@mengine/medeo-client 1.2.1-alpha.2 → 1.2.1-alpha.4

package/dist/schemas.d.ts

@@ -0,0 +1,2 @@
+import { $ as changeSpeechVoiceInputSchema, A as MoveVideoClipsInput, B as MoveSpeechesInput, C as setBgmInputSchema, D as replaceVideoClipSequenceInputSchema, E as anchoredReplacePolicySchema, F as moveAnchorSchema, G as deleteVideoClipsInputSchema, H as AnchoredDeletePolicy, I as moveVideoClipsByAnchorInputSchema, J as DeleteBgmInput, K as DeleteSpeechesInput, L as movedClipAnchoredPolicySchema, M as MoveAnchor, N as MoveVideoClipsByAnchorInput, O as ReplaceVideoClipContentInput, P as MovedClipAnchoredPolicy, Q as changeSpeechScriptInputSchema, R as MoveBgmInput, S as SetBgmInput, T as ReplaceVideoClipSequenceInput, U as DeleteVideoClipsInput, V as moveSpeechesInputSchema, W as anchoredDeletePolicySchema, X as ChangeSpeechScriptInput, Y as deleteBgmInputSchema, Z as ChangeSpeechVoiceInput, _ as setVideoClipSpeedShiftInputSchema, a as speechAssetsSchema, at as adjustSpeechVolumeInputSchema, b as SetCaptionStyleInput, c as mediaIdSchema, ct as AddVideoClipsInput, d as speechIdsSchema, dt as addSpeechesInputSchema, et as AdjustVideoClipVolumeInput, f as speedShiftSchema, g as SetVideoClipSpeedShiftInput, h as volumeSchema, i as SpeechAssets, it as AdjustSpeechVolumeInput, j as moveVideoClipsInputSchema, k as replaceVideoClipContentInputSchema, l as positiveMsSchema, lt as addVideoClipsInputSchema, m as voiceSchema, n as CaptionAsset, nt as AdjustVideoClipDurationInput, o as clipIdSchema, ot as AdjustBgmVolumeInput, p as timelineMsSchema, q as deleteSpeechesInputSchema, r as SpeechAsset, rt as adjustVideoClipDurationInputSchema, s as clipIdsSchema, st as adjustBgmVolumeInputSchema, tt as adjustVideoClipVolumeInputSchema, u as speechIdSchema, ut as AddSpeechesInput, v as SetCaptionVisibilityInput, w as AnchoredReplacePolicy, x as setCaptionStyleInputSchema, y as setCaptionVisibilityInputSchema, z as moveBgmInputSchema } from "./index-CiDSfSMi.js";
+export { AddSpeechesInput, AddVideoClipsInput, AdjustBgmVolumeInput, AdjustSpeechVolumeInput, AdjustVideoClipDurationInput, AdjustVideoClipVolumeInput, AnchoredDeletePolicy, AnchoredReplacePolicy, CaptionAsset, ChangeSpeechScriptInput, ChangeSpeechVoiceInput, DeleteBgmInput, DeleteSpeechesInput, DeleteVideoClipsInput, MoveAnchor, MoveBgmInput, MoveSpeechesInput, MoveVideoClipsByAnchorInput, MoveVideoClipsInput, MovedClipAnchoredPolicy, ReplaceVideoClipContentInput, ReplaceVideoClipSequenceInput, SetBgmInput, SetCaptionStyleInput, SetCaptionVisibilityInput, SetVideoClipSpeedShiftInput, SpeechAsset, SpeechAssets, addSpeechesInputSchema, addVideoClipsInputSchema, adjustBgmVolumeInputSchema, adjustSpeechVolumeInputSchema, adjustVideoClipDurationInputSchema, adjustVideoClipVolumeInputSchema, anchoredDeletePolicySchema, anchoredReplacePolicySchema, changeSpeechScriptInputSchema, changeSpeechVoiceInputSchema, clipIdSchema, clipIdsSchema, deleteBgmInputSchema, deleteSpeechesInputSchema, deleteVideoClipsInputSchema, mediaIdSchema, moveAnchorSchema, moveBgmInputSchema, moveSpeechesInputSchema, moveVideoClipsByAnchorInputSchema, moveVideoClipsInputSchema, movedClipAnchoredPolicySchema, positiveMsSchema, replaceVideoClipContentInputSchema, replaceVideoClipSequenceInputSchema, setBgmInputSchema, setCaptionStyleInputSchema, setCaptionVisibilityInputSchema, setVideoClipSpeedShiftInputSchema, speechAssetsSchema, speechIdSchema, speechIdsSchema, speedShiftSchema, timelineMsSchema, voiceSchema, volumeSchema };

package/dist/schemas.js

@@ -0,0 +1,541 @@
+import { t as __exportAll } from "./chunk-D7D4PA-g.js";
+import { z } from "zod";
+//#region src/editor/schemas/shared.ts
+const clipIdSchema = z.string().min(1).describe("The clip part ID on the timeline");
+const clipIdsSchema = z.array(clipIdSchema).min(1).refine((ids) => new Set(ids).size === ids.length, { message: "Duplicate clip IDs are not allowed" }).describe("List of clip part IDs (no duplicates allowed)");
+const mediaIdSchema = z.string().min(1).describe("The media asset ID");
+const speechIdSchema = z.string().min(1).describe("The speech part ID on the timeline");
+const timelineMsSchema = z.number().int().min(0).describe("Time position in milliseconds on the timeline (>= 0)");
+const positiveMsSchema = z.number().int().positive().describe("Duration in milliseconds (> 0)");
+const volumeSchema = z.number().min(-60).max(20).describe("Volume in decibels (-60.0 to 20.0; 0.0 = original, -60 = mute, +20 = max)");
+const speechIdsSchema = z.array(speechIdSchema).min(1).refine((ids) => new Set(ids).size === ids.length, { message: "Duplicate speech IDs are not allowed" }).describe("List of speech part IDs (no duplicates allowed)");
+const tangentHandleSchema = z.object({
+	x: z.number().finite(),
+	y: z.number().finite()
+}).describe("Bezier tangent handle (x, y)");
+const speedKeyframeSchema = z.object({
+	position: z.number().min(0).max(1),
+	rate: z.number().min(0),
+	in_tangent: tangentHandleSchema.optional(),
+	out_tangent: tangentHandleSchema.optional()
+}).describe("A speed keyframe: normalized position (0..1), rate, optional tangents");
+/**
+* A clip's playback-speed fact, the only thing `SetVideoClipSpeedShift` writes.
