@dxos/plugin-transcription 0.9.0 → 0.9.1-staging.ee54ba693a

package/PLUGIN.mdl

@@ -8,7 +8,7 @@ A real-time voice-to-text transcription plugin for DXOS Composer.
 Captures spoken audio from a microphone, converts it to timestamped text segments via the
 Whisper speech-recognition service, and stores the results in an ECHO `Transcript` object
 backed by a `Feed`.
-The plugin also exposes blueprint-compatible operations that let AI assistants open and
+The plugin also exposes skill-compatible operations that let AI assistants open and
 summarize transcripts inside agent workflows.
 
 ## Extensions
@@ -69,6 +69,12 @@ type WhisperSegment
     words: WhisperWord[]
 ```
 
+```mdl
+type Settings
+  fields:
+    entityExtraction: boolean    # default: true — annotate transcript with references to known objects
+```
+
 ## Components
 
 ```mdl
@@ -76,7 +82,7 @@ component Transcription
   desc: Read-only CodeMirror editor that renders transcript messages as Markdown with byline headings.
   props:
     transcript?: Transcript
-    model: SerializationModel<Message>
+    model: TranscriptModel<Message>
   slots:
     (none — self-contained editor viewport)
   layout: |
@@ -169,6 +175,20 @@ op summarizeTranscript
     and clearly assigned.
 ```
 
+```mdl
+op enrichMessage
+  desc: Extracts proper nouns from each transcript block of a Message using a small LLM, full-text-searches the space for matching objects, and rewrites the text with inline `[noun](dxn)` reference links.
+  input:
+    message: Message
+  output:
+    message: Message
+  effects: [echo:read, ai:inference]
+  requires: [AiService, Database]
+  note: |
+    Uses Claude Haiku. Falls back to the raw message if extraction fails (e.g. the AI
+    service is unavailable). Gated by the entityExtraction setting.
+```
+
 ```mdl
 op normalizeSentences
   desc: Post-processes an array of raw transcript messages, merging word-level tokens into grammatically complete sentences.
@@ -234,7 +254,7 @@ feat F-4: Transcript Display
 
   req F-4.2:
     when: new messages arrive during live recording
-    then: the SerializationModel is updated and the editor scrolls to show the latest content
+    then: the TranscriptModel is updated and the editor scrolls to show the latest content
 ```
 
 ```mdl
@@ -254,9 +274,9 @@ feat F-5: AI Summary
 ```
 
 ```mdl
-feat F-6: Blueprint Integration
+feat F-6: Skill Integration
 
-  req F-6.1: Blueprint exposes op:openTranscript and op:summarizeTranscript as AI tools.
+  req F-6.1: Skill exposes op:openTranscript and op:summarizeTranscript as AI tools.
   req F-6.2: An AI assistant can open a transcript by Ref and return formatted text to the user.
   req F-6.3: An AI assistant can invoke the summary operation and present the result as a message.
 ```

package/dist/lib/neutral/{TranscriptionArticle-L6XANBH5.mjs → TranscriptionArticle-FEWCWG4A.mjs}

@@ -1,6 +1,3 @@
-import {
-  renderByline
-} from "./chunk-KAEXRX3Y.mjs";
 import "./chunk-J5LGTIGS.mjs";
 
 // src/containers/TranscriptionArticle/TranscriptionArticle.tsx
@@ -9,9 +6,9 @@ import { Filter, Obj, Query } from "@dxos/echo";
 import { useMembers, useQuery } from "@dxos/react-client/echo";
 import { Panel } from "@dxos/react-ui";
 import { Menu, MenuBuilder, useMenuBuilder } from "@dxos/react-ui-menu";
+import { Transcription, renderByline, useFeedModelAdapter } from "@dxos/react-ui-transcription";
 import { Message } from "@dxos/types";
-import { Transcription } from "#components";
-import { useFeedModelAdapter, useTranscriptionRecording } from "#hooks";
+import { useTranscriptionRecording } from "#hooks";
 import { meta } from "#meta";
 var TranscriptionArticle = ({ role, subject: transcript, attendableId }) => {
   const db = Obj.getDatabase(transcript);
@@ -24,7 +21,7 @@ var TranscriptionArticle = ({ role, subject: transcript, attendableId }) => {
     label: [
       recording ? "stop-recording.label" : "start-recording.label",
       {
-        ns: meta.id
+        ns: meta.profile.key
       }
     ],
     icon: recording ? "ph--stop-circle--regular" : "ph--microphone--regular",
@@ -51,4 +48,4 @@ var TranscriptionArticle = ({ role, subject: transcript, attendableId }) => {
 export {
   TranscriptionArticle as default
 };
-//# sourceMappingURL=TranscriptionArticle-L6XANBH5.mjs.map
+//# sourceMappingURL=TranscriptionArticle-FEWCWG4A.mjs.map

package/dist/lib/neutral/TranscriptionArticle-FEWCWG4A.mjs.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../../src/containers/TranscriptionArticle/TranscriptionArticle.tsx"],
+  "sourcesContent": ["//\n// Copyright 2024 DXOS.org\n//\n\nimport React from 'react';\n\nimport { type AppSurface } from '@dxos/app-toolkit/ui';\nimport { Filter, Obj, Query } from '@dxos/echo';\nimport { useMembers, useQuery } from '@dxos/react-client/echo';\nimport { Panel } from '@dxos/react-ui';\nimport { Menu, MenuBuilder, useMenuBuilder } from '@dxos/react-ui-menu';\nimport { Transcription, renderByline, useFeedModelAdapter } from '@dxos/react-ui-transcription';\nimport { Message, type Transcript } from '@dxos/types';\n\nimport { useTranscriptionRecording } from '#hooks';\nimport { meta } from '#meta';\n\nexport type TranscriptionArticleProps = AppSurface.ObjectArticleProps<Transcript.Transcript>;\n\nexport const TranscriptionArticle = ({ role, subject: transcript, attendableId }: TranscriptionArticleProps) => {\n  const db = Obj.getDatabase(transcript);\n  const members = useMembers(db?.spaceId).map((member) => member.identity);\n  const feed = transcript.feed.target;\n  const messages = useQuery(\n    db,\n    feed ? Query.select(Filter.type(Message.Message)).from(feed) : Query.select(Filter.nothing()),\n  );\n  const model = useFeedModelAdapter(renderByline(members), messages);\n\n  // TODO(burdon): Remove if not mutable. E.g., finalized transcript.\n  const { recording, toggleRecording } = useTranscriptionRecording(transcript);\n  const menuActions = useMenuBuilder(\n    () =>\n      MenuBuilder.make()\n        .action(\n          'toggle-recording',\n          {\n            label: [recording ? 'stop-recording.label' : 'start-recording.label', { ns: meta.profile.key }],\n            icon: recording ? 'ph--stop-circle--regular' : 'ph--microphone--regular',\n            disposition: 'toolbar',\n            testId: 'transcription.toggle-recording',\n          },\n          toggleRecording,\n        )\n        .build(),\n    [recording, toggleRecording],\n  );\n\n  return (\n    <Panel.Root role={role}>\n      <Menu.Root {...menuActions} attendableId={attendableId}>\n        <Panel.Toolbar asChild>\n          <Menu.Toolbar />\n        </Panel.Toolbar>\n      </Menu.Root>\n      <Panel.Content asChild>\n        <Transcription model={model} transcript={transcript} />\n      </Panel.Content>\n    </Panel.Root>\n  );\n};\n"],
