merge-deps
Pass
Audited by Gen Agent Trust Hub on Oct 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from pull request titles, descriptions, and comments during the triage process.
- Ingestion points: PR metadata and comments are retrieved and processed during the detection and assessment phases (SKILL.md, Step 2 and 6).
- Boundary markers: The skill explicitly instructs the agent to treat unauthorized text as context only and to stop for a human if instructions are detected (REFERENCE.md, Author authority section).
- Capability inventory: The skill has the capability to merge code, delete branches, and execute shell commands (SKILL.md, Step 4).
- Sanitization: Strict author verification is enforced (e.g., matching app/dependabot or specific GitLab bot IDs) to determine which requests are authoritative.
- [DYNAMIC_EXECUTION]: The skill executes local repository scripts and package manager commands (e.g., npm install, pnpm check) on code from incoming pull requests to verify updates.
- Evidence: SKILL.md Step 3 and REFERENCE.md ("Running the repo's checks") describe the execution of a verify command and lockfile installations.
- Mitigation: The instructions mandate the use of an OS-enforced runner for isolation to ensure code execution does not inherit the operator's authority, credentials, or environment.
- [COMMAND_EXECUTION]: The skill utilizes forge CLI tools (gh, glab) and git commands to perform repository management and merging tasks.
- Evidence: SKILL.md Step 2 and 4 detail the use of CLI tools to list, view, and merge pull/merge requests.
Audit Metadata