+* Mirrors the IDL `SpeedShift`: `category` is `linear` | `curve`, and `config`
+* is a discriminated union — `{ linear: { speed } }` for a constant multiplier
+* (the multiplier projection reads at `config.linear.speed`) or `{ curve: {
+* keyframes } }` for a Bezier-controlled variable speed (RFC 02 / `reference/16`
+* §0). Exactly one of `linear` / `curve` is present.
+*/
+const speedShiftSchema = z.object({
+	category: z.enum(["linear", "curve"]),
+	mode: z.string(),
+	config: z.union([z.object({ linear: z.object({ speed: z.number().finite().positive() }) }), z.object({ curve: z.object({ keyframes: z.array(speedKeyframeSchema).min(2) }) })])
+}).describe("Speed shift: linear multiplier or Bezier curve, mirroring the IDL shape");
+const voiceSchema = z.object({
+	id: z.string().min(1),
+	name: z.string()
+}).describe("TTS voice summary attached to a speech");
+//#endregion
+//#region src/editor/schemas/speech-assets.ts
+/**
+* The materialized TTS result shared by `AddSpeeches` / `ChangeSpeechScript` /
+* `ChangeSpeechVoice` (see `results/phase-4-side-effect-payload-contract.md`
+* §1/§2). The side effect (TTS/ASR + billing) runs upstream; the op receives the
+* stable speech + caption parts and writes them as authoritative facts. No
+* cascade runs on write — the projection derives absolute positions on read.
+*
+* `anchor_part_id` states the root relationship explicitly: a video/BGM id
+* creates an anchored speech; `null` creates an absolute audio root.
+* `offset_ms` is relative to the anchor or absolute for a root. Captions always
+* anchor to their speech via `start_ms`.
+*/
+const speechAssetSchema = z.object({
+	speech_id: speechIdSchema.describe("The speech part ID (= side-effect speech_parts[].id)"),
+	anchor_part_id: z.string().min(1).nullable().describe("Video/BGM parent part ID, or null when the speech itself is the root"),
+	offset_ms: timelineMsSchema.describe("Relative offset from the parent, or absolute start for a root speech"),
+	audio_storage_key: z.string().min(1),
+	duration_ms: positiveMsSchema,
+	audio_script: z.string(),
+	volume: volumeSchema,
+	voice: voiceSchema.optional(),
+	origin_speech_id: z.string().min(1),
+	caption_ids: z.array(z.string().min(1)).describe("Caption part IDs owned by this speech")
+}).superRefine((speech, ctx) => {
+	if (speech.audio_storage_key === speech.speech_id || speech.audio_storage_key === speech.origin_speech_id) ctx.addIssue({
+		code: z.ZodIssueCode.custom,
+		path: ["audio_storage_key"],
+		message: "audio_storage_key must be the materialized storage key, not a speech id"
+	});
+});
+const captionAssetSchema = z.object({
+	caption_id: z.string().min(1).describe("The caption part ID (= side-effect created_caption_parts[].id)"),
+	speech_part_id: speechIdSchema.describe("The owning speech part ID"),
+	text: z.string(),
+	start_ms: timelineMsSchema.describe("Offset within the host speech (caption.abs = speech.abs + start_ms)"),
+	duration_ms: positiveMsSchema
+});
+/** A materialized speech-subtree write (speeches + their captions). */
+const speechAssetsSchema = z.object({
+	speeches: z.array(speechAssetSchema).min(1).describe("Materialized speech parts to write"),
+	captions: z.array(captionAssetSchema).describe("Materialized caption parts owned by the speeches")
+}).superRefine((assets, ctx) => {
+	const ownerByStorageKey = /* @__PURE__ */ new Map();
+	for (const [index, speech] of assets.speeches.entries()) {
+		const owner = ownerByStorageKey.get(speech.audio_storage_key);
+		if (owner != null && owner !== speech.speech_id) ctx.addIssue({
+			code: z.ZodIssueCode.custom,
+			path: [
+				"speeches",
+				index,
+				"audio_storage_key"
+			],
+			message: "one materialized audio_storage_key must belong to one speech part"
+		});
+		ownerByStorageKey.set(speech.audio_storage_key, speech.speech_id);
+	}
+});
+//#endregion
+//#region src/editor/schemas/add-speeches.ts
+/**
+* Add speeches (and their captions). TTS runs upstream; the stable speech /
+* caption parts arrive materialized (see `speech-assets.ts`). The op writes the
+* parts and each speech's `{ mode:'anchored', anchorPartId, offsetMs }` fact
+* verbatim — no write-time host-picking, no cascade (RFC 02 §4). The projection
+* derives absolute positions on read.
