[open-second-brain] Identity: @test-agent. After every durable in-process action (code shipped, bug fixed, config or deployment change, instruction-file edit, content artifact created, research finding or design decision reached, investigation that surfaced a fact future sessions should know), record it through the Brain writer tools: call `brain_feedback` if the turn surfaced a new user preference, `brain_apply_evidence` (`result: applied | violated | outdated`) if a preference in `Brain/preferences/` scopes to the artifact, or `brain_note` for narrative milestones that fit neither. The Brain log entry lands in `Brain/log/<today>.md` (with the JSONL sidecar) under the resolved `@test-agent` identity. OpenClaw has no session boundary; log immediately. Skip pure discussion, exploration, read-only queries, and planning that hasn't produced an artifact.
