First Experience Card

OME becomes valuable when the library contains lessons from your own real corrections. Create the first one through reflect and draft approval: the agent reflects, you approve or refine, then you confirm what enters the library.

1. Start a retrospective

Copy this to your agent:

Create an OME retrospective run for one reusable coding lesson:

1. Use the OME reflect flow to scan accessible coding-session sources deeply enough to find real user corrections, not just the last few messages.
2. Produce no more than 3 experience drafts.
3. Only keep lessons that would change a future agent action.
4. Give me only the draft approval page and a short summary, not JSON or internal files.

--focus is a lens, not a shortcut. Unless you explicitly limit the source set, the agent should still inspect all accessible session sources relevant to the focus.

2. Approve And Refine

Open the generated draft approval page. Ask:

• Would this situation happen again?
• Would seeing this card change the agent's next action?
• Is it specific enough to avoid broad keyword matches?
• Does it have clear ignore cases?

Reply in plain language:

Approve the first one.
The second is too broad; refine it with the boundary I just gave.
Merge these two.
Do not add it yet; here is a counterexample.

3. Confirm Library Add

When it looks right, tell the agent:

Add the approved experiences to the library.

Nothing is recalled in future tasks until you explicitly confirm adding it.

4. Verify Recall

Ask the agent to verify with a realistic future prompt, not the exact wording from the card:

Verify recall for the approved card with a realistic future task.
Tell me whether it matches too broadly or too narrowly.

The result should be precise. If a card matches too broadly, ask the agent to refine it before adding more cards.