feature-dev-mini
Pass
Audited by Gen Agent Trust Hub on Jun 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to discover and run local test commands (such as
npm testorpytest) to validate code changes. This is a legitimate functional requirement for a development-focused tool and is conducted within the local project environment. - [COMMAND_EXECUTION]: The skill requires two distinct human approval markers, 'APPROVE DISCOVERY' and 'APPROVE PLAN', before allowing any file write or creation operations. This ensures that all automated changes are reviewed and authorized by the user.
- [PROMPT_INJECTION]: The skill ingests codebase content during its discovery phase, which constitutes an indirect prompt injection surface. This potential vulnerability is mitigated by the mandatory planning and approval phases that verify the agent's proposed actions before execution.
- Ingestion points: Local source files scanned during the 'DISCOVERY_SCAN' state in Step 1a.
- Boundary markers: Use of explicit state-tracking headers (e.g., 'Current state: DISCOVERY_SCAN') and strict sequence-of-operations rules.
- Capability inventory: Local shell command execution for tests and gated file system write operations.
- Sanitization: Not explicitly implemented within the skill instructions.
Audit Metadata