+*/
+const addSpeechesInputSchema = speechAssetsSchema.describe("Materialized speeches + captions to add");
+//#endregion
+//#region src/editor/schemas/add-video-clips.ts
+/**
+* Add video clips to a track. Each clip's duration facts are separated so a
+* single number is never overloaded (RFC 02 / `reference/16` §0b):
+*
+* - `media_duration_ms` is the source media's intrinsic full length (a resource
+*   fact, written to the part);
+* - `play_in` / `play_out` are the optional trim window into that media; when
+*   omitted the whole media is used (`play_in=0`, `play_out=media_duration_ms`).
+*
+* The clip's effective timeline duration is derived by the projection from the
+* trim window and `speed_shift` — it is never an input here.
+*/
+const addVideoClipsInputSchema = z.object({
+	clips: z.array(z.object({
+		media_id: mediaIdSchema.describe("The media asset ID for the video clip"),
+		start_ms: timelineMsSchema.optional().describe("Absolute start time in milliseconds on the timeline"),
+		media_duration_ms: positiveMsSchema.describe("The source media's intrinsic full length in ms"),
+		play_in: timelineMsSchema.optional().describe("Trim window start in the media (default 0)"),
+		play_out: positiveMsSchema.optional().describe("Trim window end in the media (default media_duration_ms)"),
+		track_id: z.string().min(1).optional().describe("Target track ID (optional, defaults to main track)")
+	})).min(1).describe("List of video clips to create"),
+	before_clip_id: z.string().min(1).optional().describe("Insert new clips before this clip ID"),
+	after_clip_id: z.string().min(1).optional().describe("Insert new clips after this clip ID")
+}).superRefine((data, ctx) => {
+	if (data.before_clip_id != null && data.after_clip_id != null) {
+		ctx.addIssue({
+			code: z.ZodIssueCode.custom,
+			message: "Cannot provide both before_clip_id and after_clip_id"
+		});
+		return;
+	}
+	const hasRelative = data.before_clip_id != null || data.after_clip_id != null;
+	for (let i = 0; i < data.clips.length; i++) {
+		const clip = data.clips[i];
+		if (hasRelative && clip.start_ms != null) ctx.addIssue({
+			code: z.ZodIssueCode.custom,
+			message: `clips[${i}].start_ms must not be provided when using before_clip_id or after_clip_id`,
+			path: [
+				"clips",
+				i,
+				"start_ms"
+			]
+		});
+		if (!hasRelative && clip.start_ms == null) ctx.addIssue({
+			code: z.ZodIssueCode.custom,
+			message: `clips[${i}].start_ms is required when not using relative positioning`,
+			path: [
+				"clips",
+				i,
+				"start_ms"
+			]
+		});
+		const playIn = clip.play_in ?? 0;
+		const playOut = clip.play_out ?? clip.media_duration_ms;
+		if (playOut > clip.media_duration_ms) ctx.addIssue({
+			code: z.ZodIssueCode.custom,
+			message: `clips[${i}].play_out ${playOut}ms exceeds media_duration_ms ${clip.media_duration_ms}ms`,
+			path: [
+				"clips",
+				i,
+				"play_out"
+			]
+		});
+		if (playIn >= playOut) ctx.addIssue({
+			code: z.ZodIssueCode.custom,
+			message: `clips[${i}].play_in ${playIn}ms must be less than play_out ${playOut}ms`,
+			path: [
+				"clips",
+				i,
+				"play_in"
+			]
+		});
+	}
+});
+//#endregion
+//#region src/editor/schemas/adjust-bgm-volume.ts
+const adjustBgmVolumeInputSchema = z.object({ bgm: z.array(z.object({
+	bgm_id: clipIdSchema.describe("The bgm part ID to adjust volume for"),
+	volume: volumeSchema.describe("Volume in decibels (-60.0 to 20.0; 0.0 = original)")
+})).min(1).describe("List of bgm parts with their new volume settings") });
+//#endregion
+//#region src/editor/schemas/adjust-speech-volume.ts
+const adjustSpeechVolumeInputSchema = z.object({ speeches: z.array(z.object({
+	speech_id: speechIdSchema.describe("The speech part ID to adjust volume for"),
+	volume: volumeSchema.describe("Volume in decibels (-60.0 to 20.0; 0.0 = original)")
+})).min(1).describe("List of speeches with their new volume settings") });
+//#endregion
+//#region src/editor/schemas/adjust-video-clip-duration.ts
+/**
+* Re-trim existing video clips (the user-facing "adjust duration" gesture is a
+* trim of the source window). The new `play_in` / `play_out` are the facts; the
+* effective timeline duration is derived from them and the clip's `speed_shift`,
+* and the change reflows downstream clips, speeches, and the timeline inside the
+* op's transaction (no caller-materialized cascade).