+  "mappings": ";;;AAIA,OAAOA,WAAW;AAGlB,SAASC,QAAQC,KAAKC,aAAa;AACnC,SAASC,YAAYC,gBAAgB;AACrC,SAASC,aAAa;AACtB,SAASC,MAAMC,aAAaC,sBAAsB;AAClD,SAASC,eAAeC,cAAcC,2BAA2B;AACjE,SAASC,eAAgC;AAEzC,SAASC,iCAAiC;AAC1C,SAASC,YAAY;AAId,IAAMC,uBAAuB,CAAC,EAAEC,MAAMC,SAASC,YAAYC,aAAY,MAA6B;AACzG,QAAMC,KAAKnB,IAAIoB,YAAYH,UAAAA;AAC3B,QAAMI,UAAUnB,WAAWiB,IAAIG,OAAAA,EAASC,IAAI,CAACC,WAAWA,OAAOC,QAAQ;AACvE,QAAMC,OAAOT,WAAWS,KAAKC;AAC7B,QAAMC,WAAWzB,SACfgB,IACAO,OAAOzB,MAAM4B,OAAO9B,OAAO+B,KAAKnB,QAAQA,OAAO,CAAA,EAAGoB,KAAKL,IAAAA,IAAQzB,MAAM4B,OAAO9B,OAAOiC,QAAO,CAAA,CAAA;AAE5F,QAAMC,QAAQvB,oBAAoBD,aAAaY,OAAAA,GAAUO,QAAAA;AAGzD,QAAM,EAAEM,WAAWC,gBAAe,IAAKvB,0BAA0BK,UAAAA;AACjE,QAAMmB,cAAc7B,eAClB,MACED,YAAY+B,KAAI,EACbC,OACC,oBACA;IACEC,OAAO;MAACL,YAAY,yBAAyB;MAAyB;QAAEM,IAAI3B,KAAK4B,QAAQC;MAAI;;IAC7FC,MAAMT,YAAY,6BAA6B;IAC/CU,aAAa;IACbC,QAAQ;EACV,GACAV,eAAAA,EAEDW,MAAK,GACV;IAACZ;IAAWC;GAAgB;AAG9B,SACE,sBAAA,cAAC/B,MAAM2C,MAAI;IAAChC;KACV,sBAAA,cAACV,KAAK0C,MAAI;IAAE,GAAGX;IAAalB;KAC1B,sBAAA,cAACd,MAAM4C,SAAO;IAACC,SAAAA;KACb,sBAAA,cAAC5C,KAAK2C,SAAO,IAAA,CAAA,CAAA,GAGjB,sBAAA,cAAC5C,MAAM8C,SAAO;IAACD,SAAAA;KACb,sBAAA,cAACzC,eAAAA;IAAcyB;IAAchB;;AAIrC;",
+  "names": ["React", "Filter", "Obj", "Query", "useMembers", "useQuery", "Panel", "Menu", "MenuBuilder", "useMenuBuilder", "Transcription", "renderByline", "useFeedModelAdapter", "Message", "useTranscriptionRecording", "meta", "TranscriptionArticle", "role", "subject", "transcript", "attendableId", "db", "getDatabase", "members", "spaceId", "map", "member", "identity", "feed", "target", "messages", "select", "type", "from", "nothing", "model", "recording", "toggleRecording", "menuActions", "make", "action", "label", "ns", "profile", "key", "icon", "disposition", "testId", "build", "Root", "Toolbar", "asChild", "Content"]
+}

package/dist/lib/neutral/TranscriptionPlugin.mjs

@@ -3,17 +3,18 @@ import "./chunk-J5LGTIGS.mjs";
 // src/TranscriptionPlugin.tsx
 import { Plugin } from "@dxos/app-framework";
 import { AppActivationEvents, AppPlugin } from "@dxos/app-toolkit";
+import { MarkdownEvents } from "@dxos/plugin-markdown";
 import { Transcript } from "@dxos/types";
-import { BlueprintDefinition, OperationHandler, ReactSurface, TextContent, Transcriber } from "#capabilities";
+import { AppGraphBuilder, EntityLookup, MarkdownExtension, OperationHandler, PipelineStatus, ReactSurface, RecordingSession, SkillDefinition, TextContent, Transcriber, TranscriptionDriver, TranscriptionSettings } from "#capabilities";
 import { meta } from "#meta";
 import { translations } from "#translations";
 
-// raw-loader:/__w/dxos/dxos/packages/plugins/plugin-transcription/PLUGIN.mdl?raw
-var PLUGIN_default = '---\nid: org.dxos.plugin.transcription\nname: TranscriptionPlugin\nversion: 0.1.0\n---\n\nA real-time voice-to-text transcription plugin for DXOS Composer.\nCaptures spoken audio from a microphone, converts it to timestamped text segments via the\nWhisper speech-recognition service, and stores the results in an ECHO `Transcript` object\nbacked by a `Feed`.\nThe plugin also exposes blueprint-compatible operations that let AI assistants open and\nsummarize transcripts inside agent workflows.\n\n## Extensions\n\nThe following extension dialects are used in this document.\nEach extension is defined in the Appendix or resolved via its URI.\n\n| Term        | URI                            |\n|-------------|--------------------------------|\n| `type`      | `org.dxos.mdl.type@1.0`       |\n| `feat`      | `org.dxos.mdl.feat@1.0`       |\n| `test`      | `org.dxos.mdl.test@1.0`       |\n| `component` | `org.dxos.mdl.component@1.0`  |\n| `op`        | `org.dxos.mdl.op@1.0`         |\n\n## Types\n\n```mdl\ntype TranscriptSegment\n  fields:\n    text: string              # transcribed text for this segment\n    started: string           # ISO-8601 wall-clock timestamp of segment start\n```\n\n```mdl\ntype Message\n  fields:\n    id: string\n    created: string           # ISO-8601 creation timestamp\n    blocks: TranscriptSegment[]\n    sender:\n      identityDid?: string    # DID of the speaker, if known\n```\n\n```mdl\ntype Transcript\n  fields:\n    id: string\n    feed: Ref<Feed>           # ECHO feed that stores Message objects\n    started?: string          # ISO-8601 timestamp when recording started\n```\n\n```mdl\ntype WhisperWord\n  fields:\n    word: string\n    start: number             # seconds from start of audio chunk\n    end: number\n```\n\n```mdl\ntype WhisperSegment\n  fields:\n    text: string\n    start: number\n    end: number\n    no_speech_prob: number    # probability of silence; used to filter noise\n    words: WhisperWord[]\n```\n\n## Components\n\n```mdl\ncomponent Transcription\n  desc: Read-only CodeMirror editor that renders transcript messages as Markdown with byline headings.\n  props:\n    transcript?: Transcript\n    model: SerializationModel<Message>\n  slots:\n    (none \u2014 self-contained editor viewport)\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502 [CodeMirror editor \u2014 read-only] \u2502\n    \u2502  ## Speaker \xB7 HH:MM             \u2502\n    \u2502  Transcribed text\u2026              \u2502\n    \u2502  \u2026                              \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n```mdl\ncomponent Oracle\n  desc: |\n    Ambient display surface with an SVG graph canvas and a centered text overlay.\n    Used to show the latest transcribed utterance alongside a live visualization.\n  props:\n    text?: string             # latest utterance to display\n    delay?: number            # per-character reveal delay in ms (default 12)\n  slots:\n    canvas?: ReactNode        # replaces default SVG.Graph when provided\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502  [SVG canvas \u2014 absolutely fills]\u2502\n    \u2502         [Oracle.Text]           \u2502\n    \u2502   "latest utterance\u2026"           \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n```mdl\ncomponent TranscriptionArticle\n  desc: Full-page article panel combining the Transcription editor with a start/stop recording toolbar action.