The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to perform its core functions. It uses uv run to invoke a local Python script (_bmad/scripts/resolve_customization.py) for configuration management and relies on git (e.g., git diff) and gh (e.g., gh pr view, gh pr review --approve) for code analysis and interaction. The instructions explicitly state that certain actions, like approving a PR, should only be performed after human confirmation.
[PROMPT_INJECTION]: The skill processes untrusted external data in the form of code diffs, specifications, and project status files, which presents a surface for indirect prompt injection. However, this is inherent to the tool's primary purpose of code review.
Ingestion points: Reads output from git diff, contents of specification files, and project configuration files (config.yaml) in SKILL.md, step-01-orientation.md, and step-03-detail-pass.md.
Boundary markers: The instructions do not define specific delimiters or instructions to ignore embedded commands within the analyzed code, though the workflow is designed to be guided by a human reviewer.
Capability inventory: The agent can read local files, execute local scripts via uv, and perform network operations via the GitHub CLI (gh).
Sanitization: No explicit sanitization or filtering of the ingested code content is described before it is presented to the agent for analysis.

bmad-checkpoint-preview