The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the cli-anything-renderdoc command to perform graphics analysis. These operations involve reading local .rdc (RenderDoc capture) files and writing extracted data such as textures (PNG/JPG/DDS), shaders (HLSL/GLSL), and analysis results (JSON) to the local file system. These actions are within the expected scope of a graphics debugging tool.
[EXTERNAL_DOWNLOADS]: The skill is designed to be installed via pip install cli-anything-renderdoc. The package name matches the author's namespace and intended functionality. It also requires the renderdoc Python bindings, which are typically installed alongside the RenderDoc graphics debugger application.
[DATA_EXFILTRATION]: No network operations or unauthorized data transmission patterns were detected. The skill only interacts with local capture files and environment variables (RENDERDOC_CAPTURE, PYTHONPATH) necessary for its operation.
[INDIRECT_PROMPT_INJECTION]: The skill processes external graphics capture files (.rdc). While these files could contain attacker-controlled metadata (e.g., action names or marker labels), the risk of these strings influencing the agent's behavior is low given the specialized binary format and the primary use case for technical debugging.
Ingestion points: Reads .rdc capture files via the -c argument.
Boundary markers: Absent; the skill does not explicitly delimit data extracted from the capture when presenting it to the agent.
Capability inventory: File system read (captures), file system write (extracted assets and diffs), and execution of the local CLI harness.
Sanitization: Not explicitly implemented for the textual metadata extracted from the capture files.

cli-anything-renderdoc