bmad-code-review
Pass
Audited by Gen Agent Trust Hub on Nov 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from multiple external sources, creating a potential surface for indirect prompt injection attacks.
- Ingestion points: The skill reads and processes unified diffs (
{diff_file}), ticket descriptions ({ticket_file}), and developer narratives ({claims_file}) instep-01-gather-context.md. - Boundary markers: The subagent prompts defined in
customize.toml(e.g., for theblind-hunterandquicklenses) lack clear boundary markers or delimiters between the instructions and the untrusted content being reviewed. - Capability inventory: The skill has significant capabilities, including executing shell commands (
git,gh,uv) and the ability to modify the project's source code by applying patches instep-05-present.md. - Sanitization: There is no explicit sanitization of the content before it is passed to subagents. While
step-03-triage.mdincludes a manual verification step for findings, it relies on the agent's judgment to detect malicious intent within the findings. - [COMMAND_EXECUTION]: The skill relies on executing shell commands and local project scripts to function.
- Execution of local scripts: In
SKILL.mdandstep-01-gather-context.md, the skill usesuv runto execute scripts located within the project's own directory structure (_bmad/scripts/render_skill.pyand_bmad/method/scripts/tickets.py). This creates a dependency on the integrity of the project being reviewed. - VCS Tooling: The skill invokes
gitandgh(GitHub CLI) to retrieve diffs, logs, and pull request information.
Audit Metadata