The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute several shell commands using the wmill CLI, including wmill script preview, wmill script run, wmill generate-metadata, and wmill sync push. These commands interact with the local filesystem and the Windmill backend.
[REMOTE_CODE_EXECUTION]: The wmill script preview command is explicitly designed to execute C# code locally that the agent has just written or edited. This provides a direct path for the agent to run code it generates, which could lead to unintended side effects if the generated code is malicious or flawed.
[EXTERNAL_DOWNLOADS]: The skill instructs the agent to use the #r directive to include NuGet packages (e.g., Newtonsoft.Json, RestSharp). Additionally, the wmill generate-metadata command re-resolves and downloads dependencies to update lock files. These operations fetch external code from the public NuGet registry.
[PROMPT_INJECTION]: The skill defines a pattern for C# scripts to ingest data via parameters in the Main method (e.g., Main(string url, int count)). It lacks instructions for sanitizing these inputs or implementing boundary markers, creating a vulnerability surface for indirect prompt injection if the script processes untrusted data.
Ingestion points: Arguments passed to the Main method of the C# scripts (file: SKILL.md).
Boundary markers: None specified in the instructions or example code.
Capability inventory: File system access and network operations via RestSharp or standard libraries, coupled with local execution via wmill script preview (file: SKILL.md).
Sanitization: No mention of input validation or sanitization for the parameters received by the scripts.

write-script-csharp