diff --git a/crates/pi-walker/src/lib.rs b/crates/pi-walker/src/lib.rs
index b6c5d3c1f..6e2da0373 100644
--- a/crates/pi-walker/src/lib.rs
+++ b/crates/pi-walker/src/lib.rs
@@ -1444,10 +1444,7 @@ pub struct PreDescendDecision {
 #[derive(Clone, Debug, PartialEq, Eq)]
 enum DirectoryIdentity {
 	#[cfg(unix)]
-	Unix {
-		dev: u64,
-		ino: u64,
-	},
+	Unix { dev: u64, ino: u64 },
 	#[cfg(not(unix))]
 	Generic(std::path::PathBuf),
 }
diff --git a/packages/coding-agent/src/modes/controllers/event-controller.ts b/packages/coding-agent/src/modes/controllers/event-controller.ts
index 1842c4354..a623323ae 100644
--- a/packages/coding-agent/src/modes/controllers/event-controller.ts
+++ b/packages/coding-agent/src/modes/controllers/event-controller.ts
@@ -72,6 +72,11 @@ export class EventController {
 	// emits one read per completion — does not break it, so a run of consecutive
 	// reads collapses into one group even across completion boundaries.
 	#lastVisibleBlockCount = 0;
+	// Bumped on each assistant message_start. Scopes the placeholder key for a
+	// not-yet-id'd streamed tool block (see the toolCall loop) so a sealed,
+	// undeleted pending entry from an aborted message can never be reused by the
+	// next message's same content index.
+	#streamTurnNonce = 0;
 	#renderedCustomMessages = new Set<string>();
 	#lastIntent: string | undefined = undefined;
 	#backgroundToolCallIds = new Set<string>();
@@ -401,6 +406,7 @@ export class EventController {
 			this.ctx.ui.requestRender();
 		} else if (event.message.role === "assistant") {
 			this.#lastVisibleBlockCount = 0;
+			this.#streamTurnNonce++;
 			this.ctx.streamingComponent = createAssistantMessageComponent(this.ctx);
 			this.ctx.streamingMessage = event.message;
 			this.ctx.chatContainer.addChild(this.ctx.streamingComponent);
@@ -607,18 +613,14 @@ export class EventController {
 			if (this.ctx.streamingMessage.content.some(content => content.type === "toolCall")) {
 				this.ctx.streamingComponent.markTranscriptBlockFinalized();
 			}
-			for (const content of this.ctx.streamingMessage.content) {
+			for (const [contentIndex, content] of this.ctx.streamingMessage.content.entries()) {
 				if (content.type !== "toolCall") continue;
-				// Anthropic/OpenAI open a streamed tool block with an empty id (and
-				// `{}` args) before the id/arguments arrive; Gemini assembles the
-				// whole call first, so it never hits this. Keying `pendingTools` by
-				// "" would create a placeholder card, and the later real-id frame —
-				// `pendingTools.has(realId)` false — would create a SECOND card,
-				// orphaning the blank one (no `tool_execution_*` event ever carries
-				// "", so it is never matched, updated, or removed). Defer until the
-				// provider assigns the real id.
-				if (!content.id) continue;
 				if (content.name === "read") {
+					// Read groups key by the real id (one group spans several reads),
+					// and the owned-dialect parser delivers id + args together when
+					// the call closes, so a parseable target without an id never
+					// occurs here. Defer if it somehow does.
+					if (!content.id) continue;
 					if (!readArgsHaveTarget(content.arguments)) {
 						// Args still streaming — defer until path is parseable so we can route to the
 						// read group (regular files) vs ToolExecutionComponent (internal URLs).
@@ -641,6 +643,26 @@ export class EventController {
 					// Internal URL read falls through to ToolExecutionComponent below.
 				}
 
+				// The owned-dialect tool parser (text-based tool calls for OAuth
+				// Anthropic / OpenAI) appends a tool block when it detects the call
+				// opening and only fills `id` + arguments once the call's text
+				// closes — so the live preview must stream while `content.id` is "".
