walk-commits
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted commit messages and diff hunks, which could contain malicious instructions. Ingestion points: commit data read via git log and git show. Boundary markers: structured templates and secret redaction rules. Capability inventory: shell execution, file writes, and HTML artifact generation. Sanitization: secret redaction for credentials, but no explicit filtering for instruction-based injection.\n- [EXTERNAL_DOWNLOADS]: Fetches the Mermaid library from the JSDelivr CDN (a well-known service) to render diagrams in visual reports.\n- [COMMAND_EXECUTION]: Executes git and bash commands through the provided sidecar.sh script to manage configuration and repository state.\n- [DYNAMIC_EXECUTION]: Generates transient HTML and JavaScript artifacts to provide an interactive review interface.
Audit Metadata