\n  props:\n    subject: Transcript\n    role: string\n    attendableId?: string\n  state:\n    recording: boolean        # whether microphone capture is active\n  actions:\n    toggleRecording()\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502 [Toolbar]  [\u{1F3A4} Start/Stop]      \u2502\n    \u251C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2524\n    \u2502 [Transcription editor]          \u2502\n    \u2502  \u2026                              \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n## Operations\n\nPure operations \u2014 no side-effects.\n\n```mdl\nop openTranscript\n  desc: Loads a Transcript from ECHO, reads all Messages from its Feed, and returns the full text content rendered with speaker bylines.\n  input:\n    transcript: Ref<Transcript>\n  output: string\n  errors:\n    TranscriptNotFound: no Transcript object found for the given ref\n    FeedNotFound: the Transcript\'s feed ref cannot be resolved\n  requires: [Database, FeedService]\n```\n\nEffectful operations \u2014 produce output via AI or mutate persistent state.\n\n```mdl\nop createTranscript\n  desc: Creates a new Feed and a Transcript object linked to it, then writes both into the current ECHO space.\n  input:\n    name?: string\n    space: Space\n  output: Transcript\n  effects: [echo:write]\n```\n\n```mdl\nop summarizeTranscript\n  desc: Sends a transcript string and optional participant notes to an LLM and returns a Markdown summary with key points, verbatim quotes, and actionable tasks extracted from the conversation.\n  input:\n    transcript: string\n    notes?: string\n  output: string\n  effects: [http]\n  requires: [AiService]\n  note: |\n    Uses Claude Sonnet. Prompt instructs the model to bold speaker names,\n    include verbatim quotes, and emit only tasks that are directly actionable\n    and clearly assigned.\n```\n\n```mdl\nop normalizeSentences\n  desc: Post-processes an array of raw transcript messages, merging word-level tokens into grammatically complete sentences.\n  input:\n    messages: Message[]\n  output: Message[]\n  note: pure transformation; no effects or service dependencies\n```\n\n## Features\n\n```mdl\nfeat F-1: Recording Lifecycle\n\n  req F-1.1: User can start microphone capture from TranscriptionArticle toolbar.\n  req F-1.2: User can stop microphone capture; final buffered chunks are flushed and transcribed.\n  req F-1.3: TranscriptionManager restarts the Transcriber automatically when the audio track changes.\n```\n\n```mdl\nfeat F-2: Audio Chunking and Transcription\n\n  req F-2.1:\n    when: user is speaking and chunk count reaches transcribeAfterChunksAmount\n    then: Transcriber schedules a deferred transcription task\n\n  req F-2.2:\n    when: transcription task runs\n    then: buffered PCM chunks are merged into a WAV file and POSTed to the Whisper endpoint\n\n  req F-2.3:\n    when: Whisper response is received\n    then: segments are filtered by no_speech_prob and aligned to absolute wall-clock timestamps\n\n  req F-2.4:\n    when: user is not speaking\n    then: only prefixBufferChunksAmount most-recent chunks are retained; older chunks discarded\n```\n\n```mdl\nfeat F-3: Transcript Persistence\n\n  req F-3.1:\n    when: aligned segments are produced\n    then: a Message object with the segments is appended to the Transcript\'s ECHO Feed\n\n  req F-3.2:\n    when: a messageEnricher is configured\n    then: Message is passed through the enricher before being written to the Feed\n\n  req F-3.3:\n    when: a peer\'s ECHO space receives the Feed update\n    then: TranscriptionArticle re-renders with the new messages\n    tags: [collaborative]\n```\n\n```mdl\nfeat F-4: Transcript Display\n\n  req F-4.1:\n    when: a Transcript object is opened\n    then: all Messages are rendered in the Transcription editor with speaker bylines and timestamps\n\n  req F-4.2:\n    when: new messages arrive during live recording\n    then: the SerializationModel is updated and the editor scrolls to show the latest content\n```\n\n```mdl\nfeat F-5: AI Summary\n\n  req F-5.1:\n    when: op:summarizeTranscript is invoked\n    then: LLM returns a structured Markdown document with key points, quotes, and tasks\n\n  req F-5.2:\n    when: participant notes are provided\n    then: notes are correlated with transcript content in the summary\n\n  req F-5.3:\n    when: no actionable tasks are found\n    then: the tasks section is omitted from the summary\n```\n\n```mdl\nfeat F-6: Blueprint Integration\n\n  req F-6.1: Blueprint exposes op:openTranscript and op:summarizeTranscript as AI tools.\n  req F-6.2: An AI assistant can open a transcript by Ref and return formatted text to the user.\n  req F-6.3: An AI assistant can invoke the summary operation and present the result as a message.\n```\n\n## Acceptance\n\n```mdl\ntest T-1: Create transcript\n  given: a DXOS space is available\n  when: op:createTranscript is invoked\n  then:\n    - a new Feed is created in the space\n    - a Transcript object linked to the Feed is created in the space\n    - the returned Transcript.feed.target is not null\n```\n\n```mdl\ntest T-2: Record and persist segment\n  given: a Transcript with a Feed, microphone capture active\n  when: the user speaks and silence is detected\n  then:\n    - Whisper returns at least one segment\n    - a Message is appended to the Feed\n    - Message.blocks contains at least one TranscriptSegment with non-empty text\n```\n\n```mdl\ntest T-3: Deduplicate segments across chunk boundaries\n  given: two consecutive transcription tasks with overlapping Whisper output\n  when: the second task produces segments whose end timestamps fall before lastTimestamp\n  then:\n    - duplicate segments are filtered out\n    - only novel words (start >= lastTimestamp) are included\n```\n\n```mdl\ntest T-4: Open transcript\n  given: a Transcript with three Messages in its Feed\n  when: op:openTranscript is invoked\n  then:\n    - content string contains three byline sections\n    - each section starts with "## <speaker> \xB7 <time>"\n```\n\n```mdl\ntest T-5: Summarize transcript\n  given: a transcript string and optional notes string\n  when: op:summarizeTranscript is invoked\n  then:\n    - summary is a non-empty Markdown string\n    - speaker names appear in bold\n    - a tasks section is present if actionable tasks were detected\n```\n\n```mdl\ntest T-6: Sentence normalization is pure\n  given: an array of raw word-level Messages\n  when: op:normalizeSentences is called\n  then:\n    - output messages contain complete sentences\n    - no ECHO writes occur\n    - input array is not mutated\n```\n\n```mdl\ntest T-7: Toggle recording in UI\n  given: TranscriptionArticle is mounted with a Transcript\n  when: user clicks the Start Recording toolbar button\n  then:\n    - recording state becomes true\n    - toolbar button label changes to "Stop Recording"\n    - microphone capture begins\n```\n\n---\n\n## Appendix: Extension Definitions\n\nExtension block types used in this document are defined below using\nthe core `ext` primitive \u2014 the only construct the base language provides.