The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a local Python helper script (scripts/mix_matrix.py) to generate concept scaffolds from creative data. This script is distributed as part of the skill package and performs deterministic operations using standard libraries. A separate test script (scripts/test_mix_matrix.py) also uses subprocess.run to execute unit tests on the helper script, which is a standard development practice.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external references (e.g., videos, reels, or descriptions) provided by the user. These references represent a potential surface for indirect prompt injection, where malicious instructions embedded in the reference content could attempt to influence the agent's output.
Ingestion points: The Intake and Reference genome sections of SKILL.md define how the agent collects and deconstructs 2-6 primary references and up to 3 wildcard references.
Boundary markers: The skill employs a structured gene schema (references/reference-gene-schema.md) to categorize observations and transferable principles, which helps delimit the data being processed.
Capability inventory: The agent uses its local environment to create a gene-pool.json file and executes the mix_matrix.py script. The final output is a mix-brief.md document handed to a motion-director skill.
Sanitization: No explicit content filtering is performed on the text descriptions or metadata extracted from the user-supplied references.

mix-and-match