implement-feature

Pass

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted external data in the form of specification files and existing codebase content which could contain adversarial instructions.
  • Ingestion points: Phase 1 reads files from the path provided in arguments and analyzes the local codebase.
  • Boundary markers: The process includes a validation gate that stops execution and waits for user confirmation of the implementation plan before Phase 3.
  • Capability inventory: The skill has access to Read, Edit, Write, Bash, and Task tools, enabling filesystem modifications and local command execution.
  • Sanitization: No automated sanitization is performed on the ingested content; the primary mitigation is the manual review step.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 10, 2026, 01:21 AM
Security Audit — agent-trust-hub — implement-feature