The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill compiles project source code into shim assemblies and loads them into the running application domain at runtime. It utilizes Harmony transpilers to replace original method bodies with these dynamically generated shims.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface as it processes C# source files provided by the user or identified in the project. If these files are compromised, the skill's code-patching capabilities could be used to execute malicious logic.
Ingestion points: Local project .cs files specified via the --files parameter or automatically detected through compilation snapshots.
Boundary markers: None present; the skill treats the content of source files as trusted code to be compiled and executed.
Capability inventory: Performs assembly loading, Harmony method patching, and supports adding new fields, methods, and types to the running domain.
Sanitization: The skill uses a Roslyn worker to rewrite code into static shims, but this is a transformation for execution compatibility rather than a security filter.
[COMMAND_EXECUTION]: The mechanism involves invoking an out-of-process Roslyn worker to perform the rewriting and compilation of edited project files.

uloop-hot-reload