fix-merge-conflicts
Pass
Audited by Gen Agent Trust Hub on Jul 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: Employs git, bun install, bunx tsc, and bun run test to manage the resolution and verification of code conflicts. These operations are standard for development environments and are used as intended to maintain code integrity.
- [EXTERNAL_DOWNLOADS]: Invokes bun install which may download packages from the npm registry to synchronize lockfiles. This is an expected behavior when resolving conflicts in source control and uses well-known technology services.
- [PROMPT_INJECTION]: The skill processes code from unmerged git paths, which constitutes an indirect prompt injection surface.
- Ingestion points: Accesses untrusted file content using git show as outlined in SKILL.md.
- Boundary markers: Incorporates clear instructions to treat all incoming changes as untrusted and avoid treating embedded comments as commands.
- Capability inventory: Has shell access to execute build commands and test suites within the repository context.
- Sanitization: Explicitly directs the agent to ignore instructions within code/comments and to redact secret-like values during the resolution process.
Audit Metadata