MoonshotAI · wbxl2000 · Aug 19, 2026 · Aug 19, 2026 · Aug 19, 2026 · Aug 19, 2026
diff --git a/.changeset/transcript-live-attachments-goal-clear.md b/.changeset/transcript-live-attachments-goal-clear.md
@@ -0,0 +1,5 @@
+---
+"@moonshot-ai/kimi-code": patch
+---
+
+Project prompt attachments into the live transcript (turn.started now carries session-media references) and clear the transcript goal when the goal is cleared.
diff --git a/packages/agent-core-v2/src/agent/loop/loopService.ts b/packages/agent-core-v2/src/agent/loop/loopService.ts
@@ -59,6 +59,7 @@ import {
   isDisplayablePromptOrigin,
   ThinkingDelta,
   ToolCallDelta,
+  turnPromptAttachments,
   turnPromptText,
   TurnStarted,
   TurnStepCompleted,
@@ -454,6 +455,7 @@ export class AgentLoopService extends Disposable implements IAgentLoopService {
         turnId: job.turn.id,
         origin,
         prompt: isDisplayablePromptOrigin(origin) ? turnPromptText(job.seed.input, origin) : undefined,
+        promptAttachments: turnPromptAttachments(job.seed.input),
       }),
     );
     void this.runTurn(job.turn, job.ready).then(job.result.resolve, job.result.reject);

diff --git a/packages/agent-core-v2/src/agent/loop/turnEvents.ts b/packages/agent-core-v2/src/agent/loop/turnEvents.ts
@@ -1,5 +1,6 @@
 /* oxlint-disable typescript-eslint/no-unsafe-declaration-merging, eslint-plugin-import/namespace -- Event2 class+payload-interface declaration merging is the sanctioned event-declaration idiom. */
 import type { PromptOrigin } from '#/agent/contextMemory/types';
+import { parseDaemonFileUrl } from '#/agent/media/mediaRef';
 import { Event2 } from '#/app/event/event2';
 import type { FinishReason } from '#/kosong/contract/provider';
 import type { ContentPart, TextPart } from '#/kosong/contract/message';
@@ -19,6 +20,7 @@ export interface TurnStartedPayload {
   readonly turnId: number;
   readonly origin: PromptOrigin;
   readonly prompt?: string;
+  readonly promptAttachments?: readonly { kind: 'image' | 'video' | 'audio'; fileId: string }[];
 }
 
 export class TurnStarted extends Event2<TurnStartedPayload> {
@@ -40,6 +42,32 @@ export function turnPromptText(
   return text.length > 0 ? text : undefined;
 }
 
+/** Media parts become the turn's transcript attachments only when they point
+ *  at a session upload — the id must match the part's daemon file URL (a
+ *  provider-issued id on a remote URL is not a session-media file id). */
+export function turnPromptAttachments(
+  input: readonly ContentPart[],
+): TurnStartedPayload['promptAttachments'] {
+  const attachments: { kind: 'image' | 'video' | 'audio'; fileId: string }[] = [];
+  const sessionMediaFileId = (url: string, id: string | undefined): string | undefined => {
+    if (id === undefined) return undefined;
+    return parseDaemonFileUrl(url)?.fileId === id ? id : undefined;
+  };
+  for (const part of input) {
+    if (part.type === 'image_url') {
+      const fileId = sessionMediaFileId(part.imageUrl.url, part.imageUrl.id);
+      if (fileId !== undefined) attachments.push({ kind: 'image', fileId });
+    } else if (part.type === 'video_url') {
+      const fileId = sessionMediaFileId(part.videoUrl.url, part.videoUrl.id);
+      if (fileId !== undefined) attachments.push({ kind: 'video', fileId });
+    } else if (part.type === 'audio_url') {
+      const fileId = sessionMediaFileId(part.audioUrl.url, part.audioUrl.id);
+      if (fileId !== undefined) attachments.push({ kind: 'audio', fileId });
+    }
+  }
+  return attachments.length > 0 ? attachments : undefined;
+}
+
 export function isDisplayablePromptOrigin(origin: PromptOrigin): boolean {
   if (origin.kind === 'user') return true;
   return (

diff --git a/packages/agent-core-v2/test/agent/loop/loop.test.ts b/packages/agent-core-v2/test/agent/loop/loop.test.ts
@@ -786,6 +786,43 @@ describe('Agent loop', () => {
 
