The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from an external server that could potentially contain malicious instructions intended to influence the agent's behavior.
Ingestion points: The responses command in scripts/pages-axi.mjs retrieves data from the /api/pages/:slug/responses endpoint.
Boundary markers: The data is rendered into the agent's context without explicit delimiters or instructions to ignore embedded prompts.
Capability inventory: The skill has the ability to read local files and perform network operations.
Sanitization: No sanitization or validation of the response payload content is performed before it is displayed to the agent.
[DATA_EXFILTRATION]: The skill transmits the content of local files to a networked server, which could be a security concern if the server is misconfigured or malicious.
Evidence: The publish and publish-worksheet commands in scripts/pages-axi.mjs read local files and send their contents to the server defined by the CLAUDE_ASSIST_SERVER environment variable (defaulting to localhost).

assist-pages