hate_crack/docs/plans/2026-03-17-omen-training-fix-design.md

OMEN Training Fix + Wordlist Picker

Problem

OMEN attack (menu option 16) fails silently - hashcat exits immediately without candidates. Root causes:

1. Model validation only checks createConfig exists, not the 4 .level files
2. No way to retrain once a model exists (training permanently skipped)
3. enumNG stderr/exit code not captured - failures are invisible
4. Training uses raw path input instead of the wordlist picker menu

Design

Bug fixes (main.py)

Model validation - Add _omen_model_is_valid(model_dir) that checks all 5 required files exist and are non-empty: createConfig, CP.level, IP.level, EP.level, LN.level.

enumNG error handling - Capture stderr from enumNG subprocess. If it exits non-zero or stderr has content, print the error and return early instead of letting hashcat sit on empty stdin.

Training return value - hcatOmenTrain returns bool (True on success) so callers can abort if training fails.

Model metadata - Write model_info.json alongside model files with {"training_file": "...", "trained_at": "..."}. Add _omen_model_info(model_dir) to read it.

Menu flow (attacks.py: omen_attack)

1. Check OMEN binaries exist (unchanged)
2. Call _omen_model_is_valid() to check model
3. If valid, show status and prompt: Use existing / Train new / Cancel
4. If invalid/missing, go straight to training
5. Training wordlist picker: numbered list from list_wordlist_files() + custom path option
6. If training returns False, abort
7. Prompt for max candidates, run hcatOmen

Files modified

• hate_crack/main.py - _omen_model_is_valid, _omen_model_info, hcatOmenTrain return bool + write metadata, hcatOmen capture enumNG stderr
• hate_crack/attacks.py - Rewrite omen_attack() with train/use/cancel + wordlist picker
• tests/test_main_utils.py - Model validation, training return value, enumNG error tests
• tests/test_attacks_behavior.py - omen_attack menu flow tests