# Feedback and lesson

Create FDK with exact allowed prompt as body. Data/body context includes source/session, affected task/variant, evidence, candidate lesson, applicability, scope (project/personal), risk and confidence. Links identify source task/artifact and superseded lessons.

Example:

```json
{"type":"FDK","title":"Readable task labels","body":"Please make the task labels larger; I cannot read them comfortably.","data":{"lesson":"Use readable interface text and avoid tiny type.","scope":"project","confidence":"high","risk":"low","applicability":"Task board labels in this workspace"}}
```

Supply actual workspaceId through the writer. Read the returned status: bounded runtime recognition may apply a normalized preference; other lessons remain proposed. The standalone writer never claims automatic application. Personal scope and sensitive facts are not inferred from a single complaint.

Rollback keeps the old exact feedback/history and records which lesson supersedes which. The next brief should name the applied FDK if it actually influenced the work. “Captured” and “applied” are different outcomes.