mii-ig-migration

Pass

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]:
  • The skill executes several migration-specific utilities such as npx fsh-sushi, npx gofsh, and git clone (to re-vendor repository templates).
  • It also invokes the official HL7 FHIR validator using java -jar to generate element-tree snapshots.
  • These operations are fundamental to the skill's intended purpose and are performed with version-pinned tools where possible.
  • [EXTERNAL_DOWNLOADS]:
  • The skill fetches FHIR packages and rendered guide content from simplifier.net, which is the established platform for the German Kerndatensatz modules.
  • It downloads the HL7 FHIR validator utility (validator_cli.jar) from the official hapifhir organization on GitHub.
  • These downloads are conducted from trusted, well-known services relevant to the FHIR development domain.
  • [INDIRECT_PROMPT_INJECTION]:
  • Ingestion points: External guide narrative is scraped from Simplifier by guide-harvest.sh and converted into Markdown by guide-page-to-md.py.
  • Boundary markers: Imported content is marked with provenance comments and TODO:REVIEW tags; any content written by the migration itself is wrapped in visible DERIVED markers.
  • Capability inventory: The skill environment allows extensive capabilities including file system access, network operations, and execution of various runtimes (Node.js, Python, Java).
  • Sanitization: The guide-page-to-md.py script performs basic HTML sanitization by stripping scripts, style tags, and SVGs during the harvesting process.
  • This ingestion surface could potentially expose the agent to malicious instructions hidden in the source guide's narrative, making the skill's mandatory human review gates (Gates B and C) essential for safe operation.
  • [DYNAMIC_EXECUTION]:
  • The logging utility scripts/migration-log.sh provides a run subcommand that wraps and executes arbitrary migration steps while capturing metadata and exit statuses.
  • The bounded auto-fix loop in scripts/autofix-loop.sh dynamically plans and executes repairs based on findings detected during the verification phase.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 28, 2026, 09:56 AM
Security Audit — agent-trust-hub — mii-ig-migration