+*/
+const adjustVideoClipDurationInputSchema = z.object({ clips: z.array(z.object({
+	clip_id: clipIdSchema.describe("The video clip part ID to re-trim"),
+	play_in: timelineMsSchema.describe("New trim window start in the source media"),
+	play_out: positiveMsSchema.describe("New trim window end in the source media")
+})).min(1).describe("Video clips with their new trim windows") });
+//#endregion
+//#region src/editor/schemas/adjust-video-clip-volume.ts
+const adjustVideoClipVolumeInputSchema = z.object({ clips: z.array(z.object({
+	clip_id: clipIdSchema.describe("The video clip part ID to adjust volume for"),
+	volume: volumeSchema.describe("Volume in decibels (-60.0 to 20.0; 0.0 = original)")
+})).min(1).describe("List of video clips with their new volume settings") });
+//#endregion
+//#region src/editor/schemas/change-speech.ts
+/**
+* Change a speech's script or voice. Both re-run TTS upstream and return the
+* regenerated speech / caption parts in the same materialized shape as
+* `AddSpeeches` (`speech-assets.ts`); the op upserts them by id (the speech part
+* id is preserved across a re-TTS), re-seats at `start_ms`, and reflows. Old
+* caption parts no longer owned by the speech are removed via `caption_ids`.
+*/
+const changeSpeechScriptInputSchema = speechAssetsSchema.describe("Regenerated speeches + captions (new script)");
+const changeSpeechVoiceInputSchema = speechAssetsSchema.describe("Regenerated speeches + captions (new voice)");
+//#endregion
+//#region src/editor/schemas/delete-bgm.ts
+/**
+* Remove the document BGM. Pure document edit: clears the bgm lane and removes
+* the bgm part. Takes no input (a document holds at most one bgm); an empty
+* object keeps the op signature uniform with the rest.
+*/
+const deleteBgmInputSchema = z.object({}).describe("Remove the document BGM (no parameters)");
+//#endregion
+//#region src/editor/schemas/delete-speeches.ts
+/**
+* Delete speeches with their captions. Pure document edit (no side effect): the
+* op removes each speech part, cascade-deletes the captions it owns (via
+* `caption_ids` / `speech_part_id`), drops their track items, and reflows.
+*/
+const deleteSpeechesInputSchema = z.object({ speech_ids: speechIdsSchema.describe("Speech part IDs to delete (their captions cascade-delete)") });
+//#endregion
+//#region src/editor/schemas/delete-video-clips.ts
+/**
+* How a delete handles the anchored subtree (speeches anchored to a deleted clip,
+* and their captions) — a delete-op policy, not a data-model field (reference/17
+* §6). `cascade` (default) removes the subtree; `detach` keeps the direct
+* anchored children, re-pinning them to `absolute` so they stay on the timeline.
+*/
+const anchoredDeletePolicySchema = z.enum(["cascade", "detach"]);
+const deleteVideoClipsInputSchema = z.object({
+	clip_ids: clipIdsSchema.describe("List of video clip part IDs to delete from the main track"),
+	on_anchored: anchoredDeletePolicySchema.optional().describe("How to treat anchored children (default cascade)")
+});
+//#endregion
+//#region src/editor/schemas/move-speeches.ts
+/**
+* Move speeches in time. Pure document edit: the op re-seats each speech at its
+* new absolute `start_ms`; the cascade reassigns it to the host video clip,
+* resolves overlaps, and reflows. Captions follow their speech.
+*/
+const moveSpeechesInputSchema = z.object({ speeches: z.array(z.object({
+	speech_id: speechIdSchema.describe("The speech part ID to move"),
+	new_start_ms: timelineMsSchema.describe("New absolute start time on the timeline")
+})).min(1).describe("Speeches to move to new positions") });
+//#endregion
+//#region src/editor/schemas/move-bgm.ts
+const moveBgmInputSchema = z.object({ new_start_ms: timelineMsSchema.describe("New absolute BGM start on the timeline") });
+//#endregion
+//#region src/editor/schemas/move-video-clips-by-anchor.ts
+/**
+* Where the moved block lands on the main track.
+*
+* A discriminated union rather than two optional `before_clip_id` /
+* `after_clip_id` fields (the shape `addVideoClips` had to use, because there the
+* two modes share a whole clip description): here the alternatives carry nothing
+* in common, so making them mutually exclusive *by type* removes three runtime
+* `superRefine` checks that would otherwise have to be written and tested.
+*
+* `track_start` is an explicit member, not the absence of an anchor. The
+* agent-harness mutation this maps from treats "neither anchor given" as
+* "move to the front" (`applyBatchMoveVideoClips` falls back to `insertIndex = 0`),
+* which is a default buried in a tool description. Requiring the caller to name
+* that intent keeps a forgotten field from silently reordering the timeline.
