Experience Distiller

Distill finished work into the right OpenClaw knowledge layer. Use when a task, fix, setup, integration, cron/report workflow, repeated operation, or output-style decision has just finished and you need to decide what should be written to daily logs, the experience bank, a playbook, or a reusable skill. Triggers include requests like “沉淀一下”, “把这次经验记下来”, “应该写到哪里”, “提炼成经验”, “升级成 playbook/skill”, “这个经验记一下”, or any post-task knowledge routing decision.

Route completed work into the correct knowledge layer instead of dumping everything into one file.

Quick workflow

1. Read references/decision-rules.md.
2. Identify the finished task/result.
3. Separate:
   • dated facts/evidence
   • reusable action-level lessons
   • workflow-level changes
   • capability/package opportunities
4. Recommend one of:
   • daily-log
   • experience
   • playbook
   • skill
   • multi
   • no-op
5. If asked to execute, write the files directly.

Non-negotiables

• Do not store raw noise as long-term knowledge.
• Do not force everything into a skill.
• Prefer experience-bank for tactical reusable lessons.
• Prefer playbooks for canonical multi-step workflows.
• One task may write to multiple layers when justified.

OpenClaw default mapping

• memory/YYYY-MM-DD.md = dated facts and evidence
• memory/experience-bank/entries/ = trigger-action-failure reusable lessons
• playbooks/ = canonical workflows
• skills/ = reusable capability packages

Output pattern

Use a short recommendation block:

• route
• confidence
• why
• exact files to write/update
• draft content bullets

Bundled references

• references/decision-rules.md — routing logic
• references/template.md — lightweight invocation template
• references/examples.md — ready-to-use examples for common task types