The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions describe a workflow that ingests external data for GPU processing, creating a potential attack surface for indirect prompt injection.
Ingestion points: The shader templates and functions ingest untrusted data from texture inputs (sTD2DInputs, sTD3DInputs) and particle attributes in the glslPOP context.
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded commands within the processed data buffers.
Capability inventory: The skill utilizes filesystem write operations via set_dat_content and can trigger node execution/cooking using Python snippets like null.cook(force=True).
Sanitization: No sanitization or validation of the ingested texture data or buffer contents is mentioned, as these are typically raw numerical/color values for GPU processing.
[DYNAMIC_EXECUTION]: The skill facilitates the runtime compilation and execution of GLSL source code.
Evidence: It provides instructions for writing shader code into DAT nodes and syncing them to disk for external editing. This code is then dynamically compiled by the host application's GPU driver at runtime. This is the primary and intended purpose of the skill.
[COMMAND_EXECUTION]: The skill documentation includes Python command snippets intended for the TouchDesigner environment.
Evidence: The glsladvancedPOP.md file contains logic for refreshing cache data using the command sequence null.lock = False; null.cook(force=True); null.lock = True. These commands are standard for managing node state within the targeted application.

td-glsl-shaders