| name | model-inventory-manager |
| description | Audit and cleanup model files and configurations. Use when identifying old models, cleaning up disk space, or reconciling serving configs with physical files. |
Model Inventory Manager
This skill provides workflows for maintaining a lean and functional model storage system.
Workflow
1. Audit Inventory
Run the audit script to identify discrepancies between llama-swap.yaml, models_config.json, and the filesystem.
python3 codex-skills/model-inventory-manager/scripts/audit_models.py
The script categorizes findings into:
- Broken Serving Configs: Referenced in YAML but missing from disk.
- Potential Orphans: On disk but not referenced in the serving config (excluding known non-serving types like Image/Video).
- Dangling Disabled Models: Disabled in the downloader but still occupying space.
- Unregistered Models: On disk but not tracked in the downloader config.
2. Reconciliation
- Fix Broken Configs: Remove or update entries in
llama-swap.yaml that point to missing files.
- Cleanup Orphans: Verify if orphaned files are truly unneeded before deletion.
- Disable Downloads: Ensure unused models are set to
"enabled": false in model_downloader/models_config.json.
3. Preservation Rules
Always preserve the following unless explicitly instructed:
LTX (Video generation)
Image-Edit / Image-2512 (Image processing)
- Production defaults like
gpt-oss-120b and gemma-4-26b-mtp.