     expect(prompts).toEqual([undefined, 'hi']);
   });
+
+  it('carries session-media prompt attachments on turn.started', async () => {
+    const payloads: Array<readonly { kind: string; fileId: string }[] | undefined> = [];
+    const subscription = ctx.get(IEventBus).subscribe(TurnStarted, (event) => {
+      payloads.push(event.promptAttachments);
+    });
+    ctx.mockNextResponse({ type: 'text', text: 'seen' });
+
+    const turn = (
+      await loop.enqueue(
+        new MessageStepRequest(
+          {
+            role: 'user',
+            content: [
+              { type: 'image_url', imageUrl: { url: 'kimi-file://file_1', id: 'file_1' } },
+              { type: 'video_url', videoUrl: { url: 'kimi-file://file_2', id: 'file_2' } },
+              { type: 'image_url', imageUrl: { url: 'https://example.com/no-id.png' } },
+              { type: 'image_url', imageUrl: { url: 'ms://provider-blob', id: 'prov_1' } },
+              { type: 'text', text: 'look' },
+            ],
+            toolCalls: [],
+            origin: { kind: 'user' },
+          },
+          { admission: 'newTurn' },
+        ),
+      ).assigned
+    ).turn;
+    await turn.result;
+    subscription.dispose();
+
+    expect(payloads).toEqual([
+      [
+        { kind: 'image', fileId: 'file_1' },
+        { kind: 'video', fileId: 'file_2' },
+      ],
+    ]);
+  });
 });
 
 describe('turn telemetry', () => {

diff --git a/packages/kap-server/src/lib/promptMedia.ts b/packages/kap-server/src/lib/promptMedia.ts
@@ -74,10 +74,10 @@ export function contentToCoreParts(content: WireContent): ContentPart[] {
     if (part.type === 'text') parts.push({ type: 'text', text: part.text });
     else if (part.type === 'image' && part.source.kind === 'url') parts.push({ type: 'image_url', imageUrl: { url: part.source.url, id: part.source.id } });
     else if (part.type === 'image' && part.source.kind === 'base64') parts.push({ type: 'image_url', imageUrl: { url: `data:${part.source.media_type};base64,${part.source.data}` } });
-    else if (part.type === 'image' && part.source.kind === 'session_media') parts.push({ type: 'image_url', imageUrl: { url: buildDaemonFileUrl(part.source.file_id) } });
+    else if (part.type === 'image' && part.source.kind === 'session_media') parts.push({ type: 'image_url', imageUrl: { url: buildDaemonFileUrl(part.source.file_id), id: part.source.file_id } });
     else if (part.type === 'video' && part.source.kind === 'url') parts.push({ type: 'video_url', videoUrl: { url: part.source.url, id: part.source.id } });
     else if (part.type === 'video' && part.source.kind === 'base64') parts.push({ type: 'video_url', videoUrl: { url: `data:${part.source.media_type};base64,${part.source.data}` } });
-    else if (part.type === 'video' && part.source.kind === 'session_media') parts.push({ type: 'video_url', videoUrl: { url: buildDaemonFileUrl(part.source.file_id) } });
+    else if (part.type === 'video' && part.source.kind === 'session_media') parts.push({ type: 'video_url', videoUrl: { url: buildDaemonFileUrl(part.source.file_id), id: part.source.file_id } });
   }
   return parts;
 }

diff --git a/packages/kap-server/src/protocol/events-zod.ts b/packages/kap-server/src/protocol/events-zod.ts
@@ -696,6 +696,9 @@ export const turnStartedEventSchema = z.object({
   origin: promptOriginSchema,
   prompt: z.string().optional(),
   promptId: z.string().optional(),
+  promptAttachments: z
+    .array(z.object({ kind: z.enum(['image', 'video', 'audio']), fileId: z.string() }))
+    .optional(),
 });
 
 export const turnEndedEventSchema = z.object({

diff --git a/packages/kap-server/src/services/transcript/coreEventMap.ts b/packages/kap-server/src/services/transcript/coreEventMap.ts
@@ -1081,7 +1081,9 @@ export class AgentTranscriptProjector {
   }): TranscriptOperation[] {
     const ops: TranscriptOperation[] = [];
     const snapshot = event.snapshot;
-    if (snapshot !== null) {
+    if (snapshot === null) {
+      ops.push({ op: 'meta.merge', meta: { goal: null } });
+    } else {
       ops.push({
         op: 'meta.merge',
         meta: {

