fix-gitlab-vuln
Pass
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to clone, search, and process data from external, user-provided git repositories containing infrastructure code such as Helm values, Terraform modules, and Ansible playbooks. This creates an attack surface for indirect prompt injection if those files contain malicious instructions aimed at the agent.
- Ingestion points: Files within the target repositories are cloned and searched during the discovery phase (SKILL.md Step 1).
- Boundary markers: The instructions do not define strict boundary markers or include specific directives to treat repository content as untrusted data to prevent the agent from executing embedded instructions.
- Capability inventory: The skill utilizes significant local capabilities to perform its tasks, including shell command execution, git operations, and infrastructure validation tools like helm, terraform, and ansible-lint.
- Sanitization: There are no documented steps for sanitizing or filtering the content read from the repositories before the agent uses it to make decisions or generate pull request descriptions.
- [COMMAND_EXECUTION]: The skill requires the use of several shell-based tools to perform inventory and validation of infrastructure changes.
- Evidence: The procedure involves executing commands such as
git clone,rg(ripgrep),helm search,helm template,terraform plan, andansible-lintas part of its standard operational flow (SKILL.md Steps 1 and 9).
Audit Metadata