The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill integrates external search results from the web via the Exa API, creating a surface for untrusted data to potentially influence agent behavior through indirect prompt injection.
Ingestion points: The skill ingests external content via the exa.searchAndContents and exa.search methods as described in SKILL.md.
Boundary markers: The code examples provided do not include specific prompt delimiters or markers to isolate the retrieved web content from the agent's instructions.
Capability inventory: The skill is granted Read, Write, and Edit tool permissions in its configuration, which could be leveraged if malicious data successfully influences the agent.
Sanitization: The instructions provide procedural mitigation by explicitly warning developers not to 'turn a failed retrieval into an unreviewed automated decision' and recommending the use of 'sanitized test inputs' in SKILL.md.
[EXTERNAL_DOWNLOADS]: The skill references official API documentation from Exa and technical articles on the circuit breaker pattern from well-known sources like Martin Fowler. These are informational references for integration guidance and do not involve the execution of remote scripts.

exa-reliability-patterns