validate-changes-match-specs
Pass
Audited by Gen Agent Trust Hub on Aug 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local shell commands for repository management, including
gitoperations (diff, commit, push, merge-base) and automated testing or linting tools found within the project. All destructive or persistent actions, such as committing or pushing code, are gated behind explicit user confirmation prompts. - [DATA_EXFILTRATION]: For product validation, the skill transmits repository context, including specification excerpts and potentially environment configuration or 'safe to share' credentials, to external 'Oz' cloud agents and Figma. These interactions are core to the skill's functionality and target vendor-controlled or well-known services.
- [EXTERNAL_DOWNLOADS]: The skill retrieves external data such as pull request comments, commit history, and design specifications from GitHub and Figma to perform its comparative analysis.
- [PROMPT_INJECTION]: The skill includes defensive instructions to mitigate indirect prompt injection by explicitly telling the agent to treat specifications, PR descriptions, and comments as untrusted data and to ignore any embedded instructions attempting to override the agent's role or safety guidelines.
Audit Metadata