engineering-judgment

Pass

Audited by Gen Agent Trust Hub on Nov 6, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes standard development tools and version control commands (e.g., git, dotnet test, ng build, code) to manage project state and verify code changes. These operations are restricted to the local workspace and intended for software development tasks.
  • [DYNAMIC_EXECUTION]: The orchestrator dispatches specialized subagents (analyst, backend, frontend, reviewer, QA) to perform specific phases of the development lifecycle. This is handled via the platform's native agent invocation tools and uses isolated contexts ("fresh agents") for each task.
  • [INDIRECT_PROMPT_INJECTION]: The skill recognizes the risk of processing untrusted repository content. It implements a security boundary by explicitly instructing the agent to treat project code, comments, and files as data rather than instructions that could override the skill's operational logic.
  • [PRIVILEGE_ESCALATION]: The skill incorporates a risk rubric that mandates increased verification and human sign-off for sensitive operations (Level 2 risks like authentication, payments, or data migrations). It specifically prohibits unauthorized changes to agent definitions or repository policies.
  • [DATA_EXFILTRATION]: The skill includes instructions to prevent the exposure of sensitive information, such as redacting data from UI evidence (screenshots/recordings) and excluding secrets from the Engineering Judgment Log.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 6, 2026, 02:09 PM