the-jury

Pass

Audited by Gen Agent Trust Hub on Aug 2, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a local Python utility, scripts/tally.py, to perform weighted scoring and audit the deliberation process (detecting bandwagon effects). This script uses standard libraries and performs no network or sensitive file operations.
  • [REMOTE_CODE_EXECUTION]: The instructions direct the agent to spawn multiple subagents in parallel to act as jurors. This utilizes platform-native task delegation features (e.g., Claude Code Task tool) as intended for the skill's decision-making protocol.
  • [DATA_EXPOSURE]: The skill is designed to ingest 'shared evidence' which may include project code, configurations, or documents. This data is used solely within the agent's context to inform the deliberation process.
  • [SAFE]: An indirect prompt injection surface exists because the skill processes untrusted evidence data. However, the risk is mitigated by the skill's structured workflow, which includes blind initial rounds, anonymized deliberation, and a deterministic tallying phase that weights evidence quality.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 2, 2026, 03:22 AM
Security Audit — agent-trust-hub — the-jury