The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill demonstrates workflows where the agent fetches document data from a CouchDB instance and processes it or interpolates it into subsequent commands. This creates an attack surface where malicious content stored in the database could influence the agent's actions.
Ingestion points: Retrieval of JSON documents via curl commands as shown in the SKILL.md examples.
Capability inventory: The skill utilizes curl for network requests and data mutation, and jq for parsing database output within a shell environment.
Boundary markers: There are no explicit instructions or delimiters provided to treat the database content as untrusted or to ignore embedded instructions.
Sanitization: The skill does not provide methods for sanitizing document data before it is used in logic or interpolated into shell commands (e.g., the $REV variable in the conflict resolution example).

couchdb