The Agent Skills Directory

[DYNAMIC_EXECUTION]: The SDK includes get_shared_state_pickle and set_shared_state_pickle, which use the Python pickle module for state persistence. Pickle deserialization is inherently unsafe and can lead to arbitrary code execution if the stored state is tampered with.
[DYNAMIC_EXECUTION]: The wmill.run_inline_script_preview function allows for the execution of arbitrary script content provided as a string, which can bypass standard script deployment checks.
[CREDENTIALS_UNSAFE]: The wmill.get_id_token method facilitates the retrieval of OIDC JWT tokens for external service authentication. Additionally, wmill.get_variable and wmill.get_resource provide access to potentially sensitive workspace configuration and secrets.
[COMMAND_EXECUTION]: The skill instructs the agent to use wmill CLI commands such as script preview and script run, which involve the execution of script code on the local environment or Windmill workers.
[INDIRECT_PROMPT_INJECTION]: The skill implements handlers for external data sources, including preprocessor functions for webhooks and S3Object processing. This creates a surface where malicious input could influence the agent's behavior.
Ingestion points: The event object in preprocessor functions and S3 files loaded via wmill.load_s3_file (SKILL.md).
Boundary markers: None specified in the instructions for handling these data inputs.
Capability inventory: The SDK provides broad capabilities including network access (via httpx), S3 operations, and the ability to execute other scripts or flows.
Sanitization: The instructions do not define sanitization or validation requirements for external content before processing.

write-script-python3