post-incident-review

Pass

Audited by Gen Agent Trust Hub on Jul 15, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill is designed to ingest and process potentially untrusted external data during its analysis phase.
  • Ingestion points: The skill reads external incident reports, communication logs, and forensic findings using the Read, Grep, and Glob tools as outlined in Section 2.
  • Boundary markers: There are no explicit structural delimiters or boundary markers defined in the process to isolate untrusted content from agent instructions, although Section 8 provides a text-based safety warning to the agent.
  • Capability inventory: The skill is restricted to file reading (Read, Grep, Glob) and does not possess tools for network exfiltration, arbitrary command execution, or code evaluation.
  • Sanitization: No specific input sanitization or validation routines are specified for the external data being analyzed.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 15, 2026, 01:06 AM
Security Audit — agent-trust-hub — post-incident-review