run: wire SessionToolFactory + PostRun artifacts + AttachImages

The session-tool TYPES already lived in tool/ (P4 move) but the executor never
used them. This wires them, unblocking artifact-producing host surfaces (mort's
chat API / chatbot / .skill / scaddy) to run on executus:

- run/session.go: steerMailbox (thread-safe message queue) + runSession
  (tool.AgentSession over it: AttachImages → a user-role multimodal message
  injected before the agent's next step) + runPostRun (panic-isolated hook call).
- executor: create the mailbox + set inv.AttachImages BEFORE the toolbox build;
  add inv.ExtraTools + a SessionToolFactory's per-run Tools to the toolbox; defer
  its Cleanup; merge the session mailbox with the critic's nudges into ONE
  WithSteer; after the run, call PostRun with the full transcript
  (runRes.Messages) → Result.PostRunResult (best-effort, never fails the run).
- run.Result += PostRunResult *tool.PostRunResult.
- dropped the now-dead criticBinding.steerOptions (superseded by drainSteer).

Tests: a factory whose PostRun emits an artifact from the output+transcript +
Cleanup lands on Result.PostRunResult; a factory-added tool is callable.

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

run/session.go

@@ -0,0 +1,76 @@
+package run
+
+import (
+	"context"
+	"log/slog"
+	"strings"
+	"sync"
+
+	"gitea.stevedudenhoeffer.com/steve/majordomo/llm"
+
+	"gitea.stevedudenhoeffer.com/steve/executus/tool"
+)
+
+// runPostRun invokes a SessionToolFactory's PostRun hook with panic isolation:
+// a PostRun panic (or a slow artifact build that the hook mishandles) must not
+// fail an otherwise-successful run — artifacts are best-effort, the agent's text
+// output is the source of truth.
+func runPostRun(ctx context.Context,
+	hook func(context.Context, []llm.Message, string, error) *tool.PostRunResult,
+	transcript []llm.Message, output string, runErr error) (prr *tool.PostRunResult) {
+	defer func() {
+		if r := recover(); r != nil {
+			slog.Error("run: PostRun hook panicked; no artifacts produced", "panic", r)
+			prr = nil
+		}
+	}()
+	return hook(ctx, transcript, output, runErr)
+}
+
+// steerMailbox is a thread-safe queue of messages a session tool (via
+// tool.Invocation.AttachImages) wants injected into the agent loop before its
+// next step — the same WithSteer mechanism the critic uses for nudges, exposed
+// to ordinary tools so they can show the model content (e.g. a rendered
+// preview) it must SEE, not just be told about.
+type steerMailbox struct {
+	mu   sync.Mutex
+	msgs []llm.Message
+}
+
+func (m *steerMailbox) push(msg llm.Message) {
+	m.mu.Lock()
+	m.msgs = append(m.msgs, msg)
+	m.mu.Unlock()
+}
+
+// drain returns and clears the queued messages (nil when empty).
+func (m *steerMailbox) drain() []llm.Message {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	if len(m.msgs) == 0 {
+		return nil
+	}
+	out := m.msgs
+	m.msgs = nil
+	return out
+}
+
+// runSession implements tool.AgentSession over a steer mailbox: AttachImages
+// queues a user-role multimodal message the agent loop injects before its next
+// step. Replaces legacy agentkit's Agent.AttachImages — majordomo's *agent.Agent
+// is immutable mid-run, so mutation flows through the run-scoped steer mailbox.
+type runSession struct{ mailbox *steerMailbox }
+
+func (s *runSession) AttachImages(text string, images ...llm.ImagePart) {
+	parts := make([]llm.Part, 0, len(images)+1)
+	if strings.TrimSpace(text) != "" {
+		parts = append(parts, llm.Text(text))
+	}
+	for _, img := range images {
+		parts = append(parts, img)
+	}
+	if len(parts) == 0 {
+		return
+	}
+	s.mailbox.push(llm.UserParts(parts...))
+}