+*/
+const moveAnchorSchema = z.discriminatedUnion("position", [
+	z.object({
+		position: z.literal("before"),
+		clip_id: clipIdSchema.describe("The moved block lands immediately before this clip")
+	}),
+	z.object({
+		position: z.literal("after"),
+		clip_id: clipIdSchema.describe("The moved block lands immediately after this clip")
+	}),
+	z.object({ position: z.literal("track_start") })
+]).describe("Where the moved block lands: before/after a reference clip, or at the head of the track");
+/**
+* What happens to the speeches anchored to the clips being moved.
+*
+* - `follow` keeps each speech anchored where it is, so it travels with its clip
+*   to the new position. In the anchored model this is the *no-op* branch: a
+*   speech's authoritative fact is `{ anchorPartId, offsetMs }` and its absolute
+*   time is derived on read from the host's position, so moving the host moves
+*   the speech with no write to the speech at all.
+* - `keep_absolute` preserves each speech's current absolute landing instead, then
+*   re-anchors it to whichever clip now covers that time (RFC 02 §9.1/§11.1). This
+*   is the branch that costs an extra pass, and the one the FE timeline uses.
+*
+* **Required, with no default**, matching `deleteVideoClips` and
+* `replaceVideoClipSequence`. The two branches decide which picture the user's
+* narration ends up over, which is too consequential to infer from a missing
+* field — and neither branch is "safe enough" to be the implicit one.
+*/
+const movedClipAnchoredPolicySchema = z.enum(["follow", "keep_absolute"]);
+/**
+* Reorder a set of main-track clips relative to a reference clip.
+*
+* Distinct from `moveVideoClips`, which positions clips by absolute time
+* (`new_start_ms`) and is what the FE timeline dispatches after a drag. Main-track
+* clips are `sequential`-positioned, so absolute time is not authoritative state
+* there (RFC 02 §4/§7): `moveVideoClips` has to *guess* an index back out of the
+* time it was handed, whereas an anchor already is the ordinal fact being changed.
+* Keeping them separate also isolates blast radius — this method can diverge on
+* speech policy without touching the FE path.
+*
+* `clip_ids` need not be contiguous. They move as one block, keeping their
+* relative order, which is the agent-harness `batch_move_video_clips` contract.
+*/
+const moveVideoClipsByAnchorInputSchema = z.object({
+	clip_ids: clipIdsSchema.describe("Clips to move as one block, keeping their relative order. Need not be contiguous on the track."),
+	anchor: moveAnchorSchema,
+	on_anchored: movedClipAnchoredPolicySchema.describe("What happens to speeches anchored to the moved clips (required — see the policy doc)")
+});
+//#endregion
+//#region src/editor/schemas/move-video-clips.ts
+const moveVideoClipsInputSchema = z.object({ clips: z.array(z.object({
+	clip_id: clipIdSchema.describe("The video clip part ID to move"),
+	new_start_ms: timelineMsSchema.describe("New absolute start time in milliseconds on the timeline"),
+	new_track_id: z.string().min(1).optional().describe("Target track ID to move the clip to (optional)")
+})).min(1).describe("List of video clips to move to new positions") });
+//#endregion
+//#region src/editor/schemas/replace-video-clip-content.ts
+/**
+* Replace the media backing existing video clips. The media import runs upstream
+* (Director); its stable result — the new media id, intrinsic length, and the
+* reset trim window — arrives materialized (see
+* `results/phase-4-side-effect-payload-contract.md` §4). Director resets
+* `play_in=0` / `play_out=media_duration_ms` and clears `speed_shift` on
+* replacement. The clip `part_id`s (hence their track items) are unchanged; the
+* editor reflows the main track from the new effective durations.
+*/
+const replaceVideoClipContentInputSchema = z.object({ clips: z.array(z.object({
+	clip_id: clipIdSchema.describe("Existing video clip part ID to re-point"),
+	origin_media_id: mediaIdSchema.describe("The new media asset ID"),
+	media_duration_ms: positiveMsSchema.describe("The new media's intrinsic full length"),
+	play_in: timelineMsSchema.describe("Trim window start in the new media (usually 0)"),
+	play_out: positiveMsSchema.describe("Trim window end in the new media (usually = media_duration_ms)"),
+	volume: volumeSchema
+})).min(1).describe("Video clips whose media is being replaced") });
+//#endregion
+//#region src/editor/schemas/replace-video-clip-sequence.ts
+/**
+* What happens to the speeches anchored to the clips being replaced.
+*
+* - `remap` re-anchors each surviving speech to the new clip in the SAME POSITION
+*   of the sequence, keeping its offset — old[i]'s children become new[i]'s
+*   children. An old clip with no counterpart (fewer new clips than old) has its
+*   subtree deleted, because there is nothing left to anchor to.
+* - `cascade` deletes every anchored speech (and its captions) outright, like
+*   `deleteVideoClips`.
+*
+* **Required, with no default.** The two branches differ in whether the user's
+* narration survives, and the agent tool that drives this op makes its
+* `preserve_speeches` flag required for that reason. A default here would let a
+* caller that forgot the field silently delete speech.
+*/
+const anchoredReplacePolicySchema = z.enum(["remap", "cascade"]);
+/**
+* Replace a contiguous run of main-track clips with a new run.
