phase-2-refine

Pass

Audited by Gen Agent Trust Hub on Jul 11, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill demonstrates a mature security posture with clear boundaries and restricted file system access. It explicitly prohibits mutations outside of plan-specific local directories.
  • [PROMPT_INJECTION]: The 'Untrusted external-issue-body ingestion' protocol (defined in standards/source-premise-verification.md) provides significant protection against indirect prompt injection. It employs a read-only reader for isolation and a deterministic validator gate to sanitize external issue bodies before they are consumed by the write-capable context.
  • [COMMAND_EXECUTION]: Shell execution is performed through a controlled script executor (.plan/execute-script.py). The operations are limited to project metadata management, git synchronization (git fetch), and internal state logging.
  • [DATA_EXFILTRATION]: The skill includes strict 'Prohibited actions' and 'Allowed write paths' sections in SKILL.md, ensuring that sensitive project configuration and production code remain isolated from the refinement process.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 11, 2026, 07:12 AM
Security Audit — agent-trust-hub — phase-2-refine