The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches installation scripts, UI source code, and large model weights from GitHub (citronlegacy, aitrepreneur, kohya-ss) and Hugging Face (circlestone-labs).
[REMOTE_CODE_EXECUTION]: Clones the kohya-ss/sd-scripts repository and executes the anima_train_network.py script via the accelerate launch utility.
[COMMAND_EXECUTION]: Executes shell commands such as git, pip, winget, and wget to configure the Python environment and download dependencies. It also runs generated batch files and shell scripts for launching the trainer.
[DYNAMIC_EXECUTION]: At runtime, the skill generates platform-specific launcher scripts (.bat for Windows, .sh for Linux) and configuration files in TOML and YAML formats. It also dynamically patches the app.py source code to set defaults.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied image captions from a local directory as input for the model training process.
Ingestion points: Captions are read from .txt files located in the image_dir specified in SKILL.md.
Boundary markers: The skill does not document specific delimiters or instructions to ignore potential commands within the caption text.
Capability inventory: The environment includes subprocess execution (accelerate launch), file system write access, and network capabilities (git, wget).
Sanitization: No sanitization or validation of the text content within the dataset captions is described.

anima-lora-trainer