The Agent Skills Directory

[DATA_EXFILTRATION]: The skill reads user pairing data from a local file located at ~/.tinyclaw/pairing.json to identify message recipients. This is a standard operation within the intended framework.
[DATA_EXFILTRATION]: The script send_message.ts allows the agent to specify absolute file paths via the --files argument, which are then passed to a local API server. While this allows referencing any file on the system, the script itself does not read the file contents, and the destination is restricted to localhost.
[COMMAND_EXECUTION]: The skill uses npx ts-node within send-message.sh to execute its TypeScript-based logic, which is a common pattern for executing scripts in this environment.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection because it accepts unvalidated message content and file paths as input from the agent. Ingestion points: The message and files arguments processed in scripts/send_message.ts. Boundary markers: None are present to distinguish between agent instructions and user-provided data. Capability inventory: The skill can send network requests to a local API and reference any file path on disk. Sanitization: No sanitization or path validation is performed on the provided inputs.

send-user-message