om-app-spec-writing
Pass
Audited by Gen Agent Trust Hub on Jul 17, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill defines a clear 'Untrusted content boundary' for all repository data (issues, PRs, logs). It explicitly identifies and instructs the agent to ignore prompt injection patterns like 'ignore previous instructions' or 'run this command' found in external data.
- [COMMAND_EXECUTION]: The skill mandates strict input validation for any shell command interpolation, using an alphanumeric allowlist regex and requiring shell quoting. Commands are restricted to building, testing, or reviewing the local project.
- [DATA_EXFILTRATION]: There are explicit prohibitions against data exfiltration and accessing sensitive files (e.g., credential stores) outside the repository's scope. Network access for tracking tools is restricted to read-only operations.
- [SAFE]: The skill uses a 'Hard Gate' logic that prevents any implementation or code generation tools from being called until the user manually confirms the business specification, ensuring human oversight.
Audit Metadata