The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves external documentation which is then processed by the agent. If the documentation source is untrusted or compromised, it could contain malicious instructions designed to manipulate the agent.
Ingestion points: Documentation content is fetched from a remote server via the scripts/query.mjs script.
Boundary markers: The instructions do not specify any delimiters or directives for the agent to treat the retrieved content as untrusted data.
Capability inventory: The agent is authorized to execute shell commands to run the retrieval script, which has network access.
Sanitization: The script outputs the raw response body from the server directly to stdout without filtering or sanitization.
[COMMAND_EXECUTION]: The skill requires the agent to execute a local Node.js script (scripts/query.mjs) to perform library discovery and document retrieval.
[DATA_EXFILTRATION]: The skill performs network operations using the fetch API to an external URL defined by the DOCS_SERVER_URL environment variable, which is not restricted to a whitelisted domain.

docs-search