OpenClaw LESSONS.md Plugins

Drop-in hooks that give your OpenClaw agent swarm shared institutional memory via LESSONS.md.

Quick Install

npm install @jeremymandile/openclaw-lessons

Plugins

lesson-guard — Pre-Action Guard

Reads LESSONS.md before every non-trivial tool call. If a DO NOT entry matches the planned action (via tag overlap), the agent stops and substitutes the safe alternative. No human intervention required.

Hook: pre_tool_call

lesson-reporter — Post-Failure Reporter

When an action fails, formats a new DO NOT entry, appends it to LESSONS.md, and opens a PR. Every other agent in the fleet learns from the mistake automatically.

Hook: post_failure

lesson-digest — Weekly Compaction

Runs once a week to archive the oldest entries to LESSONS_ARCHIVE.md when the active file exceeds 100 entries. Keeps retrieval fast.

Hook: scheduled (recommended: Monday 9 AM)

Configuration

Add to your openclaw.yaml:

hooks:
  pre_tool_call:
    - name: lesson-guard
      command: "python ./node_modules/@jeremymandile/openclaw-lessons/lesson_guard.py"
  post_failure:
    - name: lesson-reporter
      command: "python ./node_modules/@jeremymandile/openclaw-lessons/lesson_reporter.py"
  scheduled:
    - name: lesson-digest
      command: "python ./node_modules/@jeremymandile/openclaw-lessons/lesson_digest.py"
      schedule: "0 9 * * 1"

Requirements

• Python 3.6+
• OpenClaw with plugin hook support
• A LESSONS.md at your project root — create one with:

  python tools/lesson_tool.py init-memory

Real-World Results

After six months of use, an OpenClaw agent fleet went from ~12 recurring mistake classes per month to zero of those same classes repeating. New mistakes still happen — but each one only happens once.

License

MIT