\n\n```mdl\next type\n  uri: org.dxos.mdl.type@1.0\n  desc: A named data structure with typed fields and optional literals.\n  fields:\n    desc?: Prose\n    fields?: FieldMap        # name[?]: TypeExpr  (# inline comment)\n    literals?: UnionList     # a | b | c\n    extends?: TypeRef[]\n```\n\n```mdl\next feat\n  uri: org.dxos.mdl.feat@1.0\n  desc: A named feature grouping one or more requirements.\n  fields:\n    desc?: Prose\n    req: RequirementList\n  nesting: self              # feat blocks may contain feat blocks\n```\n\n```mdl\next test\n  uri: org.dxos.mdl.test@1.0\n  desc: An acceptance scenario expressed as given / when / then steps.\n  fields:\n    given?: Step | Step[]\n    when?: Step | Step[]\n    then: Step | Step[]\n    tags?: TagList\n```\n\n```mdl\next component\n  uri: org.dxos.mdl.component@1.0\n  desc: A UI component with props, internal state, slots, actions, and events.\n  fields:\n    desc?: Prose\n    props?: FieldMap            # external inputs (immutable inside component)\n    state?: FieldMap            # internal reactive state\n    slots?: FieldMap            # named ReactNode injection points\n    actions?: ActionMap         # methods the component exposes or handles\n    emits?: EventMap            # events the component raises to its parent\n    layout?: CodeBlock          # ASCII sketch of visual structure (non-normative)\n```\n\n```mdl\next op\n  uri: org.dxos.mdl.op@1.0\n  desc: |\n    A named operation with typed inputs, outputs, and declared errors.\n    Pure ops have no effects or requires. Effectful ops declare both.\n  fields:\n    desc?: Prose\n    input?: FieldMap            # named input parameters\n    output?: TypeExpr           # return type\n    errors?: ErrorMap           # name: Prose  (when this error occurs)\n    effects?: EffectList        # echo:read | echo:write | http | fs | ...\n    requires?: ServiceList      # injected service dependencies\n    note?: Prose                # implementation guidance (non-normative)\n```\n';
+// raw-loader:/__w/dxos/dxos/packages/plugins/plugin-transcription/PLUGIN.mdl
+var PLUGIN_default = '---\nid: org.dxos.plugin.transcription\nname: TranscriptionPlugin\nversion: 0.1.0\n---\n\nA real-time voice-to-text transcription plugin for DXOS Composer.\nCaptures spoken audio from a microphone, converts it to timestamped text segments via the\nWhisper speech-recognition service, and stores the results in an ECHO `Transcript` object\nbacked by a `Feed`.\nThe plugin also exposes skill-compatible operations that let AI assistants open and\nsummarize transcripts inside agent workflows.\n\n## Extensions\n\nThe following extension dialects are used in this document.\nEach extension is defined in the Appendix or resolved via its URI.\n\n| Term        | URI                            |\n|-------------|--------------------------------|\n| `type`      | `org.dxos.mdl.type@1.0`       |\n| `feat`      | `org.dxos.mdl.feat@1.0`       |\n| `test`      | `org.dxos.mdl.test@1.0`       |\n| `component` | `org.dxos.mdl.component@1.0`  |\n| `op`        | `org.dxos.mdl.op@1.0`         |\n\n## Types\n\n```mdl\ntype TranscriptSegment\n  fields:\n    text: string              # transcribed text for this segment\n    started: string           # ISO-8601 wall-clock timestamp of segment start\n```\n\n```mdl\ntype Message\n  fields:\n    id: string\n    created: string           # ISO-8601 creation timestamp\n    blocks: TranscriptSegment[]\n    sender:\n      identityDid?: string    # DID of the speaker, if known\n```\n\n```mdl\ntype Transcript\n  fields:\n    id: string\n    feed: Ref<Feed>           # ECHO feed that stores Message objects\n    started?: string          # ISO-8601 timestamp when recording started\n```\n\n```mdl\ntype WhisperWord\n  fields:\n    word: string\n    start: number             # seconds from start of audio chunk\n    end: number\n```\n\n```mdl\ntype WhisperSegment\n  fields:\n    text: string\n    start: number\n    end: number\n    no_speech_prob: number    # probability of silence; used to filter noise\n    words: WhisperWord[]\n```\n\n```mdl\ntype Settings\n  fields:\n    entityExtraction: boolean    # default: true \u2014 annotate transcript with references to known objects\n```\n\n## Components\n\n```mdl\ncomponent Transcription\n  desc: Read-only CodeMirror editor that renders transcript messages as Markdown with byline headings.\n  props:\n    transcript?: Transcript\n    model: TranscriptModel<Message>\n  slots:\n    (none \u2014 self-contained editor viewport)\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502 [CodeMirror editor \u2014 read-only] \u2502\n    \u2502  ## Speaker \xB7 HH:MM             \u2502\n    \u2502  Transcribed text\u2026              \u2502\n    \u2502  \u2026                              \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n```mdl\ncomponent Oracle\n  desc: |\n    Ambient display surface with an SVG graph canvas and a centered text overlay.\n    Used to show the latest transcribed utterance alongside a live visualization.\n  props:\n    text?: string             # latest utterance to display\n    delay?: number            # per-character reveal delay in ms (default 12)\n  slots:\n    canvas?: ReactNode        # replaces default SVG.Graph when provided\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502  [SVG canvas \u2014 absolutely fills]\u2502\n    \u2502         [Oracle.Text]           \u2502\n    \u2502   "latest utterance\u2026"           \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n```mdl\ncomponent TranscriptionArticle\n  desc: Full-page article panel combining the Transcription editor with a start/stop recording toolbar action.\n  props:\n    subject: Transcript\n    role: string\n    attendableId?: string\n  state:\n    recording: boolean        # whether microphone capture is active\n  actions:\n    toggleRecording()\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502 [Toolbar]  [\u{1F3A4} Start/Stop]      \u2502\n    \u251C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2524\n    \u2502 [Transcription editor]          \u2502\n    \u2502  \u2026                              \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n## Operations\n\nPure operations \u2014 no side-effects.