+*
+* A composite of delete + insert that cannot be expressed as the two ops in
+* sequence, because the anchored speeches have to survive *across* the swap: with
+* `remap` they are re-anchored positionally, which needs both the old and the new
+* ids in the same transaction (ADR 0009 — the cascade stays in the editor, callers
+* never re-wire anchors themselves).
+*
+* Duration facts follow `addVideoClips`: `media_duration_ms` is the source's
+* intrinsic length and the trim window defaults to the whole media. The effective
+* timeline duration is derived by the projection, never an input.
+*
+* `media_id` is optional: omitting it creates a **deliberate empty placeholder
+* clip** (`origin_media_id: ''`) — structure with no picture. This is the only op
+* that can produce one, and it is authoritative state, unlike the gap fillers the
+* read-side solve mints (which never enter the document).
+*/
+const replaceVideoClipSequenceInputSchema = z.object({
+	old_clip_ids: clipIdsSchema.describe("The clips being replaced: a contiguous main-track run, listed in timeline order"),
+	new_clips: z.array(z.object({
+		media_id: mediaIdSchema.optional().describe("The replacement media asset ID. Omit to create an empty placeholder clip."),
+		media_duration_ms: positiveMsSchema.describe("The source media's intrinsic full length in ms"),
+		play_in: timelineMsSchema.optional().describe("Trim window start in the media (default 0)"),
+		play_out: positiveMsSchema.optional().describe("Trim window end in the media (default media_duration_ms)")
+	})).min(1).describe("The replacement clips, in the order they take on the track"),
+	on_anchored: anchoredReplacePolicySchema.describe("What happens to speeches anchored to the replaced clips (required — see the policy doc)")
+}).superRefine((data, ctx) => {
+	for (let i = 0; i < data.new_clips.length; i++) {
+		const clip = data.new_clips[i];
+		const playIn = clip.play_in ?? 0;
+		const playOut = clip.play_out ?? clip.media_duration_ms;
+		if (playOut > clip.media_duration_ms) ctx.addIssue({
+			code: z.ZodIssueCode.custom,
+			message: `new_clips[${i}].play_out ${playOut}ms exceeds media_duration_ms ${clip.media_duration_ms}ms`,
+			path: [
+				"new_clips",
+				i,
+				"play_out"
+			]
+		});
+		if (playIn >= playOut) ctx.addIssue({
+			code: z.ZodIssueCode.custom,
+			message: `new_clips[${i}].play_in ${playIn}ms must be less than play_out ${playOut}ms`,
+			path: [
+				"new_clips",
+				i,
+				"play_in"
+			]
+		});
+	}
+});
+//#endregion
+//#region src/editor/schemas/set-bgm.ts
+/**
+* Set the document BGM. The media's stable result (storage key) arrives
+* materialized from upstream (see
+* `results/phase-4-side-effect-payload-contract.md` §3). The op upserts the bgm
+* part and seats it on the bgm lane; its effective length is always the whole
+* timeline, derived by the projection on read — so there is no `duration_ms`
+* input or fact (RFC 02 / `reference/16` §0b). A `bgm_id` lets the op replace an
+* existing bgm part by id.
+*/
+const setBgmInputSchema = z.object({
+	bgm_id: z.string().min(1).describe("The bgm part ID to write"),
+	audio_storage_key: z.string().min(1),
+	origin_media_id: mediaIdSchema,
+	volume: volumeSchema
+});
+//#endregion
+//#region src/editor/schemas/set-caption-style.ts
+/**
+* Set the caption visual style. GLOBAL by design: the style applies to every
+* caption part in the document — it carries NO `caption_id`. This mirrors the FE,
+* whose caption-style store (`caption-style.ts:persistCaptionStylePatch`) iterates
+* ALL captions and writes the same normalized style to each; the product has a
+* single document-wide caption style, not per-caption styling.
+*
+* Every field is optional and maps to a `CaptionStyle` attribute (snake_case
+* IDL). A field present in the input is written to every caption; a field ABSENT
+* from the input is left untouched on each caption (the editor merges the patch
+* onto each caption's existing style — this is a value edit, not a full-style
+* replace, so a partial patch such as "recolor only" does not wipe font size).
+*
+* Pure document edit, no cascade — captions keep their positions; only the style
+* sub-map of each caption part changes.
+*/
+const setCaptionStyleInputSchema = z.object({
+	font_id: z.string().min(1).optional().describe("Font ID referencing a font from the font library"),
+	font_size: z.number().positive().optional().describe("Font size in points"),
+	font_color: z.string().min(1).optional().describe("Font color as hex string, e.g. \"#FFFFFF\""),
+	font_weight: z.number().int().optional().describe("Numeric font weight, e.g. 400 or 700"),
+	entrance_animation: z.string().optional().describe("Entrance animation preset ID, e.g. \"fade\" or \"none\""),
+	entrance_animation_duration_ms: z.number().min(0).optional().describe("Entrance animation duration in ms"),
+	stroke_color: z.string().min(1).optional().describe("Outline/stroke color as hex string, e.g. \"#000000\""),
+	stroke_width: z.number().min(0).optional().describe("Outline/stroke width in pixels"),
+	position_x: z.number().optional().describe("Caption center X as a fraction (0.0 to 1.0)"),
+	position_y: z.number().optional().describe("Caption center Y as a fraction (0.0 to 1.0)")
+}).describe("Document-wide caption style patch (no caption_id; applies to every caption)");
+//#endregion
+//#region src/editor/schemas/set-caption-visibility.ts
+/**
+* Toggle caption visibility (the caption track's `is_hidden` flag). Pure
+* document edit, no cascade — captions keep their positions; only the lane's
+* hidden flag changes.
