusted platforms which could be manipulated to include hidden instructions targeting the AI agent's behavior.
Ingestion points: In references/newsjacking.md, the skill fetches external data from Google News RSS feeds, Hacker News (via the Algolia API), and Reddit JSON endpoints.
Boundary markers: The skill does not include specific delimiters or instructions to the agent to disregard potentially malicious commands embedded within the retrieved news content.
Capability inventory: The skill uses shell commands (curl, jq, xmllint) and browser automation tools (dev-browser) to gather and analyze this external content.
Sanitization: There is no documentation of sanitization or input validation performed on the fetched external strings before they are presented to the agent for strategy development.
[EXTERNAL_DOWNLOADS]: The skill initiates network requests to retrieve real-time news data.
Evidence: references/newsjacking.md contains curl commands targeting news.google.com, hn.algolia.com, and reddit.com. These are well-known services used for legitimate news monitoring purposes.

public-relations