\n\n```mdl\nop openTranscript\n  desc: Loads a Transcript from ECHO, reads all Messages from its Feed, and returns the full text content rendered with speaker bylines.\n  input:\n    transcript: Ref<Transcript>\n  output: string\n  errors:\n    TranscriptNotFound: no Transcript object found for the given ref\n    FeedNotFound: the Transcript\'s feed ref cannot be resolved\n  requires: [Database, FeedService]\n```\n\nEffectful operations \u2014 produce output via AI or mutate persistent state.\n\n```mdl\nop createTranscript\n  desc: Creates a new Feed and a Transcript object linked to it, then writes both into the current ECHO space.\n  input:\n    name?: string\n    space: Space\n  output: Transcript\n  effects: [echo:write]\n```\n\n```mdl\nop summarizeTranscript\n  desc: Sends a transcript string and optional participant notes to an LLM and returns a Markdown summary with key points, verbatim quotes, and actionable tasks extracted from the conversation.\n  input:\n    transcript: string\n    notes?: string\n  output: string\n  effects: [http]\n  requires: [AiService]\n  note: |\n    Uses Claude Sonnet. Prompt instructs the model to bold speaker names,\n    include verbatim quotes, and emit only tasks that are directly actionable\n    and clearly assigned.\n```\n\n```mdl\nop enrichMessage\n  desc: Extracts proper nouns from each transcript block of a Message using a small LLM, full-text-searches the space for matching objects, and rewrites the text with inline `[noun](dxn)` reference links.\n  input:\n    message: Message\n  output:\n    message: Message\n  effects: [echo:read, ai:inference]\n  requires: [AiService, Database]\n  note: |\n    Uses Claude Haiku. Falls back to the raw message if extraction fails (e.g. the AI\n    service is unavailable). Gated by the entityExtraction setting.\n```\n\n```mdl\nop normalizeSentences\n  desc: Post-processes an array of raw transcript messages, merging word-level tokens into grammatically complete sentences.\n  input:\n    messages: Message[]\n  output: Message[]\n  note: pure transformation; no effects or service dependencies\n```\n\n## Features\n\n```mdl\nfeat F-1: Recording Lifecycle\n\n  req F-1.1: User can start microphone capture from TranscriptionArticle toolbar.\n  req F-1.2: User can stop microphone capture; final buffered chunks are flushed and transcribed.\n  req F-1.3: TranscriptionManager restarts the Transcriber automatically when the audio track changes.\n```\n\n```mdl\nfeat F-2: Audio Chunking and Transcription\n\n  req F-2.1:\n    when: user is speaking and chunk count reaches transcribeAfterChunksAmount\n    then: Transcriber schedules a deferred transcription task\n\n  req F-2.2:\n    when: transcription task runs\n    then: buffered PCM chunks are merged into a WAV file and POSTed to the Whisper endpoint\n\n  req F-2.3:\n    when: Whisper response is received\n    then: segments are filtered by no_speech_prob and aligned to absolute wall-clock timestamps\n\n  req F-2.4:\n    when: user is not speaking\n    then: only prefixBufferChunksAmount most-recent chunks are retained; older chunks discarded\n```\n\n```mdl\nfeat F-3: Transcript Persistence\n\n  req F-3.1:\n    when: aligned segments are produced\n    then: a Message object with the segments is appended to the Transcript\'s ECHO Feed\n\n  req F-3.2:\n    when: a messageEnricher is configured\n    then: Message is passed through the enricher before being written to the Feed\n\n  req F-3.3:\n    when: a peer\'s ECHO space receives the Feed update\n    then: TranscriptionArticle re-renders with the new messages\n    tags: [collaborative]\n```\n\n```mdl\nfeat F-4: Transcript Display\n\n  req F-4.1:\n    when: a Transcript object is opened\n    then: all Messages are rendered in the Transcription editor with speaker bylines and timestamps\n\n  req F-4.2:\n    when: new messages arrive during live recording\n    then: the TranscriptModel is updated and the editor scrolls to show the latest content\n```\n\n```mdl\nfeat F-5: AI Summary\n\n  req F-5.1:\n    when: op:summarizeTranscript is invoked\n    then: LLM returns a structured Markdown document with key points, quotes, and tasks\n\n  req F-5.2:\n    when: participant notes are provided\n    then: notes are correlated with transcript content in the summary\n\n  req F-5.3:\n    when: no actionable tasks are found\n    then: the tasks section is omitted from the summary\n```\n\n```mdl\nfeat F-6: Skill Integration\n\n  req F-6.1: Skill exposes op:openTranscript and op:summarizeTranscript as AI tools.\n  req F-6.2: An AI assistant can open a transcript by Ref and return formatted text to the user.\n  req F-6.3: An AI assistant can invoke the summary operation and present the result as a message.\n```\n\n## Acceptance\n\n```mdl\ntest T-1: Create transcript\n  given: a DXOS space is available\n  when: op:createTranscript is invoked\n  then:\n    - a new Feed is created in the space\n    - a Transcript object linked to the Feed is created in the space\n    - the returned Transcript.feed.target is not null\n```\n\n```mdl\ntest T-2: Record and persist segment\n  given: a Transcript with a Feed, microphone capture active\n  when: the user speaks and silence is detected\n  then:\n    - Whisper returns at least one segment\n    - a Message is appended to the Feed\n    - Message.blocks contains at least one TranscriptSegment with non-empty text\n```\n\n```mdl\ntest T-3: Deduplicate segments across chunk boundaries\n  given: two consecutive transcription tasks with overlapping Whisper output\n  when: the second task produces segments whose end timestamps fall before lastTimestamp\n  then:\n    - duplicate segments are filtered out\n    - only novel words (start >= lastTimestamp) are included\n```\n\n```mdl\ntest T-4: Open transcript\n  given: a Transcript with three Messages in its Feed\n  when: op:openTranscript is invoked\n  then:\n    - content string contains three byline sections\n    - each section starts with "## <speaker> \xB7 <time>"\n```\n\n```mdl\ntest T-5: Summarize transcript\n  given: a transcript string and optional notes string\n  when: op:summarizeTranscript is invoked\n  then:\n    - summary is a non-empty Markdown string\n    - speaker names appear in bold\n    - a tasks section is present if actionable tasks were detected\n```\n\n```mdl\ntest T-6: Sentence normalization is pure\n  given: an array of raw word-level Messages\n  when: op:normalizeSentences is called\n  then:\n    - output messages contain complete sentences\n    - no ECHO writes occur\n    - input array is not mutated\n```\n\n```mdl\ntest T-7: Toggle recording in UI\n  given: TranscriptionArticle is mounted with a Transcript\n  when: user clicks the Start Recording toolbar button\n  then:\n    - recording state becomes true\n    - toolbar button label changes to "Stop Recording"\n    - microphone capture begins\n```\n\n---\n\n## Appendix: Extension Definitions\n\nExtension block types used in this document are defined below using\nthe core `ext` primitive \u2014 the only construct the base language provides.