When to use it

Use lessons-learned when you correct an agent’s behavior or diagnose a recurring workflow mistake. It helps you avoid explaining the same expectation in every conversation, while keeping the reason for that expectation visible.

How it works

Figure 01

Turn a correction into something you can reuse.

  1. Notice

    Identify a behavior correction or diagnosed workflow mistake.

  2. Match

    Find an existing lesson or create a new entry.

  3. Reinforce

    Update the record when the same lesson recurs.

Result.ai/memory/LESSONS.md
At three reinforcements, the skill can propose a project instruction. Applying that promotion requires approval of the exact change.

The skill records the correction in .ai/memory/LESSONS.md, compares it with existing entries by meaning, and updates a matching entry instead of creating a duplicate. It counts repeated evidence and shows you the entry it changed.

At three reinforcements, it can propose promoting the lesson into project instructions. It first checks that the lesson still applies, then requests approval for the exact instruction change.

An example request

Ask your agent

“You reported the deploy as successful before checking the service. Record the lesson: verify the health endpoint before claiming the deployment worked.”

This illustrates a request you can adapt; it is not a transcript of a completed run.

The correction is saved in an entry you can review and the agent can retrieve for similar work. You can also explicitly ask the agent to read or maintain the ledger.

Install and use

The skill’s identifier is lessons-learned. It ships in the session-memory plugin. In Claude Code, first add the Overclock marketplace, then install the package:

/plugin install session-memory@overclock

If you only want the lessons ledger, install learning-loop instead. It contains the same skill without handoffs or solutions. Choose one package; installing both duplicates the skill.

For a standalone installation, keep the skill directory and its bundled resources together. The repository also includes per-skill Codex metadata. Read the full skill instructions for its exact workflow and supporting files.

What to keep in mind

After installation, the host may select the skill when a matching correction occurs. This is model-based routing, not a guarantee that every lesson will be captured; ask explicitly when the correction matters.

The startup hook only reminds the agent that memory exists. It does not inject the ledger or rewrite your skills in the background. Requirement changes and one-off preferences do not automatically belong in this ledger.

  • Session handoff — Resume a long task without reconstructing the previous conversation.
  • Solutions — Save a verified fix so the same problem is easier to solve next time.
  • Project vocabulary — Keep the meaning of your project’s terms consistent as the code and team evolve.