mr-buddy
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external GitLab Merge Requests, creating a vulnerability surface for indirect instructions to influence the agent.\n
- Ingestion points: Fetches MR metadata, commit history, and file change lists via the
glab apiin Step 1 (SKILL.md).\n - Boundary markers: Absent; there are no instructions to the model to ignore or delimit potentially malicious instructions within the fetched data.\n
- Capability inventory: The skill has significant capabilities including file system modification via
git, network operations viaglab, and execution of local scripts viamake verify(SKILL.md).\n - Sanitization: Absent; commit messages and MR descriptions are ingested and reused to plan slices and generate new MR descriptions.\n- [COMMAND_EXECUTION]: The skill frequently executes shell commands to perform its core tasks. This includes widespread use of
gitfor branching and history manipulation, andglabfor GitLab API operations and MR creation. These commands interpolate variables such as project paths and Merge Request IDs, which could be malicious if the source metadata is manipulated.\n- [DYNAMIC_EXECUTION]: In Step 4, the skill is directed to execute the project's local verification routine (e.g.,make verify) as specified in the repository'sAGENTS.md. This involves running local code found within the repository being processed, which represents a runtime execution surface for potentially malicious project code.
Audit Metadata