Function worker_degradation 

pub async fn worker_degradation(
    __arg0: State<AppState>,
    headers: HeaderMap,
    __arg2: Json<DegradationBody>,
) -> Result<StatusCode, ApiError>

POST /v1/worker/degradation (GH #32). Bearer = same short-handle / full-CapToken forms as worker_submit. Body = JSON, not raw bytes — this endpoint carries structured data, unlike its raw-bytes siblings.

Independent channel: appends a DegradationEntry to RunRecord.degradations via RunStore::append_degradation directly. Never touches OutputStore / the fold path (Crux invariant 2 — a degradation must not surface as step OUTPUT / $.step.parts).

Behavior:

• task_id is auto-looked-up server-side from the token/handle, same as worker_submit / worker_artifact.
• GH #37 terminal-run guard applies first — a degradation addressed at an already-terminal Run is rejected with 410 Gone ([reject_if_run_terminal]), same as a submit/artifact would be.
• step_ref / attempt / at are server-injected — step_ref is the fetching agent’s resolved name (AgentContextView.agent, the best proxy for Step.ref available at this layer), attempt is the task’s current attempt, at is now (Unix epoch seconds). The client body never supplies any of the three.
• No Run linkage in agent_ctx (a pre-run-tracking dispatch), an unparseable run_id, or an append_degradation call against a Run the store doesn’t actually hold (RunStoreError::NotFound — the same condition [reject_if_run_terminal] itself fails open on) all take the same silent 204 No Content path, logged via tracing::warn! — this is a legitimate no-tracking codepath, not a client error. Any other RunStore failure propagates as ApiError::engine.