+*/
+const setCaptionVisibilityInputSchema = z.object({ is_hidden: z.boolean().describe("Whether the caption track is hidden") });
+//#endregion
+//#region src/editor/schemas/set-video-clip-speed-shift.ts
+/**
+* Set the playback speed of existing video clips. Per the speed-shift decision
+* (`reference/16` §0): the op writes only the `speed_shift` fact — it does NOT
+* store an effective `duration_ms` (projection derives it from the trim window /
+* speed) and does NOT scale anchored speeches' relative offsets (offsets stay
+* put; the cascade reflows absolute positions). A `null` speed_shift clears the
+* speed back to original (1×).
+*/
+const setVideoClipSpeedShiftInputSchema = z.object({ clips: z.array(z.object({
+	clip_id: clipIdSchema.describe("The video clip part ID to set speed for"),
+	speed_shift: speedShiftSchema.nullable().describe("The new speed setting, or null to reset to 1×")
+})).min(1).describe("Video clips with their new speed settings") });
+//#endregion
+//#region src/editor/schemas/index.ts
+var schemas_exports = /* @__PURE__ */ __exportAll({
+	addSpeechesInputSchema: () => addSpeechesInputSchema,
+	addVideoClipsInputSchema: () => addVideoClipsInputSchema,
+	adjustBgmVolumeInputSchema: () => adjustBgmVolumeInputSchema,
+	adjustSpeechVolumeInputSchema: () => adjustSpeechVolumeInputSchema,
+	adjustVideoClipDurationInputSchema: () => adjustVideoClipDurationInputSchema,
+	adjustVideoClipVolumeInputSchema: () => adjustVideoClipVolumeInputSchema,
+	anchoredDeletePolicySchema: () => anchoredDeletePolicySchema,
+	anchoredReplacePolicySchema: () => anchoredReplacePolicySchema,
+	changeSpeechScriptInputSchema: () => changeSpeechScriptInputSchema,
+	changeSpeechVoiceInputSchema: () => changeSpeechVoiceInputSchema,
+	clipIdSchema: () => clipIdSchema,
+	clipIdsSchema: () => clipIdsSchema,
+	deleteBgmInputSchema: () => deleteBgmInputSchema,
+	deleteSpeechesInputSchema: () => deleteSpeechesInputSchema,
+	deleteVideoClipsInputSchema: () => deleteVideoClipsInputSchema,
+	mediaIdSchema: () => mediaIdSchema,
+	moveAnchorSchema: () => moveAnchorSchema,
+	moveBgmInputSchema: () => moveBgmInputSchema,
+	moveSpeechesInputSchema: () => moveSpeechesInputSchema,
+	moveVideoClipsByAnchorInputSchema: () => moveVideoClipsByAnchorInputSchema,
+	moveVideoClipsInputSchema: () => moveVideoClipsInputSchema,
+	movedClipAnchoredPolicySchema: () => movedClipAnchoredPolicySchema,
+	positiveMsSchema: () => positiveMsSchema,
+	replaceVideoClipContentInputSchema: () => replaceVideoClipContentInputSchema,
+	replaceVideoClipSequenceInputSchema: () => replaceVideoClipSequenceInputSchema,
+	setBgmInputSchema: () => setBgmInputSchema,
+	setCaptionStyleInputSchema: () => setCaptionStyleInputSchema,
+	setCaptionVisibilityInputSchema: () => setCaptionVisibilityInputSchema,
+	setVideoClipSpeedShiftInputSchema: () => setVideoClipSpeedShiftInputSchema,
+	speechAssetsSchema: () => speechAssetsSchema,
+	speechIdSchema: () => speechIdSchema,
+	speechIdsSchema: () => speechIdsSchema,
+	speedShiftSchema: () => speedShiftSchema,
+	timelineMsSchema: () => timelineMsSchema,
+	voiceSchema: () => voiceSchema,
+	volumeSchema: () => volumeSchema
+});
+//#endregion
+export { addSpeechesInputSchema, addVideoClipsInputSchema, adjustBgmVolumeInputSchema, adjustSpeechVolumeInputSchema, adjustVideoClipDurationInputSchema, adjustVideoClipVolumeInputSchema, anchoredDeletePolicySchema, anchoredReplacePolicySchema, changeSpeechScriptInputSchema, changeSpeechVoiceInputSchema, clipIdSchema, clipIdsSchema, deleteBgmInputSchema, deleteSpeechesInputSchema, deleteVideoClipsInputSchema, mediaIdSchema, moveAnchorSchema, moveBgmInputSchema, moveSpeechesInputSchema, moveVideoClipsByAnchorInputSchema, moveVideoClipsInputSchema, movedClipAnchoredPolicySchema, positiveMsSchema, replaceVideoClipContentInputSchema, replaceVideoClipSequenceInputSchema, setBgmInputSchema, setCaptionStyleInputSchema, setCaptionVisibilityInputSchema, setVideoClipSpeedShiftInputSchema, speechAssetsSchema, speechIdSchema, speechIdsSchema, speedShiftSchema, schemas_exports as t, timelineMsSchema, voiceSchema, volumeSchema };

