The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests structured data from an external server that is populated by human users, creating a surface for indirect instructions.\n
Ingestion points: The responses command in scripts/pages-axi.mjs fetches data from the /api/pages/:slug/responses endpoint.\n
Boundary markers: Absent. The data is rendered as a table in the agent's context without delimiters or instructions to ignore embedded commands.\n
Capability inventory: The agent can execute the bundled CLI to publish pages, mark responses as processed, and interact with other tools in its environment.\n
Sanitization: The script parses JSON responses and uses a custom toonValue function to format them as strings for display, but does not filter or sanitize the content for malicious instructions.\n- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill provides the ability to read local file contents and transmit them to a remote server.\n
The publish and publish-worksheet commands read files from the local filesystem and send the contents via an HTTP POST request to the server specified by the CLAUDE_ASSIST_SERVER environment variable.\n
While intended for local or private use (defaulting to localhost), a misconfigured or malicious server address could lead to file data exfiltration.\n- [DYNAMIC_EXECUTION]: The skill relies on the agent executing a bundled Bash script (scripts/pages-axi) which acts as a wrapper for a Node.js CLI tool.

assist-pages