The Agent Skills Directory

[COMMAND_EXECUTION]: The download action in scripts/run_livedoc.mjs writes files to the local system using a filename extracted from the Content-Disposition header of the API response. The script does not perform path sanitization, which could allow a malicious or compromised API response to use path traversal sequences (e.g., ../../) to overwrite arbitrary files on the user's system.
[DATA_EXFILTRATION]: The upload action in scripts/run_livedoc.mjs reads local files based on paths provided to the tool and transmits them to openapi.felo.ai. While this is core functionality, it creates a risk where the agent could be manipulated into reading and exfiltrating sensitive local files if prompted by malicious input.
[PROMPT_INJECTION]: The skill ingests and processes data from an external API, including document snippets and search results, which are then presented to the agent. This creates a surface for indirect prompt injection.
Ingestion points: Data returned by the retrieve, content, and get-readme actions in scripts/run_livedoc.mjs.
Boundary markers: Output from the API is not enclosed in delimiters or accompanied by instructions for the agent to ignore embedded commands.
Capability inventory: The skill possesses file read, file write, and network communication capabilities.
Sanitization: No validation, filtering, or escaping is performed on the content received from the remote API before it is rendered into the agent's context.

felo-livedoc