bmad-architecture
Pass
Audited by Gen Agent Trust Hub on Nov 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external ticket files, existing architecture documents, and codebases, which are potential ingestion points for untrusted data.
- Ingestion points: The agent reads ticket files (
<type>-<slug>.md), architecture spines, and project source code to derive invariants. - Boundary markers: The skill employs a 'Reviewer Gate' (references/reviewer-gate.md) that uses parallel subagents to independently verify the spine against a rubric and adversarial lenses, preventing the agent from blindly following embedded instructions.
- Capability inventory: The skill can write architecture documents and memlogs to the project workspace and produce human-facing HTML reports.
- Sanitization: It requires user confirmation for Assumptions and Open Questions before artifacts are finalized.
- [COMMAND_EXECUTION]: The skill executes utility scripts located within the project's
_bmad/scriptsdirectory to manage metadata and perform linting. - Evidence: Subprocess execution of
scripts/memlog.py,scripts/resolve_customization.py, andscripts/lint_spine.pyviauv run. - [DYNAMIC_EXECUTION]: The workflow allows for the execution of custom steps defined in the
customize.tomlconfiguration file. - Evidence: The skill executes
workflow.activation_steps_prependandworkflow.activation_steps_appendas part of its initialization sequence. - [EXTERNAL_DOWNLOADS]: The skill documentation suggests installing the core BMad method components from the vendor's official repository.
- Evidence: Instructions include
npx skills add bmad-code-org/BMAD-METHOD --skill bmad, which targets the author's own trusted organization resources.
Audit Metadata