The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes an ingestion surface by instructing the agent to fetch evidence from external sources such as official documentation, Reddit, and community reports in references/external.md and references/comparison.md. Mandatory evidence: 1) Ingestion points: External URLs and community sources (Reddit) are accessed for discovery. 2) Boundary markers: Absent; there are no instructions to use delimiters or ignore embedded commands. 3) Capability inventory: The skill can trace code behavior, access library contracts, and run local probes. 4) Sanitization: Absent; there is no instruction to escape or filter external content.
[COMMAND_EXECUTION]: The guidelines for tool comparison and troubleshooting in references/comparison.md and references/troubleshooting.md describe running 'bounded local probes' and 'reproducing the reported case at its actual failing boundary.' These activities involve executing commands or code to verify runtime behavior, which constitutes a security risk if the inputs or environments are not isolated.

research