@@ -7,7 +7,8 @@ import {AuthorityStoreProtocolError, canonicalAuthorityObject, canonicalAuthorit
   requireAuthorityStoreId} from "./authority_store_codec.ts";
 import {normalizeRegisteredTodoAgents, normalizeTodoAgent} from "./todo_agents.ts";
 import {canonicalTodoRecord} from "./todo_presentation.ts";
-import {TODO_OWNERSHIP_INTENT_FIELDS, planTodoAuthoringScope, TODO_AUTHORING_SCOPE_REQUEST_SCHEMA} from "../todos/authoring_scope.ts";
+import {planTodoAuthoringScope, TODO_AUTHORING_SCOPE_REQUEST_SCHEMA} from "../todos/authoring_scope.ts";
+import {shouldRecordTodoMutationActor} from "../todos/public_update.ts";
 import {normalizeNativePlanningIntent, planNativeTodoUpdate} from "../todos/native_update_plan.ts";
 import {decodeMonitorPollObservation, type MonitorPollObservation} from "../todos/monitor_metadata.ts";
 import {
@@ -156,10 +157,8 @@ export function prepareUpdatedTodo(
   // Preserve the public planner's legacy metadata semantics. Raw copy edits
   // already carry actor attribution, while planning-only updates historically
   // leave last_actor_agent_id untouched.
-  const rawCopyChanged = Object.entries(input.patch).some(([field, value]) =>
-    !Object.hasOwn(todo, field) || !canonicalAuthorityBytes(todo[field]).equals(canonicalAuthorityBytes(value))) ||
-    input.clear_fields.some(field => Object.hasOwn(todo, field));
-  if (rawCopyChanged || input.monitor_observation !== undefined || TODO_OWNERSHIP_INTENT_FIELDS.some(field => Object.hasOwn(input.planning_intent ?? {}, field))) {
+  if (shouldRecordTodoMutationActor(todo, input.patch, input.clear_fields,
+      input.planning_intent ?? {}, input.monitor_observation)) {
     next.last_actor_agent_id = input.actor_agent_id;
   }
   next.updated_at = input.now.toISOString().replace(/\.\d{3}Z$/u, "Z");
 
@@ -1000,6 +1000,11 @@ def _metadata_value_is_present(value: Any) -> bool:
             "claimed_by must be a public-safe agent token such as codex-main-control"
         ),
     ),
+    _TodoMetadataField(
+        "last_actor_agent_id",
+        normalize_todo_claimed_by,
+        invalid_message="last_actor_agent_id must be a public-safe agent token",
+    ),
     _TodoMetadataField(
         "bound_agent",
         normalize_todo_bound_agent,
@@ -1247,6 +1252,7 @@ def format_todo_metadata_line(
     decision_outcome: str | None = None,
     decision_scope_outcomes: Any = None,
     claimed_by: str | None = None,
+    last_actor_agent_id: str | None = None,
     bound_agent: str | None = None,
     goal_bound: bool | None = None,
     blocks_agent: str | None = None,
 
@@ -899,7 +899,7 @@ def _update_goal_todo_legacy(
             monitor_metadata=monitor_intent["metadata"],
             public_context={
                 "goal_id": goal_id, "role": target_role,
-                "actor_agent_id": effective_agent_id,
+                "actor_agent_id": mutation_authority.get("actor_agent_id"),
                 "registered_agents": registered_agent_ids_from_registry(registry_path, goal_id),
                 "monitor_observation": monitor_intent["observation"],
                 "enforce_monitor_boundedness": enforce_monitor_boundedness,
@@ -986,7 +986,7 @@ def _complete_goal_todo_legacy(
     state_file: Path | None = None,
     dry_run: bool = False,
 ) -> dict[str, Any]:
-    from .line_update import apply_todo_update_to_lines, link_generated_successor_todo_ids
+    from .line_update import apply_todo_update_to_lines, link_generated_successor_todo_ids, record_todo_mutation_actor
     from ..coordination.runtime_shadow_writer_adapter import begin_todo_runtime_shadow_capture, write_captured_todo_state, settle_todo_runtime_shadow_capture
 
     shadow_runtime_root = effective_runtime_root(registry_path, runtime_root_arg)
@@ -1186,6 +1186,7 @@ def _complete_goal_todo_legacy(
             successor_todo_ids=normalized_successor_todo_ids if successor_todo_ids is not None else None,
             updated_at=updated_at,
         )
+        record_todo_mutation_actor(lines, update_result, mutation_authority.get("actor_agent_id"))
         unblock_resume, decision_scope_resolution = (
             apply_completed_user_todo_lifecycle(
                 lines,
@@ -1319,7 +1320,7 @@ def _supersede_goal_todo_legacy(
     state_file: Path | None = None,
     dry_run: bool = False,
 ) -> dict[str, Any]:
-    from .line_update import apply_todo_update_to_lines, link_superseding_todo_id
+    from .line_update import apply_todo_update_to_lines, link_superseding_todo_id, record_todo_mutation_actor
     from ..coordination.runtime_shadow_writer_adapter import begin_todo_runtime_shadow_capture, write_captured_todo_state, settle_todo_runtime_shadow_capture
 
     if successor_todo_ids:
@@ -1387,6 +1388,7 @@ def _supersede_goal_todo_legacy(
             note="superseded",
             updated_at=updated_at,
         )
+        record_todo_mutation_actor(lines, update_result, mutation_authority.get("actor_agent_id"))
         registered_agents = registered_agent_ids_from_registry(registry_path, goal_id)
         successor_intents = build_successor_intents(
             next_agent_todo=next_agent_todo,
 
