The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied reference images that could contain embedded instructions or metadata designed to influence agent behavior during the code generation phase.
Ingestion points: Reference images are ingested into the workspace via python make.py new <name> --refs <image> as defined in SKILL.md.
Boundary markers: There are no specific instructions or delimiters provided to the agent to distinguish between visual character data and potential embedded textual instructions in the images.
Capability inventory: The agent has the ability to write Python scripts (draw.py, mascot.py) and execute them using python make.py draw, as well as access external voice design services.
Sanitization: The instructions do not specify any validation or sanitization steps for the image content or metadata before the agent processes them.
[DYNAMIC_EXECUTION]: The skill requires the agent to generate Python source code (mascots/<name>/draw.py) based on external inputs and then execute it.
The instructions direct the agent to "write mascots/<name>/draw.py" and subsequently run python make.py draw <name> and python make.py check <name> to verify the output.
Although the agent is instructed to follow established local templates (e.g., mascots/sparky/draw.py), the pattern of writing and then executing code derived from untrusted inputs is a dynamic execution surface.
[COMMAND_EXECUTION]: The skill involves multiple shell command executions to manage the mascot build lifecycle, including creating directories, running drawing scripts, and performing validation checks.
Evidence includes calls to python make.py new, python make.py draw, python make.py check, and python make.py video.
[EXTERNAL_DOWNLOADS]: The skill suggests installing an external library and using third-party voice services.
It recommends pip install skia-python for graphics rendering.
It references using ElevenLabs Voice Design via python make.py design-voice for voice generation.

draw-mascot