The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied search queries that are interpolated into API calls and used in logic branching, creating a surface for potential instruction injection through data.
Ingestion points: The query parameter is ingested in various functions throughout SKILL.md, including costAwareSearch, cachedSearch, deduplicateQueries, and selectCostEffectiveType.
Boundary markers: Absent. The skill does not implement delimiters or explicit instructions for the agent to ignore embedded commands within the search query.
Capability inventory: The skill uses network operations via the exa-js library and performs shell command execution using curl to interact with api.exa.ai (documented in SKILL.md).
Sanitization: Basic regex matching is used in selectCostEffectiveType to determine search type, but there is no escaping or sanitization of external content before interpolation into API calls.
[DATA_EXFILTRATION]: The skill performs network requests to the official Exa API usage endpoint to monitor search limits and budget status.
Evidence: The shell command curl -s https://api.exa.ai/v1/usage -H "x-api-key: $EXA_API_KEY" in SKILL.md interacts with the service's management endpoint.

exa-cost-tuning