/inference-optimizer

Production-oriented OpenClaw skill for auditing runtime health, cleaning stale session state, and applying safer inference tuning with approval.

What it does

inference-optimizer does not start with token tuning guesses. It checks the runtime first, then recommends cleanup or tuning only after the obvious operational failures are ruled out.

Audit order:

1. Gateway ownership and duplicate supervisors
2. Restart loops and failed services
3. Resolved openclaw binary path and install type
4. Updater status and allowlist coverage for the resolved path
5. Plugin provenance and unused local extensions
6. Context pressure, stale sessions, cache-trace, pruning, and concurrency

Core commands:

/preflight
/audit
/optimize
purge sessions

Install

clawhub install inference-optimizer

Manual install:

git clone https://github.com/vitalyis/inference-optimizer.git ~/clawd/skills/inference-optimizer
bash ~/clawd/skills/inference-optimizer/scripts/setup.sh
bash ~/clawd/skills/inference-optimizer/scripts/setup.sh --apply

Usage

/preflight              # backup + audit + setup preview
/audit                  # analyze-only runtime audit
/optimize               # audit, then propose approved actions
purge sessions          # archive stale sessions after approval

Operational checks:

• Treat warnings as signals, not proof.
• If updater output is partial or truncated, verify installed version, service status, and logs before claiming root cause.
• Resolve the real executable path before changing the allowlist.
• Use bash scripts/verify.sh after install; it now fails on legacy skills/public and repo-local workspace wiring.

For script details and safety constraints, see SECURITY.md.

Update

cd ~/clawd/skills/inference-optimizer && git pull

If installed elsewhere, pull from that skill directory instead.

Uninstall

rm -rf ~/clawd/skills/inference-optimizer

Safety

• /audit is analyze-only and should not mutate workspace files.
• purge-stale-sessions.sh archives by default. Use --delete only for intentional immediate removal.
• setup.sh previews by default. Use --apply only after reviewing changes.
• For allowlists, prefer exact executable paths or bounded NVM wildcards such as /home/ubuntu/.nvm/versions/node/*/bin/openclaw *.
• Do not rely on basename-only allowlist entries like openclaw.

Additional background and remediation notes live in SECURITY.md.

Contributing

Contributions are welcome. See CONTRIBUTING.md for workflow and release-note expectations.

License

MIT