incident-commander

Pass

Audited by Gen Agent Trust Hub on Apr 5, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local Python scripts (scripts/incident_classifier.py, scripts/timeline_reconstructor.py, and scripts/pir_generator.py) to process incident descriptions, log events, and generate summary reports.
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests and processes untrusted data from incident logs and user-provided descriptions without sufficient isolation.
  • Ingestion points: Data is read from files in the assets/ directory (e.g., db_incident_events.json, api_incident_logs.json) and through standard input.
  • Boundary markers: No explicit boundary markers or instructions to disregard embedded commands are provided in the orchestration logic.
  • Capability inventory: The skill can execute local Python scripts and write files to the local file system (e.g., timeline.md, pir.md).
  • Sanitization: No input sanitization or validation steps are defined for the ingested data.
  • [PROMPT_INJECTION]: There is a metadata discrepancy where the SKILL.md file identifies the author as "Claude Skills Team," which conflicts with the provided author context of "Roger-235."
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 5, 2026, 08:07 PM
Security Audit — agent-trust-hub — incident-commander