bmad-agent-architect

Fail

Audited by Gen Agent Trust Hub on Nov 13, 2026

Risk Level: HIGHCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill's activation sequence involves executing entries from activation_steps_prepend and activation_steps_append. These fields are sourced from configuration files within the project root. If a project contains malicious entries in these lists, the agent will execute them automatically during setup.\n- [DYNAMIC_EXECUTION]: The agent is instructed to use uv run to execute Python scripts found in the project's _bmad/scripts/ directory. This allows for the execution of arbitrary code provided by the repository being worked on.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple sources within the project directory.\n
  • Ingestion points: Merges data from customize.toml and project-specific override files (.toml and .user.toml). It also loads file contents into memory via persistent_facts using path globs.\n
  • Boundary markers: No delimiters or instructions are provided to ensure the agent treats this external content strictly as data.\n
  • Capability inventory: Includes file reading, shell command execution via uv, and skill installation via npx.\n
  • Sanitization: The skill does not define any validation or filtering logic for the merged configuration or the facts loaded from the filesystem.\n- [EXTERNAL_DOWNLOADS]: The skill prompts the user to use npx skills add to fetch additional components from the developer's repository if they are not already present. This pattern facilitates the download and installation of external code into the environment.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Nov 13, 2026, 10:20 PM