The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/crash_trace.py executes the rr replay command via subprocess.run. The implementation follows security best practices by using a list-based command argument structure to prevent shell injection, sanitizing the environment variables to remove high-risk variables like LD_PRELOAD and PYTHONPATH, and setting a known-safe working directory to prevent the automatic loading of potentially malicious local configuration files.
[INDIRECT_PROMPT_INJECTION]: The skill processes external crash traces and outputs the results, including source code and variable values, to the agent. While these traces represent untrusted external data, the script explicitly disables GDB's automated script-loading features using set auto-load no and set auto-load python-scripts off. This significantly reduces the risk of malicious code execution during the trace extraction phase. The skill provides clear boundary markers in the output trace to help the agent distinguish between debugger metadata and the trace content.

rr-debugger