The Agent Skills Directory

[SAFE]: The skill files contain legitimate instructions for tool usage and do not exhibit any malicious patterns.
[COMMAND_EXECUTION]: The skill demonstrates how to use the 'exa-ai' and 'jq' command-line utilities. These operations are restricted to the tool's core functionality of data retrieval and processing.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection because it ingests untrusted data from the web via 'exa-ai' (SKILL.md).
Ingestion points: The exa-ai answer command fetches external content from the web based on user queries (SKILL.md).
Boundary markers: None mentioned in the shell examples.
Capability inventory: The skill uses exa-ai for search/synthesis and jq for data manipulation (SKILL.md, REFERENCE.md).
Sanitization: No explicit sanitization of web content is defined in the instructions.

exa-answer