The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script scripts/mp-api.py to perform all API interactions.
[REMOTE_CODE_EXECUTION]: The skill provides access to the /api/v1/plugin/install endpoint, which allows the agent to install plugins from external repository URLs on the target MoviePilot server. This represents a significant capability for remote code execution.
[DATA_EXFILTRATION]: The skill exposes endpoints that return sensitive information, including /api/v1/system/env (system configuration and environment variables), /api/v1/system/logging (application logs), and /api/v1/user/ (user account details).
[CREDENTIALS_UNSAFE]: The script scripts/mp-api.py explicitly disables SSL/TLS certificate verification by setting ssl.CERT_NONE and check_hostname = False. This makes the connection vulnerable to Man-in-the-Middle (MITM) attacks, potentially allowing an attacker to intercept the API key or other sensitive data transmitted to the MoviePilot host.
[EXTERNAL_DOWNLOADS]: The /api/v1/storage/download endpoint allows the agent to trigger file downloads on the target system via the API.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8).
Ingestion points: Data retrieved from the API (e.g., torrent titles, media metadata, log contents) enters the agent's context.
Boundary markers: The instructions do not define boundary markers or safety instructions for handling data returned from the API.
Capability inventory: The script scripts/mp-api.py can perform destructive actions (delete files), modify system state (update environment, create users), and install code (plugins).
Sanitization: There is no evidence of sanitization or validation of data retrieved from external sources before it is used in subsequent API calls.

moviepilot-api