The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection via the processing of untrusted TouchDesigner network data.
Ingestion points: Untrusted data enters the agent context through tools like td_operators, td_pane, and td_selection, which inspect the state of the TouchDesigner network. Specifically, content from DAT operators (e.g., textDAT.text) is read and processed.
Boundary markers: The instructions lack explicit delimiters or warnings to ignore embedded instructions within ingested data.
Capability inventory: The skill possesses significant capabilities, including arbitrary Python execution within the application via td_execute and the ability to write files to the local system using op.save().
Sanitization: No sanitization, escaping, or validation logic is provided for external content before it is used to influence agent actions or code generation.
[DYNAMIC_EXECUTION]: The skill is designed to dynamically generate Python scripts and GLSL shaders at runtime. These scripts are intended for execution within the TouchDesigner process using the td_execute MCP tool. While this is the intended function of the developer tool, it represents a risk factor for executing logic derived from potentially untrusted inputs.

touchdesigner