diff --git a/packages/kap-server/test/prompts.test.ts b/packages/kap-server/test/prompts.test.ts
@@ -730,6 +730,7 @@ describe('server-v2 /api/v1 prompts', () => {
         type: 'image_url',
         imageUrl: {
           url: `kimi-file://${uploaded.id}`,
+          id: uploaded.id,
         },
       });
     });

diff --git a/packages/kap-server/test/services/transcript.test.ts b/packages/kap-server/test/services/transcript.test.ts
@@ -997,7 +997,9 @@ describe('AgentTranscriptProjector', () => {
     expect(marker).toMatchObject({ marker: 'goal', payload: { snapshot } });
 
     const clearedOps = projector.map(ev({ type: 'goal.updated', snapshot: null }));
-    expect(clearedOps.every((op) => op.op === 'marker.upsert')).toBe(true);
+    expect(clearedOps[0]).toEqual({ op: 'meta.merge', meta: { goal: null } });
+    tx.apply(clearedOps);
+    expect(tx.getMeta().goal).toBeUndefined();
   });
 
   it('mirrors plan / swarm mode slices into meta.modes (only when provided)', () => {

diff --git a/packages/protocol/src/events.ts b/packages/protocol/src/events.ts
@@ -682,6 +682,8 @@ export interface TurnStartedEvent {
   readonly prompt?: string;
   /** The prompt record id when the turn was opened by a prompt submission. */
   readonly promptId?: string;
+  /** Session-media references carried by the prompt (transcript attachments). */
+  readonly promptAttachments?: readonly { kind: 'image' | 'video' | 'audio'; fileId: string }[];
 }
 
 export interface TurnEndedEvent {
@@ -1647,6 +1649,9 @@ export const turnStartedEventSchema = z.object({
   origin: promptOriginSchema,
   prompt: z.string().optional(),
   promptId: z.string().optional(),
+  promptAttachments: z
+    .array(z.object({ kind: z.enum(['image', 'video', 'audio']), fileId: z.string() }))
+    .optional(),
 }) satisfies z.ZodType<TurnStartedEvent>;
 
 export const turnEndedEventSchema = z.object({

diff --git a/packages/transcript/src/contract/schema.ts b/packages/transcript/src/contract/schema.ts
@@ -322,7 +322,9 @@ export const transcriptMetaSchema = z.object({
   agent: agentStatusMetaSchema.optional(),
 });
 
+/** `goal` set to `null` in a merge clears the goal (same convention as mode keys). */
 export const transcriptMetaMergeSchema = transcriptMetaSchema.extend({
+  goal: goalMetaSchema.nullable().optional(),
   modes: modesMetaMergeSchema.optional(),
 });
 

diff --git a/packages/transcript/src/model/meta.ts b/packages/transcript/src/model/meta.ts
@@ -130,7 +130,8 @@ export interface TranscriptMeta {
   readonly agent?: AgentStatusMeta;
 }
 
-/** Contract shape of a `meta.merge` payload — like {@link TranscriptMeta}, but mode keys may be `null` to clear. */
-export type TranscriptMetaMerge = Omit<TranscriptMeta, 'modes'> & {
+/** Contract shape of a `meta.merge` payload — like {@link TranscriptMeta}, but mode keys and `goal` may be `null` to clear. */
+export type TranscriptMetaMerge = Omit<TranscriptMeta, 'modes' | 'goal'> & {
   readonly modes?: ModesMetaMerge;
+  readonly goal?: GoalMeta | null;
 };
diff --git a/packages/transcript/src/ops/apply.ts b/packages/transcript/src/ops/apply.ts
@@ -578,7 +578,7 @@ function applyMetaMerge(state: AgentState, meta: TranscriptMetaMerge): ApplyResu
   const agent =
     meta.agent !== undefined ? { ...state.meta.agent, ...meta.agent } : state.meta.agent;
   const next: TranscriptMeta = {
-    goal: meta.goal ?? state.meta.goal,
+    goal: meta.goal === null ? undefined : (meta.goal ?? state.meta.goal),
     activity: meta.activity ?? state.meta.activity,
     modes: modes !== undefined && modes.plan === undefined && modes.swarm === undefined ? undefined : modes,
     agent,