dsh-code-review
Pass
Audited by Gen Agent Trust Hub on Sep 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions include running a repository-local script using
pnpm --silent run change-scope. This involves command execution with variable arguments (baseandheadreferences) derived from the pull request being reviewed. While this is a common developer workflow, it involves executing shell commands with parameters derived from external PR data. - [INDIRECT_PROMPT_INJECTION]: The agent is instructed to read and analyze pull request diffs and surrounding source code, which are untrusted inputs. This exposes the agent to indirect prompt injection attacks where malicious code or comments could attempt to subvert the review process.
- Ingestion points: Pull request diffs, repository source files, and base/head branch references.
- Boundary markers: No markers or instructions to treat code as data are present.
- Capability inventory: Local command execution via
pnpmand repository file access. - Sanitization: No sanitization of the processed code content is specified.
Audit Metadata