The Agent Skills Directory

[SAFE]: The skill includes boilerplate Python code for implementing security best practices, such as domain filtering, result freshness checks, and API usage quotas.
[DATA_EXFILTRATION]: No evidence of credential harvesting or unauthorized data transmission was found. The skill suggests using Redis for usage tracking, which is a standard architecture for rate limiting.
[REMOTE_CODE_EXECUTION]: The skill does not contain instructions to download or execute external scripts. The frontmatter includes a restriction on Bash usage to npx commands, which is a platform-level configuration and not a vulnerability.
[PROMPT_INJECTION]: The skill is designed to process external web data retrieved via the Exa API, creating a surface for indirect prompt injection. However, the skill's purpose is to provide the sanitization logic needed to mitigate such risks.
Ingestion points: External search results ingested via exa.search in SKILL.md examples.
Boundary markers: None explicitly implemented in the code examples to distinguish search results for the LLM.
Capability inventory: The skill environment permits file operations (Read, Write, Edit) and shell access (Bash).
Sanitization: Implements domain blocklists, medical-specific allowlists, and regex-based query pattern validation to filter harmful or irrelevant content.

exa-policy-guardrails