using-git-worktrees

Pass

Audited by Gen Agent Trust Hub on Aug 6, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes standard Git commands and custom shell scripts (copy-env-local.sh, dehydrate-worktree.sh, storage-preflight.sh) for workspace management. These operations are scoped to the project's repository and dedicated worktree directories.
  • [CREDENTIALS_UNSAFE]: The copy-env-local.sh script specifically includes logic to strip sensitive Convex API keys and admin credentials when copying environment files between workspaces, demonstrating a security-first approach to credential handling.
  • [DATA_EXFILTRATION]: No patterns of unauthorized data exfiltration were detected. The skill enforces local-only Convex deployments and prevents the import of production or shared development data into the isolated workspaces.
  • [EXTERNAL_DOWNLOADS]: The skill uses bun install --frozen-lockfile for dependency management. This is a standard and safe practice for deterministic builds, ensuring that only expected packages from the project's lockfile are installed.
  • [REMOTE_CODE_EXECUTION]: There is no evidence of remote code execution. All scripts are local to the skill directory, and command execution is strictly limited to managing the local Git worktree lifecycle.
  • [PROMPT_INJECTION]: The instructions are instructional and focused on operational safety. No attempts to override agent behavior or bypass safety guidelines were found.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 6, 2026, 10:11 PM
Security Audit — agent-trust-hub — using-git-worktrees