The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/install-skill.sh script executes shell commands to copy files, create metadata, and run a local validator script (quick_validate.py) located in the vendor's system directory. These actions are intended for local environment maintenance.
[PERSISTENCE]: The model installation guides suggest adding export commands to shell profiles (~/.bashrc or ~/.config/fish/config.fish). This is a standard method for persisting environment variables like OLLAMA_MODELS and HF_HOME across terminal sessions.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided directory paths when installing new skills. The risk of unintended command execution through these paths is managed by the agent's requirement to perform a --dry-run and obtain user approval before applying changes. \n
Ingestion points: Path and name arguments provided to scripts/install-skill.sh (referenced in SKILL.md). \n
Boundary markers: The agent prompt requires a mandatory --dry-run step and user confirmation. \n
Capability inventory: File system writes, shell command execution via eval, and modification of system-level configuration files. \n
Sanitization: The scripts use standard shell quoting to handle variable interpolation.
[EXTERNAL_DOWNLOADS]: The skill facilitates the setup of external resources from well-known platforms like HuggingFace, Ollama, and GitHub. These references are used to download models and configuration files as part of the system installation process.
[DYNAMIC_EXECUTION]: The shell script scripts/install-skill.sh uses the eval command to execute dynamically constructed file management strings. This usage is restricted to local installation tasks and includes safeguards for review.

lora-tooling