ai-incident-response

Pass

Audited by Gen Agent Trust Hub on Aug 6, 2026

Risk Level: SAFENO_CODE
Full Analysis
  • [NO_CODE]: The skill consists solely of a markdown file containing instructional content and workflow procedures. No scripts or executable files are provided.
  • [SAFE]: No patterns of hardcoded credentials, sensitive data access, or network exfiltration were found. The skill references internal project documentation and tools for legitimate diagnostic purposes.
  • [PROMPT_INJECTION]: The skill defines a process for analyzing production failures which involves ingesting potentially untrusted data like model outputs and user inputs. This is a potential surface for indirect prompt injection.
  • Ingestion points: The agent is instructed to read 'bad outputs' and 'inputs that triggered the failure' from the production environment logs mentioned in SKILL.md.
  • Boundary markers: The instructions do not specify the use of delimiters or safety warnings for the agent when processing these files.
  • Capability inventory: The skill is limited to Read and Write tools, and model invocation is disabled, reducing the risk of autonomous malicious action.
  • Sanitization: While the skill mentions using 'sanitized' samples, it does not define or enforce a specific sanitization process.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 6, 2026, 01:53 PM
Security Audit — agent-trust-hub — ai-incident-response