The Agent Skills Directory

[CREDENTIALS_UNSAFE] (HIGH): The skill documentation instructs the agent to run scripts/search setup <api-key> after collecting a key from the user. Passing sensitive credentials as command-line arguments is a major security risk as they can be leaked in shell history files, system logs, and process monitoring tools like ps or top.
[COMMAND_EXECUTION] (MEDIUM): The skill relies on local script execution for its core functionality. Specifically, the 'Script Path Resolution' section suggests using a complex find command to locate and execute scripts in a plugin cache. This behavior makes it harder to audit the exact code being executed and could lead to execution of unexpected files if the directory structure is manipulated.
[DATA_EXFILTRATION] (LOW): The skill transmits natural language queries and a user-provided API key to https://api.valyu.ai/v1. While this is documented as the intended behavior for the service, it constitutes an external data flow to a non-whitelisted third-party domain.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill processes untrusted external content from clinical trial databases.
Ingestion points: External data enters the context via the results array (specifically the content and title fields) from the Valyu API.
Boundary markers: Absent. There are no instructions to the agent to ignore or delimit embedded instructions within the trial data.
Capability inventory: The agent can execute the scripts/search Bash wrapper and the scripts/search.mjs Node.js script.
Sanitization: Absent. The documentation suggests piping output directly to jq or Python/TS SDKs without content filtering.

clinical-trials-search