The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the web via web_search and curl to generate news summaries. A malicious article could contain hidden instructions that influence the agent's behavior during processing.
Ingestion points: SKILL.md (Step 4, Step 9), references/pbsa-hk-search-config.md (Search Methods), references/selfhosting-digest-config.md (Must-Track RSS Sources).
Boundary markers: The skill does not define clear delimiters or instructions to ignore embedded commands within the ingested article text.
Capability inventory: The skill uses web_search, terminal (for curl), and cronjob management.
Sanitization: There is no mention of sanitizing or filtering input text for potential prompt injection patterns.
[COMMAND_EXECUTION]: The skill makes extensive use of curl within the terminal to perform searches and fetch RSS data. It also references the execution of a Python script belonging to another skill.
curl is used for discovery across various search engines and RSS feeds in SKILL.md and configuration files.
references/pbsa-hk-search-config.md references a local script execution: python -u ~/.hermes/skills/google-news-api-skill/scripts/google_news_api.py.
[EXTERNAL_DOWNLOADS]: The skill fetches data from multiple external sources to discover news content.
Interacts with SearXNG instances and various RSS feeds including selfh.st, mariushosting.com, and noted.lol.
Uses Google News and Bing News discovery endpoints.

news-monitoring