The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/setup.sh script is designed to download a twozero.tox binary from https://www.404zero.com/pisang/twozero.tox. This resource is hosted on a third-party domain that does not match the provided vendor context.
[COMMAND_EXECUTION]: The skill instructs users to run an automated setup.sh script via bash. This script executes system commands (such as pgrep, nc, and curl) and utilizes a Python subprocess to programmatically modify the agent's config.yaml file, injecting a new MCP server entry into the configuration.
[INDIRECT_PROMPT_INJECTION]: The skill is heavily focused on ingesting data from external sources and using it to influence application behavior, creating a vulnerability surface for indirect injection.
Ingestion points: As described in references/external-data.md and references/midi-osc.md, the skill reads data from HTTP APIs, WebSockets, MQTT brokers, Serial devices, and OSC network messages.
Boundary markers: The skill body lacks specific delimiters or "ignore instructions" warnings for data being passed from external inputs into TouchDesigner parameters.
Capability inventory: The td_execute_python tool, as documented in references/mcp-tools.md, provides full access to the TouchDesigner Python API and the local filesystem as the process user.
Sanitization: While the documentation mentions basic JSON error handling, it does not provide comprehensive sanitization or escaping mechanisms to prevent malicious data from manipulating the Python execution context.
[PRIVILEGE_ESCALATION]: The references/external-data.md file advises users to use sudo chmod 666 on macOS to modify device permissions for serial port access, which is a significant privilege modification.

touchdesigner-mcp