bmad-ux

Pass

Audited by Gen Agent Trust Hub on Nov 13, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses uv run to execute local Python scripts (resolve_customization.py, resolve_config.py, and memlog.py) located within the project's _bmad/scripts/ directory. These scripts are used for environment setup, configuration resolution, and maintaining a decision log.
  • [COMMAND_EXECUTION]: The skill employs platform-specific openers (open on macOS, xdg-open on Linux, start on Windows) to automatically display generated HTML artifacts, such as color themes, design directions, and validation reports, in the user's browser.
  • [EXTERNAL_DOWNLOADS]: Instructions are provided to install the foundational bmad skill using npx skills add bmad-code-org/BMAD-METHOD if the environment is not yet configured. It also references the Google Labs DESIGN.md specification and the Google Stitch tool for design handoffs.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses a documented attack surface for indirect prompt injection as it ingests untrusted data from project tickets, external source documents, and user-supplied assets during the discovery phase.
  • Ingestion points: Requirements are read from files such as initiative-*.md, epic-*.md, and other ticket types, as well as documents stored in imports/ or linked in the sources: frontmatter.
  • Boundary markers: The skill uses a 'Reviewer Gate' and a 'Rubric Walker' (defined in references/validate.md) to perform coverage checks and validation, though specific text delimiters for untrusted input interpolation are not explicitly defined in the provided markdown instructions.
  • Capability inventory: The skill can write files to the local workspace, execute local Python scripts via uv run, and trigger system commands to open files in a web browser.
  • Sanitization: The skill relies on subagents to distill and reconcile inputs into structured markdown documents, providing an inherent layer of human-readable abstraction before finalized content is committed.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 13, 2026, 10:20 PM