The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use the wmill CLI for several operations: metadata generation (wmill generate-metadata), script execution (wmill script preview, wmill script run), and workspace synchronization (wmill sync push). These are official platform commands used for their intended purpose within the Windmill development workflow.
[REMOTE_CODE_EXECUTION]: The skill facilitates the execution of TypeScript/Bun code using wmill script preview. This is a core feature of the platform designed to allow developers to test scripts locally before deployment. The instructions emphasize that this executes code and should be done when testing is intended.
[DATA_EXFILTRATION]: While the skill interacts with sensitive data via resources (e.g., RT.Stripe, wmill.getResource) and S3 storage (wmill.loadS3File), these operations are performed through the official windmill-client SDK. These interactions are standard for the platform's functionality and do not exhibit exfiltration patterns to unauthorized domains.
[INDIRECT_PROMPT_INJECTION]: The skill describes processing external data (e.g., webhook events in preprocessor scripts).
Ingestion points: Event bodies and headers are ingested in preprocessor functions.
Boundary markers: None explicitly defined in the provided code templates.
Capability inventory: The agent can execute scripts and system commands via the wmill CLI.
Sanitization: Standard TypeScript typing is used, but the skill does not explicitly detail sanitization of the input data. However, as this is a developer-centric coding tool, this surface is considered a standard functional risk rather than a malicious injection.

write-script-bun