The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Python's subprocess.run to execute local commands such as git and bmad-loop. These calls use list-based arguments without the shell=True flag, which is a safe practice that prevents shell injection attacks. The commands are scoped to repository status and tool monitoring.
[INDIRECT_PROMPT_INJECTION]: The skill processes terminal captures generated by an autonomous agent. It implements robust sanitization and reconstruction logic—such as stripping ANSI escape sequences and using normalized prefix bucketing—to ensure the analyzing agent is not misled by terminal redraw artifacts, hidden text patterns, or collapsed output markers common in terminal-based TUIs.
[DATA_EXPOSURE]: Analysis of the included scripts (run_probe.py, extract_transcript.py, bootstrap_adapter.py) confirms that data is read only from the local .bmad-loop run directories and internal project configuration files. There are no network requests, external data uploads, or attempts to access sensitive system files like credentials or environment variables.
[PRIVILEGE_ESCALATION]: The skill uses ps and os.kill with signal 0 to check if the engine process is still alive. These operations are restricted to standard user permissions and do not attempt to gain administrative or root access or bypass system security controls.

bmad-run-inspector