run: critic parity — fuller RecordStep + cause-carrying Kill (distinct status)

Completes the run-critic seam so a host adapter (mort's agentcritic) has full
fidelity, closing the two limitations gadfly surfaced on mort #1334.

- RecordStep(iter int, resp *llm.Response): the completed step's model response
  is now passed to the critic (was index-only), so a host that records a trace
  (mort's ProgressRecorder) can show what the agent actually produced, not just
  an iteration count. The executor forwards s.Response; the battery ignores it
  (its Progress is count-based).
- CriticHandle.KillCause() error + ErrCriticKill: the executor now distinguishes
  an explicit critic KILL from a natural backstop expiry. runCtx uses a
  cause-carrying cancel (WithCancelCause + a MaxRuntime timer cancelling with
  DeadlineExceeded); the deadline-watch cancels with ErrCriticKill when
  KillCause()!=nil, else DeadlineExceeded. statusFor reads context.Cause →
  killed / timeout / cancelled are now distinct (were all "cancelled"). The
  battery sets killCause from Decision.KillReason on a Kill.

Tests: statusFor "killed" case (cause=ErrCriticKill, err=Canceled); fake handle
+ battery RecordStep/KillCause signatures. Core stays battery-free.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

critic/critic.go

@@ -18,6 +18,7 @@ package critic
 
 import (
 	"context"
+	"errors"
 	"log/slog"
 	"math"
 	"sync"
@@ -162,12 +163,15 @@ type handle struct {
 	iterations   int
 	maxSteps     int // current tool-dispatch ceiling (base MaxIterations, raised by RaiseStepsBy)
 	lastTool     string
-	killed       bool // sticky: once an Escalator kills, no later decision un-kills it
+	killed       bool  // sticky: once an Escalator kills, no later decision un-kills it
+	killCause    error // non-nil once killed; surfaced via KillCause for "killed" status
 	stopped      bool
 	stopCh       chan struct{}
 }
 
-func (h *handle) RecordStep(iter int) {
+func (h *handle) RecordStep(iter int, _ *llm.Response) {
+	// This battery's Progress tracks iteration count + activity, not per-step
+	// payload, so the response is unused here; a richer Escalator could record it.
 	h.mu.Lock()
 	h.iterations = iter
 	h.lastActivity = h.now()
@@ -204,6 +208,12 @@ func (h *handle) MaxSteps() int {
 	return h.maxSteps
 }
 
+func (h *handle) KillCause() error {
+	h.mu.Lock()
+	defer h.mu.Unlock()
+	return h.killCause
+}
+
 func (h *handle) Stop() {
 	h.mu.Lock()
 	if !h.stopped {
@@ -266,8 +276,13 @@ func (h *handle) tick(ctx context.Context) {
 	}
 	if d.Kill {
 		h.killed = true
-		h.deadline = h.now() // immediate hard deadline → executor cancels
-		return               // ignore any Nudge/ExtendBy paired with a Kill
+		reason := d.KillReason
+		if reason == "" {
+			reason = "critic killed the run"
+		}
+		h.killCause = errors.New(reason) // surfaced via KillCause → "killed" status
+		h.deadline = h.now()             // immediate hard deadline → executor cancels
+		return                           // ignore any Nudge/ExtendBy paired with a Kill
 	}
 	if len(d.Nudge) > 0 {
 		h.steer = append(h.steer, d.Nudge...)

critic/critic_test.go

@@ -51,7 +51,7 @@ func TestMonitorEscalatesOncePerIdlePeriodAndExtends(t *testing.T) {
 		t.Error("deadline should have been extended past the original")
 	}
 	// A fresh step re-arms; another idle period escalates again.
-	h.RecordStep(1)
+	h.RecordStep(1, nil)
 	time.Sleep(60 * time.Millisecond)
 	mu.Lock()
 	c2 := calls