Files
AWatch-rus/.planning/phases/17-hayabusa-production-validation/SUMMARY.md
T

2.2 KiB

Phase 17 Summary: Production Validation

Result

Phase 17 is closed.

What was proven live

A real end-to-end forensic path was executed and closed on production infrastructure:

  1. Windows EVTX package was exported on SHARKON2025.
  2. The real zip package was transferred to 10.10.10.13.
  3. aw-hayabusa accept and aw-hayabusa process-inbox --mode incident were run through the standard wrapper.
  4. Hayabusa generated a real report set with bounded traceability metadata.
  5. The result was linked back into AW-rus case management as bounded forensics.hayabusa metadata.

Live proof record

  • host: SHARKON2025
  • case id: 30
  • intake id: 20260521T125653Z_SHARKON2025-phase17-rerun3
  • package path: /opt/hayabusa/archive/packages/SHARKON2025/20260521T125653Z_SHARKON2025-phase17-rerun3.zip
  • sha256: e86b9abbfc1d706ac706c6c8a89509ab17023344c50880641e9175f73f1198d4
  • report dir: /opt/hayabusa/reports/SHARKON2025/20260521T125654Z_incident_20260521T125653Z_SHARKON2025-phase17-rerun3
  • report artifacts:
    • summary.html
    • manifest.json
    • run.log
    • timeline.jsonl
    • logon-summary-successful.csv
    • logon-summary-failed.csv

Production bugs found and fixed during validation

  • aw-hayabusa treated unzip warning return code 1 as a hard failure for Windows-created zip archives that use backslashes as path separators.
  • timeline modes were using the wrong Hayabusa config path; the wrapper must pass rules/config, not the rules root.

Both issues were fixed in aw-server/hayabusa/aw-hayabusa.sh and retested live against the same package.

Why this closes the phase

  • the path is no longer theoretical or docs-only; it was proven on a real Windows export package
  • traceability from host to package to report directory is explicit
  • AW-rus case linkage now stores bounded forensic metadata exactly as designed
  • the remaining gaps are operational tuning items, not missing core implementation

Tuning backlog after the live run

  • keep at least one preserved sample EVTX zip for future regression reruns
  • consider a self-check in aw-hayabusa doctor for rules/config completeness
  • optionally persist a compact machine-readable proof manifest for future audits