The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script (resolve_customization.py) located in the project's _bmad/scripts directory during the activation and completion phases to manage configuration overrides.
[COMMAND_EXECUTION]: The workflow relies on standard development CLI tools including git for diff generation and branch context, as well as gh for interacting with GitHub Pull Requests.
[COMMAND_EXECUTION]: The skill features a dynamic 'on_complete' hook that resolves a command string from the user's local configuration and executes it as a terminal instruction upon workflow completion.
[PROMPT_INJECTION]: The automation-mode.md file defines a set of behavioral overrides that instruct the agent to suppress standard conversational elements (e.g., 'Never HALT for input', 'skip the greeting') when the BMAD_AUTO_MODE environment variable is active. These are functional overrides for automated orchestration.
[PROMPT_INJECTION]: The skill processes untrusted external data including git diffs and development specifications which are then passed to various LLM subagents.
Ingestion points: Untrusted data enters via diff_output (from git diff) and {spec_file} (Markdown/YAML) in Step 1.
Boundary markers: The instructions do not define explicit delimiters or 'ignore' instructions for the subagent prompts.
Capability inventory: The agent has capabilities for file system writes (Step 4), local script execution (python3), and git command execution.
Sanitization: There is no explicit evidence of sanitization or escaping of the ingested code or specification content before it is processed by the review layers.

bmad-auto-review