The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local commands such as python3, git, and code as part of its standard workflow. These operations are used for resolving project-level configurations, managing version control commits, and facilitating developer review by opening files in an editor.
[COMMAND_EXECUTION]: The skill references and executes vendor-specific scripts (e.g., resolve_customization.py) located within the project's own directory structure (_bmad/scripts/). These scripts are used for deep-merging configuration files and post-completion actions.
[PROMPT_INJECTION]: The skill processes untrusted external data such as user intent and planning artifacts (PRDs, architecture guides) to generate implementation plans. This creates a surface for indirect prompt injection.
Ingestion points: Ingests content from {planning_artifacts} (PRD, architecture, epics) and project-wide documentation (e.g., **/project-context.md, CLAUDE.md).
Boundary markers: Implements a <frozen-after-approval> tag system in the specification files to protect human-validated intent from being modified by subsequent agent iterations.
Capability inventory: The skill can read/write files and execute subprocesses including python3, git, and code commands.
Sanitization: Employs an iterative review process including multi-agent adversarial checks (bmad-review-adversarial-general, etc.) and mandatory human approval checkpoints before transitioning from planning to implementation.

gds-quick-dev