@@ -50,6 +50,32 @@ def upsert_todo_metadata(
     return True
 
 
+def record_todo_mutation_actor(
+    lines: list[str],
+    update_result: dict[str, Any],
+    actor: str | None,
+) -> None:
+    """Transport accepted mutation provenance inside the held writer transaction.
+
+    Authority remains with the existing lifecycle decision. An unchanged replay
+    must not mint a new actor or advance the source timestamp.
+    """
+    if not update_result.get("changed"):
+        return
+    if actor is None:
+        return
+    match = find_todo_block(
+        lines, todo_id=update_result["todo_id"], role=update_result["role"],
+    )
+    if match is None:
+        raise RuntimeError("Authorized Todo disappeared before provenance writeback")
+    block = match[4]
+    metadata_line = metadata_line_for_todo_block(block, {"last_actor_agent_id": actor})
+    updated = upsert_todo_metadata(lines, block, metadata_line)
+    update_result["metadata_updated"] = updated or update_result["metadata_updated"]
+    update_result["last_actor_agent_id"] = actor
+
+
 def link_generated_successor_todo_ids(
     lines: list[str],
     *,
@@ -152,6 +178,7 @@ def _field_update_plan(
                     **{key: block.get(key)
                     for key in (
                         "text",
+                        "note",
                         "todo_id",
                         "status",
                         "claimed_by",
@@ -336,6 +363,7 @@ def apply_todo_update_to_lines(
     normalized_status = plan["normalized_status"]
     target_status = plan["target_status"]
     updates = plan["metadata_updates"]
+    accepted_actor = updates.pop("last_actor_agent_id", None)
     text = updates.pop("text", text)
     updated_priority = updates.pop("priority", todo_priority_label(block))
     updates.pop("title", None)
@@ -354,11 +382,13 @@ def apply_todo_update_to_lines(
         metadata_line = metadata_line_for_todo_block(block, updates)
     metadata_updated = upsert_todo_metadata(lines, block, metadata_line)
     effective_metadata = parse_todo_metadata_line(metadata_line or "") or {}
-    return serialize_todo_update_result(
+    result = serialize_todo_update_result(
         role=resolved_role, section=section, todo=block.get("text"),
         todo_id=normalized_todo_id, status=target_status, priority=updated_priority,
         status_changed=status_changed, text_changed=text_changed,
         metadata_updated=metadata_updated, metadata=effective_metadata,
         monitor_poll_transition=plan.get("monitor_poll_transition"),
         external_wait_transition=plan.get("external_wait_transition"),
     )
+    record_todo_mutation_actor(lines, result, accepted_actor)
+    return result
@@ -15,9 +15,22 @@ import {
 } from "./decision_metadata.ts";
 
 import { normalizeMonitorConfiguration, validateMonitorConfigurationTarget } from "./monitor_metadata.ts";
+import {canonicalAuthorityBytes} from "../coordination/authority_store_codec.ts";
 
 export const TODO_PUBLIC_UPDATE_REQUEST_SCHEMA = "todo_public_update_request_v0";
 
+/** Shared provenance intent; callers still suppress unchanged replay and own admission. */
+export function shouldRecordTodoMutationActor(
+  todo: JsonObject, patch: JsonObject, clearFields: readonly string[],
+  planningIntent: JsonObject, monitorObservation: unknown,
+): boolean {
+  const rawCopyChanged = Object.entries(patch).some(([field, value]) =>
+    !Object.hasOwn(todo, field) || !canonicalAuthorityBytes(todo[field]).equals(canonicalAuthorityBytes(value))) ||
+    clearFields.some(field => Object.hasOwn(todo, field));
+  return rawCopyChanged || monitorObservation !== undefined ||
+    TODO_OWNERSHIP_INTENT_FIELDS.some(field => Object.hasOwn(planningIntent, field));
+}
+
 const SCOPE_INTENT_FIELDS = ["task_class", "status", "claimed_by", "bound_agent", "goal_bound",
   "blocks_agent", "clear_blocks_agent", "global_gate", "clear_global_gate", "excluded_agents",
   "task_repository", "task_domain", "resume_when", "clear_resume_when", "clear_claim"] as const;
@@ -100,5 +113,13 @@ export function planPublicTodoUpdate(value: unknown): JsonObject {
     monitor_context: {metadata: intent.monitor_metadata ?? null,
       observation: context.monitor_observation ?? null, role, task_class: scope.task_class,
       resume_when: scope.effective_resume_when, enforce_boundedness: context.enforce_monitor_boundedness}});
+  const copyPatch = Object.fromEntries(["text", "note"].flatMap(field =>
+    Object.hasOwn(intent, field) ? [[field, intent[field]]] : []));
+  if (context.actor_agent_id != null && shouldRecordTodoMutationActor(
+      todo, copyPatch, [], intent, context.monitor_observation ?? undefined)) {
+    // The legacy writer transports this after its material-change check; actor
+    // provenance alone must never turn an unchanged replay into a mutation.
+    plan.metadata_updates.last_actor_agent_id = context.actor_agent_id;
+  }
   return {...plan, ...(transition ? {external_wait_transition: transition} : {})};
 }