The Agent Skills Directory

[COMMAND_EXECUTION]: Executes a local bash script (kokoro-install.sh) with several flags (--health, --upgrade, --uninstall, --install) to manage the TTS environment.
[EXTERNAL_DOWNLOADS]: Downloads machine learning model weights from the mlx-community repository on Hugging Face and updates Python packages (mlx-audio, soundfile, numpy) via PyPI.
[PROMPT_INJECTION]: Includes 'Self-Evolving Skill' and 'Post-Execution Reflection' instructions that direct the agent to modify the SKILL.md file itself if issues are encountered. This creates a persistence mechanism where the agent can overwrite its own instructions.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect injection because it reads untrusted output from shell commands and version files, then uses that information to self-modify its instructions.
Ingestion points: Output from the --health check command, the content of ~/.local/share/kokoro/version.json, and results from the test synthesis command.
Boundary markers: None present; the agent is not instructed to distinguish between legitimate system output and potential embedded instructions.
Capability inventory: The skill possesses full shell access via the Bash tool and the ability to modify its own source files.
Sanitization: No validation or sanitization is performed on the data ingested before it is used to update the skill's logic.

augment