The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from external news platforms and podcast metadata, which could be used to influence agent behavior.
Ingestion points: Fetches content from Google News, Hacker News, and Reddit in references/newsjacking.md, and RSS feeds in references/podcast-guest-prep.md.
Boundary markers: The instructions do not define clear delimiters or instruct the agent to ignore instructions within the retrieved data.
Capability inventory: The skill utilizes curl, xmllint, jq, and dev-browser to interact with this data.
Sanitization: No sanitization or validation logic is applied to the external content before processing.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to retrieve content from well-known news and information services.
Evidence: references/newsjacking.md includes curl requests to news.google.com, hn.algolia.com, and reddit.com.
Evidence: references/podcast-guest-prep.md uses RSS feeds and web scraping to gather show information.
[COMMAND_EXECUTION]: The skill uses shell commands for data retrieval and transformation.
Evidence: references/newsjacking.md provides templates for using curl piped into xmllint and jq for parsing RSS and JSON data.

public-relations