The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides procedures for executing Python debugging tools via the shell.
Evidence: Commands such as python3 -m pdb path/to/script.py and python3 -m debugpy --listen 127.0.0.1:5678 are documented for interactive use.
Mitigation: The skill explicitly warns to confirm the exact PID and owner before attaching and to keep ports on loopback (127.0.0.1).
[DYNAMIC_EXECUTION]: The skill utilizes debugpy's capability to inject into running processes for debugging purposes.
Evidence: references/upstream/SKILL.source.md contains the command python3 -m debugpy --listen 127.0.0.1:5678 --pid <pid>.
Context: While process injection is a sensitive capability, it is the primary intended function of the debugger and is accompanied by instructions to verify target PID ownership and ptrace privileges.
[INDIRECT_PROMPT_INJECTION]: The skill processes external Python scripts and provides the agent with access to runtime state, which represents a potential attack surface for malformed data to influence the agent.
Ingestion points: The agent is instructed to run and debug arbitrary Python scripts (path/to/script.py) provided by the user.
Boundary markers: SKILL.md contains instructions to "report the failing frame, observed values with private data excluded".
Capability inventory: The skill has the capability to execute shell commands (python3), attach to PIDs, and listen on local ports.
Sanitization: Instructions strictly prohibit using the debugger to bypass credential checks or filesystem/network restrictions and mandate the removal of all debugging statements before completion.

openclaw-python-debugpy