\n\n```mdl\next type\n  uri: org.dxos.mdl.type@1.0\n  desc: A named data structure with typed fields and optional literals.\n  fields:\n    desc?: Prose\n    fields?: FieldMap        # name[?]: TypeExpr  (# inline comment)\n    literals?: UnionList     # a | b | c\n    extends?: TypeRef[]\n```\n\n```mdl\next feat\n  uri: org.dxos.mdl.feat@1.0\n  desc: A named feature grouping one or more requirements.\n  fields:\n    desc?: Prose\n    req: RequirementList\n  nesting: self              # feat blocks may contain feat blocks\n```\n\n```mdl\next test\n  uri: org.dxos.mdl.test@1.0\n  desc: An acceptance scenario expressed as given / when / then steps.\n  fields:\n    given?: Step | Step[]\n    when?: Step | Step[]\n    then: Step | Step[]\n    tags?: TagList\n```\n\n```mdl\next component\n  uri: org.dxos.mdl.component@1.0\n  desc: A UI component with props, internal state, slots, actions, and events.\n  fields:\n    desc?: Prose\n    props?: FieldMap            # external inputs (immutable inside component)\n    state?: FieldMap            # internal reactive state\n    slots?: FieldMap            # named ReactNode injection points\n    actions?: ActionMap         # methods the component exposes or handles\n    emits?: EventMap            # events the component raises to its parent\n    layout?: CodeBlock          # ASCII sketch of visual structure (non-normative)\n```\n\n```mdl\next op\n  uri: org.dxos.mdl.op@1.0\n  desc: |\n    A named operation with typed inputs, outputs, and declared errors.\n    Pure ops have no effects or requires. Effectful ops declare both.\n  fields:\n    desc?: Prose\n    input?: FieldMap            # named input parameters\n    output?: TypeExpr           # return type\n    errors?: ErrorMap           # name: Prose  (when this error occurs)\n    effects?: EffectList        # echo:read | echo:write | http | fs | ...\n    requires?: ServiceList      # injected service dependencies\n    note?: Prose                # implementation guidance (non-normative)\n```\n';
 
 // src/TranscriptionPlugin.tsx
-var TranscriptionPlugin = Plugin.define(meta).pipe(AppPlugin.addBlueprintDefinitionModule({
-  activate: BlueprintDefinition
+var TranscriptionPlugin = Plugin.define(meta).pipe(AppPlugin.addSkillDefinitionModule({
+  activate: SkillDefinition
 }), AppPlugin.addTextContentModule({
   activate: TextContent
 }), AppPlugin.addOperationHandlerModule({
@@ -26,13 +27,35 @@ var TranscriptionPlugin = Plugin.define(meta).pipe(AppPlugin.addBlueprintDefinit
   activate: ReactSurface
 }), AppPlugin.addTranslationsModule({
   translations
+}), AppPlugin.addSettingsModule({
+  activate: TranscriptionSettings
+}), AppPlugin.addAppGraphModule({
+  activate: AppGraphBuilder
+}), AppPlugin.addReactContextModule({
+  activate: TranscriptionDriver
 }), Plugin.addModule({
   id: "transcription",
   activatesOn: AppActivationEvents.SetupAppGraph,
   activate: Transcriber
+}), Plugin.addModule({
+  id: "entity-lookup",
+  activatesOn: AppActivationEvents.SetupAppGraph,
+  activate: EntityLookup
+}), Plugin.addModule({
+  id: "recording-session",
+  activatesOn: AppActivationEvents.SetupSettings,
+  activate: RecordingSession
+}), Plugin.addModule({
+  id: "pipeline-status",
+  activatesOn: AppActivationEvents.SetupSettings,
+  activate: PipelineStatus
+}), Plugin.addModule({
+  id: "markdown",
+  activatesOn: MarkdownEvents.SetupExtensions,
+  activate: MarkdownExtension
 }), AppPlugin.addPluginAssetModule({
   asset: {
-    pluginId: meta.id,
+    pluginId: meta.profile.key,
     path: "PLUGIN.mdl",
     content: PLUGIN_default,
     mimeType: "application/x-mdl"

package/dist/lib/neutral/TranscriptionPlugin.mjs.map

@@ -1,7 +1,7 @@
 {
   "version": 3,
-  "sources": ["../../../src/TranscriptionPlugin.tsx", "raw-loader:/__w/dxos/dxos/packages/plugins/plugin-transcription/PLUGIN.mdl?raw"],
-  "sourcesContent": ["//\n// Copyright 2023 DXOS.org\n//\n\nimport { Plugin } from '@dxos/app-framework';\nimport { AppActivationEvents, AppPlugin } from '@dxos/app-toolkit';\nimport { Transcript } from '@dxos/types';\n\nimport { BlueprintDefinition, OperationHandler, ReactSurface, TextContent, Transcriber } from '#capabilities';\nimport { meta } from '#meta';\nimport { translations } from '#translations';\n\n// eslint-disable-next-line import/no-relative-packages\nimport pluginSpec from '../PLUGIN.mdl?raw';\n\nexport const TranscriptionPlugin = Plugin.define(meta).pipe(\n  AppPlugin.addBlueprintDefinitionModule({ activate: BlueprintDefinition }),\n  AppPlugin.addTextContentModule({ activate: TextContent }),\n  AppPlugin.addOperationHandlerModule({ activate: OperationHandler }),\n  AppPlugin.addSchemaModule({ schema: [Transcript.Transcript] }),\n  AppPlugin.addSurfaceModule({ activate: ReactSurface }),\n  AppPlugin.addTranslationsModule({ translations }),\n  Plugin.addModule({\n    id: 'transcription',\n    activatesOn: AppActivationEvents.SetupAppGraph,\n    activate: Transcriber,\n  }),\n  AppPlugin.addPluginAssetModule({\n    asset: { pluginId: meta.id, path: 'PLUGIN.mdl', content: pluginSpec, mimeType: 'application/x-mdl' },\n  }),\n  Plugin.make,\n);\n\nexport default TranscriptionPlugin;\n", "---\nid: org.dxos.plugin.transcription\nname: TranscriptionPlugin\nversion: 0.1.0\n---\n\nA real-time voice-to-text transcription plugin for DXOS Composer.\nCaptures spoken audio from a microphone, converts it to timestamped text segments via the\nWhisper speech-recognition service, and stores the results in an ECHO `Transcript` object\nbacked by a `Feed`.\nThe plugin also exposes blueprint-compatible operations that let AI assistants open and\nsummarize transcripts inside agent workflows.\n\n## Extensions\n\nThe following extension dialects are used in this document.\nEach extension is defined in the Appendix or resolved via its URI.