The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill fetches an external repository and executes code from it.
Evidence: The instructions require git clone https://github.com/hamelsmu/hamelnb.git followed by uv run "$SCRIPT", where the script is part of the cloned repository.
[EXTERNAL_DOWNLOADS]: The skill relies on code downloaded from a third-party source not identified as a trusted vendor.
Evidence: git clone https://github.com/hamelsmu/hamelnb.git ~/.agent-skills/hamelnb.
[CREDENTIALS_UNSAFE]: The setup instructions explicitly disable Jupyter's built-in security features.
Evidence: jupyter-lab --no-browser ... --IdentityProvider.token='' --ServerApp.password=''. Disabling tokens and passwords allows any process with local access to the port to execute arbitrary code in the kernel.
[COMMAND_EXECUTION]: The skill facilitates the execution of arbitrary shell and Python commands.
Evidence: Frequent use of uv run to execute the downloaded script with user-supplied arguments and --code flags.
[INDIRECT_PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection by ingesting and executing content from notebook files and code strings.
Ingestion points: SKILL.md (via --path and --code arguments).
Boundary markers: None present.
Capability inventory: Arbitrary Python execution, server management, and file editing.
Sanitization: None provided; the skill is designed to execute raw code inputs.
[DYNAMIC_EXECUTION]: The core functionality involves the dynamic execution of Python code within a persistent kernel session.
Evidence: The execute subcommand takes arbitrary strings to run in the Jupyter environment.

jupyter-live-kernel