error-recovery

Pass

Audited by Gen Agent Trust Hub on Jul 8, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill framework possesses an indirect prompt injection surface as it is designed to ingest and process error event payloads and tool outputs from potentially untrusted external sources.
  • Ingestion points: The skill architecture in SKILL.md is built to process 'Error Event Payloads' and 'Agent tool call failures' which enter the agent's context from external dependencies.
  • Boundary markers: The provided implementation patterns do not explicitly define strict delimiters for these payloads, though they suggest classification before recovery.
  • Capability inventory: The skill scripts include file system write capabilities for state persistence in './checkpoints' and generic examples for network data fetching.
  • Sanitization: SKILL.md identifies sensitive data leakage in error logs as a security risk and explicitly recommends stripping sensitive information before logging or reporting, which is a key mitigation for this attack surface.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 8, 2026, 12:23 PM
Security Audit — agent-trust-hub — error-recovery