\n\n| Term        | URI                            |\n|-------------|--------------------------------|\n| `type`      | `org.dxos.mdl.type@1.0`       |\n| `feat`      | `org.dxos.mdl.feat@1.0`       |\n| `test`      | `org.dxos.mdl.test@1.0`       |\n| `component` | `org.dxos.mdl.component@1.0`  |\n| `op`        | `org.dxos.mdl.op@1.0`         |\n\n## Types\n\n```mdl\ntype TranscriptSegment\n  fields:\n    text: string              # transcribed text for this segment\n    started: string           # ISO-8601 wall-clock timestamp of segment start\n```\n\n```mdl\ntype Message\n  fields:\n    id: string\n    created: string           # ISO-8601 creation timestamp\n    blocks: TranscriptSegment[]\n    sender:\n      identityDid?: string    # DID of the speaker, if known\n```\n\n```mdl\ntype Transcript\n  fields:\n    id: string\n    feed: Ref<Feed>           # ECHO feed that stores Message objects\n    started?: string          # ISO-8601 timestamp when recording started\n```\n\n```mdl\ntype WhisperWord\n  fields:\n    word: string\n    start: number             # seconds from start of audio chunk\n    end: number\n```\n\n```mdl\ntype WhisperSegment\n  fields:\n    text: string\n    start: number\n    end: number\n    no_speech_prob: number    # probability of silence; used to filter noise\n    words: WhisperWord[]\n```\n\n## Components\n\n```mdl\ncomponent Transcription\n  desc: Read-only CodeMirror editor that renders transcript messages as Markdown with byline headings.\n  props:\n    transcript?: Transcript\n    model: SerializationModel<Message>\n  slots:\n    (none \u2014 self-contained editor viewport)\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502 [CodeMirror editor \u2014 read-only] \u2502\n    \u2502  ## Speaker \u00B7 HH:MM             \u2502\n    \u2502  Transcribed text\u2026              \u2502\n    \u2502  \u2026                              \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n```mdl\ncomponent Oracle\n  desc: |\n    Ambient display surface with an SVG graph canvas and a centered text overlay.\n    Used to show the latest transcribed utterance alongside a live visualization.\n  props:\n    text?: string             # latest utterance to display\n    delay?: number            # per-character reveal delay in ms (default 12)\n  slots:\n    canvas?: ReactNode        # replaces default SVG.Graph when provided\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502  [SVG canvas \u2014 absolutely fills]\u2502\n    \u2502         [Oracle.Text]           \u2502\n    \u2502   \"latest utterance\u2026\"           \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n```mdl\ncomponent TranscriptionArticle\n  desc: Full-page article panel combining the Transcription editor with a start/stop recording toolbar action.\n  props:\n    subject: Transcript\n    role: string\n    attendableId?: string\n  state:\n    recording: boolean        # whether microphone capture is active\n  actions:\n    toggleRecording()\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502 [Toolbar]  [\uD83C\uDFA4 Start/Stop]      \u2502\n    \u251C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2524\n    \u2502 [Transcription editor]          \u2502\n    \u2502  \u2026                              \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n## Operations\n\nPure operations \u2014 no side-effects.\n\n```mdl\nop openTranscript\n  desc: Loads a Transcript from ECHO, reads all Messages from its Feed, and returns the full text content rendered with speaker bylines.\n  input:\n    transcript: Ref<Transcript>\n  output: string\n  errors:\n    TranscriptNotFound: no Transcript object found for the given ref\n    FeedNotFound: the Transcript's feed ref cannot be resolved\n  requires: [Database, FeedService]\n```\n\nEffectful operations \u2014 produce output via AI or mutate persistent state.\n\n```mdl\nop createTranscript\n  desc: Creates a new Feed and a Transcript object linked to it, then writes both into the current ECHO space.\n  input:\n    name?: string\n    space: Space\n  output: Transcript\n  effects: [echo:write]\n```\n\n```mdl\nop summarizeTranscript\n  desc: Sends a transcript string and optional participant notes to an LLM and returns a Markdown summary with key points, verbatim quotes, and actionable tasks extracted from the conversation.\n  input:\n    transcript: string\n    notes?: string\n  output: string\n  effects: [http]\n  requires: [AiService]\n  note: |\n    Uses Claude Sonnet. Prompt instructs the model to bold speaker names,\n    include verbatim quotes, and emit only tasks that are directly actionable\n    and clearly assigned.\n```\n\n```mdl\nop normalizeSentences\n  desc: Post-processes an array of raw transcript messages, merging word-level tokens into grammatically complete sentences.\n  input:\n    messages: Message[]\n  output: Message[]\n  note: pure transformation; no effects or service dependencies\n```\n\n## Features\n\n```mdl\nfeat F-1: Recording Lifecycle\n\n  req F-1.1: User can start microphone capture from TranscriptionArticle toolbar.\n  req F-1.2: User can stop microphone capture; final buffered chunks are flushed and transcribed.\n  req F-1.3: TranscriptionManager restarts the Transcriber automatically when the audio track changes.\n```\n\n```mdl\nfeat F-2: Audio Chunking and Transcription\n\n  req F-2.1:\n    when: user is speaking and chunk count reaches transcribeAfterChunksAmount\n    then: Transcriber schedules a deferred transcription task\n\n  req F-2.2:\n    when: transcription task runs\n    then: buffered PCM chunks are merged into a WAV file and POSTed to the Whisper endpoint\n\n  req F-2.3:\n    when: Whisper response is received\n    then: segments are filtered by no_speech_prob and aligned to absolute wall-clock timestamps\n\n  req F-2.4:\n    when: user is not speaking\n    then: only prefixBufferChunksAmount most-recent chunks are retained; older chunks discarded\n```\n\n```mdl\nfeat F-3: Transcript Persistence\n\n  req F-3.1:\n    when: aligned segments are produced\n    then: a Message object with the segments is appended to the Transcript's ECHO Feed\n\n  req F-3.2:\n    when: a messageEnricher is configured\n    then: Message is passed through the enricher before being written to the Feed\n\n  req F-3.3:\n    when: a peer's ECHO space receives the Feed update\n    then: TranscriptionArticle re-renders with the new messages\n    tags: [collaborative]\n```\n\n```mdl\nfeat F-4: Transcript Display\n\n  req F-4.1:\n    when: a Transcript object is opened\n    then: all Messages are rendered in the Transcription editor with speaker bylines and timestamps\n\n  req F-4.2:\n    when: new messages arrive during live recording\n    then: the SerializationModel is updated and the editor scrolls to show the latest content\n```\n\n```mdl\nfeat F-5: AI Summary\n\n  req F-5.1:\n    when: op:summarizeTranscript is invoked\n    then: LLM returns a structured Markdown document with key points, quotes, and tasks\n\n  req F-5.2:\n    when: participant notes are provided\n    then: notes are correlated with transcript content in the summary\n\n  req F-5.3:\n    when: no actionable tasks are found\n    then: the tasks section is omitted from the summary\n```\n\n```mdl\nfeat F-6: Blueprint Integration\n\n  req F-6.1: Blueprint exposes op:openTranscript and op:summarizeTranscript as AI tools.