doris-repo-review
Pass
Audited by Gen Agent Trust Hub on Aug 20, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- Repository Workspace Integrity: The skill manages local repository state by detaching to specific PR commits. It includes rigorous safety checks in
align-to-pr.sh, ensuring that the agent does not overwrite or discard uncommitted local work. It explicitly refuses to proceed if tracked files are modified, handing control back to the user. - Secure GitHub Interactions: All communications with GitHub are performed through the official
ghCLI. The skill targets the official project repository (apache/doris) and restricts write operations to a single pass-commenting function. These interactions are consistent with the intended developer workflow and follow the principle of least privilege. - Controlled Remote Write Pattern: The
post-pass-comment.shscript facilitates the only external write action. This process is gated by multiple conditions: the review must not have high-severity findings, the PR head must remain unchanged, and most importantly, the user must manually review and approve the exact comment body before it is posted. - Indirect Prompt Injection Surface: As a code review tool, the skill naturally ingests untrusted data from PR diffs and existing comments. This represents a standard surface for indirect prompt injection. The skill mitigates this by using structured prompt templates for sub-tasks and maintaining human oversight for any resulting actions that affect external state.
- Execution of Local Utility Scripts: The skill utilizes helper scripts in Python and Bash to perform diff analysis and line-number verification. These scripts are self-contained within the skill package and operate only on the local context directory or the repository worktree to provide metadata for the review process.
Audit Metadata