The Agent Skills Directory

[COMMAND_EXECUTION]: The run_probe.py script executes local system commands such as git branch, git status, and bmad-loop status using subprocess.run to gather environment data. These operations are restricted to a predefined list of binaries in the skill configuration and are used exclusively for monitoring purposes.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted terminal logs and session journal files generated during autonomous coding runs. An attacker who controls the output of a process being executed within the monitored loop could potentially inject malicious text into the log files that this skill parses.
Ingestion points: .bmad-loop/runs/<run-id>/logs/*.log, journal.jsonl, and state.json files.
Boundary markers: The skill does not implement explicit delimiters or safety warnings for the data read from logs.
Capability inventory: The skill has access to tools including git, bmad-loop, tmux, and ps via its allowed-tools configuration.
Sanitization: The extract_transcript.py script strips ANSI terminal escape sequences and reconstructs logical lines but does not filter for natural language instructions.
[DATA_EXPOSURE]: The skill reads internal artifacts from .bmad-loop directories, including task prompts, status JSON files, and terminal captures. This data access is necessary for the skill's primary function of providing run forensics and health monitoring to the user.

bmad-run-inspector