openclaw-agent-optimize

OpenClaw Agent Optimization

Use this skill to tune an OpenClaw workspace for cost-aware routing, parallel-first delegation, and lean context.

Default posture

This skill is advisory first. It should produce:

• audit,
• options,
• recommended plan,
• exact patch proposal,
• rollback,
• verification plan.

No persistent mutations without explicit approval.

Quick start

1. Full audit (safe, no changes)

Audit my OpenClaw setup for cost, reliability, and context bloat. Output a prioritized plan with rollback notes. Do NOT apply changes.

2. Context bloat / transcript noise

My OpenClaw context is bloating (slow replies / high cost / lots of transcript noise). Identify the top offenders (tools, crons, bootstrap files, skills) and propose the smallest reversible fixes first. Do NOT apply changes.

3. Model routing / delegation posture

Propose a model routing plan for (a) coding/engineering, (b) short notifications/reminders, (c) reasoning-heavy research/writing. Include an exact config patch + rollback plan, but do NOT apply changes.

What good output looks like

• Executive summary
• Top drivers
  • cost
  • context
  • reliability
  • operator friction
• Options A/B/C with tradeoffs
• Recommended plan (smallest safe change first)
• Exact proposals + rollback + verify

Safety contract

• Do not mutate persistent settings without explicit approval.
• Do not create/update/remove cron jobs without explicit approval.
• If an optimization reduces monitoring coverage, present options and require choice.
• Before any approved change, show:
  1. exact change,
  2. expected impact,
  3. rollback plan,
  4. post-change verification.

High-ROI optimization levers

1) Output discipline for automation

Make maintenance loops truly silent on success.

2) Separate work from notification

If you want alerts but want interactive context lean:

• do the work quietly
• notify out-of-band with a short human receipt

3) Bootstrap discipline

Keep always-injected files short and load-bearing only. Move long runbooks into references/ or adjacent notes.

4) Ambient specialist surface reduction

A common hidden tax is too many always-visible specialist skills. If a workflow is low-frequency or specialist:

• prefer on-demand worker/subagent usage,
• do not keep it permanently ambient in main-chat prompt surface.

5) Measure optimizations authoritatively

Prefer fresh-session /context json or equivalent receipts over “feels better”. High-signal fields include:

• eligible skills
• skills.promptChars
• projectContextChars
• systemPrompt.chars
• promptTokens

6) Verification-first ops hygiene

After any approved optimization, verify:

• core chat still works
• recall/behavior did not degrade
• new session actually picks up the change
• rollback path is proven, not theoretical

Workflow (concise)

1. Audit rules + memory: keep restart-critical facts only.
2. Audit skill surface: trim ambient specialists before touching tool surface.
3. Audit transcripts/noise: silence cron and heartbeat success paths.
4. Audit model routing and delegation posture.
5. Recommend the smallest viable change first.
6. Verify on a new session when skill/bootstrap snapshotting exists.

Notes

• Some runtimes snapshot skills/config per session. If you install/update skills and do not see changes, start a new session.
• Prefer short SKILL.md + references/ for long runbooks.
• If context bloat is the main complaint, pair this skill with context-clean-up (audit-only).

References

• references/optimization-playbook.md
• references/model-selection.md
• references/context-management.md
• references/agent-orchestration.md
• references/cron-optimization.md
• references/heartbeat-optimization.md
• references/memory-patterns.md
• references/continuous-learning.md
• references/safeguards.md