The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from external, untrusted RSS and Atom feeds, creating a surface for indirect prompt injection attacks where malicious instructions could be hidden within news items or summaries.
Ingestion points: The agent fetches XML content from external URLs provided by the user or discovered on websites using the curl tool (SKILL.md).
Boundary markers: No specific delimiters, framing, or explicit "ignore embedded instructions" warnings are provided to the agent to separate retrieved feed data from its core instructions.
Capability inventory: The skill uses curl to perform network requests and retrieve data from the internet (SKILL.md).
Sanitization: The instructions recommend stripping HTML from item descriptions and decoding entities, which helps mitigate technical payloads but does not prevent natural language instruction overrides (SKILL.md).
[COMMAND_EXECUTION]: The skill utilizes the curl command-line utility to retrieve feed data.
Evidence: The instructions direct the agent to execute curl -sL with user-supplied or discovered URLs to fetch XML feeds (SKILL.md).
[EXTERNAL_DOWNLOADS]: The skill performs network operations to retrieve content from external domains for processing.
Evidence: The skill fetches data from sources such as hnrss.org and other feed URLs provided during execution (SKILL.md).

rss-digest