The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documentation describes actions such as run_command and run_script within JSON configurations. Examples demonstrate the execution of shell commands like ls -l and rsync on the host system.
[REMOTE_CODE_EXECUTION]: The workflow engine can be triggered via external webhooks and can execute scripts defined in configurations loaded from local paths or remote APIs. This allows for the execution of code provided at runtime.
[DATA_EXFILTRATION]: The combination of system command execution and built-in support for network-based actions (such as HTTP webhooks and email) provides a mechanism for reading sensitive data and sending it to external destinations.
[PROMPT_INJECTION]: The skill processes untrusted workflow configurations that could contain malicious instructions. Ingestion points: Configuration files (--config) and API responses (POST /api/workflows). Boundary markers: None present; the skill directly executes command and action fields from the JSON input. Capability inventory: Subprocess command execution, script execution, and network operations (webhooks, email). Sanitization: No evidence of input validation or command escaping for values defined in configurations.

clawflows