iterate-until-verified
Pass
Audited by Gen Agent Trust Hub on Aug 26, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill is designed to ingest and process untrusted user-supplied data as the 'authoritative task' for its iterative loop. This creates a surface for indirect prompt injection.
- Ingestion points: Untrusted tasks are captured in the 'Lock the original task' phase and used as the primary input for all subsequent processing steps.
- Boundary markers: The skill employs specific boundary markers, such as the
AUTHORITATIVE TASKtags in its template, to isolate user input from the meta-instructions. - Capability inventory: The skill enables high-impact capabilities including task decomposition, fan-out to subagents, and potentially code execution for verification (tests, builds, runtime behavior).
- Sanitization: The instructions explicitly warn against expanding permissions or relaxing constraints based on the task content, which serves as a logical integrity check against malicious instructions embedded in the user task.
Audit Metadata