package/dist/testing.d.ts

@@ -1,4 +1,4 @@
-import { i as MirrorVideoDocumentAdapter, nt as VideoDraft } from "./index-DrEHTguy.js";
+import { c as MirrorVideoDocumentAdapter, ct as VideoDraft } from "./index-DkJ0JK2t.js";
 
 //#region src/testing/in-memory-mengine-server.d.ts
 /**

package/dist/testing.js

@@ -1,4 +1,4 @@
-import { B as base64ToBytes, F as VIDEO_DOCUMENT_SCHEMA_VERSION, V as bytesToBase64, r as createMirrorVideoDocumentAdapter } from "./document-C5KDvMMR.js";
+import { G as base64ToBytes, K as bytesToBase64, o as createMirrorVideoDocumentAdapter, z as VIDEO_DOCUMENT_SCHEMA_VERSION } from "./document-C-h5-rCG.js";
 import { t as classifyUpdate } from "./loro-relay-doc-ssdYpuef.js";
 import { LoroDoc, VersionVector, encodeFrontiers } from "loro-crdt";
 //#region src/testing/in-memory-mengine-server.ts
@@ -36,6 +36,10 @@ var InMemoryMengineServer = class {
 		if (path.startsWith("sync") && method === "GET") return this.handleSync(url);
 		if (path === "updates" && method === "POST") return this.handlePush(init);
 		if (path === "events" && method === "GET") return this.handleEvents(init);
+		if (path === "audit" && method === "GET") return Response.json({ entries: extractAllMeta(this.doc).map((meta, index) => ({
+			...meta,
+			update_seq: index
+		})) });
 		return new Response("not found", { status: 404 });
 	};
 	parsePath(url) {
@@ -218,6 +222,41 @@ function extractMeta(doc, baseVV) {
 		frontiers
 	};
 }
+/** Mirror the real server audit: one deterministic entry per Loro Change. */
+function extractAllMeta(doc) {
+	const json = doc.exportJsonUpdates(new VersionVector(/* @__PURE__ */ new Map()));
+	const parsed = typeof json === "string" ? JSON.parse(json) : json;
+	const peers = parsed.peers ?? [];
+	return (parsed.changes ?? []).map((change) => {
+		const [counterPart, peerPart] = change.id.split("@");
+		return {
+			peer: peers[Number(peerPart)] ?? peerPart ?? "0",
+			counter: Number(counterPart),
+			opsLen: change.ops?.length ?? 1,
+			lamport: change.lamport ?? 0,
+			timestamp: change.timestamp ?? 0,
+			message: change.msg ?? change.message ?? null
+		};
+	}).sort((a, b) => a.lamport - b.lamport || a.peer.localeCompare(b.peer)).map((change) => {
+		const { semantic_op, payload, intent, actor, parse_error } = parseMessage(change.message);
+		return {
+			semantic_op,
+			payload,
+			intent,
+			actor,
+			message: change.message,
+			parse_error,
+			peer: change.peer,
+			counter: change.counter,
+			lamport: change.lamport,
+			timestamp: change.timestamp,
+			frontiers: bytesToBase64(encodeFrontiers([{
+				peer: change.peer,
+				counter: change.counter + change.opsLen - 1
+			}]))
+		};
+	});
+}
 /**
 * Mirrors `apps/mengine-server`'s `parseMessage` / `parseActor`. Kept in step
 * deliberately: this double is what harness tests push against, so a divergence

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mengine/medeo-client",
-  "version": "1.2.1-alpha.2",
+  "version": "1.2.1-alpha.4",
   "license": "UNLICENSED",
   "repository": {
     "type": "git",
@@ -15,6 +15,7 @@
   "exports": {
     ".": "./dist/index.js",
     "./relay": "./dist/relay.js",
+    "./schemas": "./dist/schemas.js",
     "./testing": "./dist/testing.js",
     "./package.json": "./package.json"
   },
@@ -23,16 +24,18 @@
     "registry": "https://registry.npmjs.org/"
   },
   "dependencies": {
+    "immer": "^10.2.0",
     "loro-mirror": "^2.2.0",
     "zod": "^4.4.3",
-    "@mengine/storage": "1.2.1-alpha.2",
-    "@mengine/sync": "1.2.1-alpha.2",
-    "@mengine/utils": "1.2.1-alpha.2"
+    "@mengine/storage": "1.2.1-alpha.4",
+    "@mengine/utils": "1.2.1-alpha.4",
+    "@mengine/sync": "1.2.1-alpha.4"
   },
   "devDependencies": {
     "@types/node": "^25.9.1",
     "@typescript/native-preview": "7.0.0-dev.20260521.1",
     "loro-crdt": "^1.13.6",
+    "tsx": "^4.22.3",
     "typescript": "^6.0.3",
     "vite-plugin-wasm": "^3.6.0",
     "vite-plus": "^0.1.23",