The Agent Skills Directory

REMOTE_CODE_EXECUTION (CRITICAL): The troubleshooting section explicitly instructs users to execute a script from an untrusted source via shell piping.
Evidence: curl -LsSf https://astral.sh/uv/install.sh | sh in SKILL.md.
Risk: This pattern allows the remote server to execute arbitrary code on the local system with the current user's privileges. The source 'astral.sh' is not in the defined trusted scope.
COMMAND_EXECUTION (HIGH): The skill defines multiple 'just' and 'uv' commands that execute arbitrary shell commands and Python scripts.
Evidence: Justfile commands such as uv run python src/discohy_thread_operad.py and uv run python -c ....
Risk: If these scripts or their inputs are compromised, it leads to full local execution capability.
EXTERNAL_DOWNLOADS (MEDIUM): The skill installs several third-party Python packages from PyPI.
Evidence: uv pip install discopy>=1.1.0, uv pip install hy>=1.0.0.
Risk: Unverifiable dependencies may introduce supply chain risks if packages are malicious or hijacked.
INDIRECT_PROMPT_INJECTION (HIGH): The skill processes 'thread' data which is likely external content, and has the capability to execute commands based on system state.
Ingestion points: threads list passed to build_operad_from_threads in SKILL.md.
Boundary markers: Absent.
Capability inventory: uv run, uvx, ruff, and just commands allow subprocess execution.
Sanitization: Absent. Untrusted thread data could influence the agent's logic in the 'discohy' environment.

uv-discohy