The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/setup.sh file downloads a TouchDesigner component (twozero.tox) from an external domain https://www.404zero.com/pisang/twozero.tox. This domain is not identified as a trusted vendor or well-known service.
[REMOTE_CODE_EXECUTION]: The td_execute_python tool allows for the execution of arbitrary Python code within the TouchDesigner environment. The documentation explicitly states it has unrestricted access to the filesystem and TD Python API.
[COMMAND_EXECUTION]: The td_input_execute tool allows the agent to simulate mouse and keyboard inputs on the host system. This capability, combined with arbitrary script execution, can be used to perform actions outside the intended scope of creative coding.
[DATA_EXFILTRATION]: The skill provides tools like td_read_dat, td_read_chop, and td_get_screenshot which can access internal project data and system state. Combined with the td_execute_python tool and network capabilities, these could be leveraged to exfiltrate sensitive information.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data (operator names, script content, and network metadata) from the TouchDesigner environment.
Ingestion points: td_read_dat (reads script code), td_get_network (reads operator names/types), and td_get_operator_info (reads comments and parameters).
Boundary markers: Absent. The skill does not use delimiters or instructions to ignore embedded commands in the data it reads.
Capability inventory: Includes td_execute_python (RCE), td_write_dat (script modification), and td_input_execute (system input simulation).
Sanitization: Absent. Data fetched from the TouchDesigner instance is interpolated directly into the context without validation.

touchdesigner-mcp