+				// Key the preview by stable content position until the id lands,
+				// then migrate the pending entry + reveal state onto the real id
+				// (so `tool_execution_*`, which always carries the real id, matches
+				// the same component instead of orphaning a blank card). Native
+				// structured tool calls — every Gemini call, and Anthropic/OpenAI
+				// function calls — carry the id from the first frame, so
+				// `pendingKey === content.id` throughout and the migration no-ops.
+				const streamKey = `\u0000stream:${this.#streamTurnNonce}:${contentIndex}`;
+				const pendingKey = content.id || streamKey;
+				if (content.id && this.ctx.pendingTools.has(streamKey)) {
+					const migrated = this.ctx.pendingTools.get(streamKey);
+					this.ctx.pendingTools.delete(streamKey);
+					if (migrated) this.ctx.pendingTools.set(pendingKey, migrated);
+					this.#toolArgsReveal.rekey(streamKey, pendingKey);
+				}
+
 				// Preserve the raw partial JSON only for renderers that need to surface fields before the JSON object closes.
 				// Bash uses this to show inline env assignments during streaming instead of popping them in at completion.
 				// While the JSON is still open, ToolArgsRevealController paces the
@@ -652,16 +674,16 @@ export class EventController {
 				const rawInput = content.customWireName !== undefined;
 				const tool = this.ctx.viewSession.getToolByName(content.name);
 				if (partialJson) {
-					renderArgs = this.#toolArgsReveal.setTarget(content.id, partialJson, {
+					renderArgs = this.#toolArgsReveal.setTarget(pendingKey, partialJson, {
 						rawInput,
 						exposeRawPartialJson: exposesRawPartialJson(content.name, rawInput, tool),
 						fullArgs: content.arguments,
 					});
 				} else {
-					this.#toolArgsReveal.finish(content.id);
+					this.#toolArgsReveal.finish(pendingKey);
 					renderArgs = content.arguments;
 				}
-				if (!this.ctx.pendingTools.has(content.id)) {
+				if (!this.ctx.pendingTools.has(pendingKey)) {
 					this.#resolveDisplaceablePoll(content.name);
 					this.#resetReadGroup();
 					const component = new ToolExecutionComponent(
@@ -680,13 +702,13 @@ export class EventController {
 					);
 					component.setExpanded(this.ctx.toolOutputExpanded);
 					this.ctx.chatContainer.addChild(component);
-					this.ctx.pendingTools.set(content.id, component);
-					this.#toolArgsReveal.bind(content.id, component);
+					this.ctx.pendingTools.set(pendingKey, component);
+					this.#toolArgsReveal.bind(pendingKey, component);
 				} else {
-					const component = this.ctx.pendingTools.get(content.id);
+					const component = this.ctx.pendingTools.get(pendingKey);
 					if (component) {
 						component.updateArgs(renderArgs, content.id);
-						this.#toolArgsReveal.bind(content.id, component);
+						this.#toolArgsReveal.bind(pendingKey, component);
 					}
 				}
 			}
diff --git a/packages/coding-agent/src/modes/controllers/tool-args-reveal.ts b/packages/coding-agent/src/modes/controllers/tool-args-reveal.ts
index d5e1d3516..6df7548ab 100644
--- a/packages/coding-agent/src/modes/controllers/tool-args-reveal.ts
+++ b/packages/coding-agent/src/modes/controllers/tool-args-reveal.ts
@@ -177,6 +177,18 @@ export class ToolArgsRevealController {
 		if (entry) entry.component = component;
 	}
 
+	/** Migrate a live reveal entry from a placeholder key onto the real
+	 *  tool-call id once the owned-dialect parser assigns it. No-op when no
+	 *  entry exists under `from` (smoothing disabled, or the JSON already
+	 *  closed and `finish` cleared it). */
+	rekey(from: string, to: string): void {
+		if (from === to) return;
+		const entry = this.#entries.get(from);
+		if (!entry) return;
+		this.#entries.delete(from);
+		this.#entries.set(to, entry);
+	}
+
 	/** Final arguments arrived (the JSON closed): drop the reveal so the
 	 *  caller's final-args render wins immediately, mirroring how assistant
 	 *  text snaps to the full message at message_end. */