\n  req F-6.2: An AI assistant can open a transcript by Ref and return formatted text to the user.\n  req F-6.3: An AI assistant can invoke the summary operation and present the result as a message.\n```\n\n## Acceptance\n\n```mdl\ntest T-1: Create transcript\n  given: a DXOS space is available\n  when: op:createTranscript is invoked\n  then:\n    - a new Feed is created in the space\n    - a Transcript object linked to the Feed is created in the space\n    - the returned Transcript.feed.target is not null\n```\n\n```mdl\ntest T-2: Record and persist segment\n  given: a Transcript with a Feed, microphone capture active\n  when: the user speaks and silence is detected\n  then:\n    - Whisper returns at least one segment\n    - a Message is appended to the Feed\n    - Message.blocks contains at least one TranscriptSegment with non-empty text\n```\n\n```mdl\ntest T-3: Deduplicate segments across chunk boundaries\n  given: two consecutive transcription tasks with overlapping Whisper output\n  when: the second task produces segments whose end timestamps fall before lastTimestamp\n  then:\n    - duplicate segments are filtered out\n    - only novel words (start >= lastTimestamp) are included\n```\n\n```mdl\ntest T-4: Open transcript\n  given: a Transcript with three Messages in its Feed\n  when: op:openTranscript is invoked\n  then:\n    - content string contains three byline sections\n    - each section starts with \"## <speaker> \u00B7 <time>\"\n```\n\n```mdl\ntest T-5: Summarize transcript\n  given: a transcript string and optional notes string\n  when: op:summarizeTranscript is invoked\n  then:\n    - summary is a non-empty Markdown string\n    - speaker names appear in bold\n    - a tasks section is present if actionable tasks were detected\n```\n\n```mdl\ntest T-6: Sentence normalization is pure\n  given: an array of raw word-level Messages\n  when: op:normalizeSentences is called\n  then:\n    - output messages contain complete sentences\n    - no ECHO writes occur\n    - input array is not mutated\n```\n\n```mdl\ntest T-7: Toggle recording in UI\n  given: TranscriptionArticle is mounted with a Transcript\n  when: user clicks the Start Recording toolbar button\n  then:\n    - recording state becomes true\n    - toolbar button label changes to \"Stop Recording\"\n    - microphone capture begins\n```\n\n---\n\n## Appendix: Extension Definitions\n\nExtension block types used in this document are defined below using\nthe core `ext` primitive \u2014 the only construct the base language provides.\n\n```mdl\next type\n  uri: org.dxos.mdl.type@1.0\n  desc: A named data structure with typed fields and optional literals.\n  fields:\n    desc?: Prose\n    fields?: FieldMap        # name[?]: TypeExpr  (# inline comment)\n    literals?: UnionList     # a | b | c\n    extends?: TypeRef[]\n```\n\n```mdl\next feat\n  uri: org.dxos.mdl.feat@1.0\n  desc: A named feature grouping one or more requirements.\n  fields:\n    desc?: Prose\n    req: RequirementList\n  nesting: self              # feat blocks may contain feat blocks\n```\n\n```mdl\next test\n  uri: org.dxos.mdl.test@1.0\n  desc: An acceptance scenario expressed as given / when / then steps.\n  fields:\n    given?: Step | Step[]\n    when?: Step | Step[]\n    then: Step | Step[]\n    tags?: TagList\n```\n\n```mdl\next component\n  uri: org.dxos.mdl.component@1.0\n  desc: A UI component with props, internal state, slots, actions, and events.\n  fields:\n    desc?: Prose\n    props?: FieldMap            # external inputs (immutable inside component)\n    state?: FieldMap            # internal reactive state\n    slots?: FieldMap            # named ReactNode injection points\n    actions?: ActionMap         # methods the component exposes or handles\n    emits?: EventMap            # events the component raises to its parent\n    layout?: CodeBlock          # ASCII sketch of visual structure (non-normative)\n```\n\n```mdl\next op\n  uri: org.dxos.mdl.op@1.0\n  desc: |\n    A named operation with typed inputs, outputs, and declared errors.\n    Pure ops have no effects or requires. Effectful ops declare both.\n  fields:\n    desc?: Prose\n    input?: FieldMap            # named input parameters\n    output?: TypeExpr           # return type\n    errors?: ErrorMap           # name: Prose  (when this error occurs)\n    effects?: EffectList        # echo:read | echo:write | http | fs | ...\n    requires?: ServiceList      # injected service dependencies\n    note?: Prose                # implementation guidance (non-normative)\n```\n"],
-  "mappings": ";;;AAIA,SAASA,cAAc;AACvB,SAASC,qBAAqBC,iBAAiB;AAC/C,SAASC,kBAAkB;AAE3B,SAASC,qBAAqBC,kBAAkBC,cAAcC,aAAaC,mBAAmB;AAC9F,SAASC,YAAY;AACrB,SAASC,oBAAoB;;;ACV7B;;;ADeO,IAAMC,sBAAsBC,OAAOC,OAAOC,IAAAA,EAAMC,KACrDC,UAAUC,6BAA6B;EAAEC,UAAUC;AAAoB,CAAA,GACvEH,UAAUI,qBAAqB;EAAEF,UAAUG;AAAY,CAAA,GACvDL,UAAUM,0BAA0B;EAAEJ,UAAUK;AAAiB,CAAA,GACjEP,UAAUQ,gBAAgB;EAAEC,QAAQ;IAACC,WAAWA;;AAAY,CAAA,GAC5DV,UAAUW,iBAAiB;EAAET,UAAUU;AAAa,CAAA,GACpDZ,UAAUa,sBAAsB;EAAEC;AAAa,CAAA,GAC/ClB,OAAOmB,UAAU;EACfC,IAAI;EACJC,aAAaC,oBAAoBC;EACjCjB,UAAUkB;AACZ,CAAA,GACApB,UAAUqB,qBAAqB;EAC7BC,OAAO;IAAEC,UAAUzB,KAAKkB;IAAIQ,MAAM;IAAcC,SAASC;IAAYC,UAAU;EAAoB;AACrG,CAAA,GACA/B,OAAOgC,IAAI;AAGb,IAAA,8BAAejC;",
-  "names": ["Plugin", "AppActivationEvents", "AppPlugin", "Transcript", "BlueprintDefinition", "OperationHandler", "ReactSurface", "TextContent", "Transcriber", "meta", "translations", "TranscriptionPlugin", "Plugin", "define", "meta", "pipe", "AppPlugin", "addBlueprintDefinitionModule", "activate", "BlueprintDefinition", "addTextContentModule", "TextContent", "addOperationHandlerModule", "OperationHandler", "addSchemaModule", "schema", "Transcript", "addSurfaceModule", "ReactSurface", "addTranslationsModule", "translations", "addModule", "id", "activatesOn", "AppActivationEvents", "SetupAppGraph", "Transcriber", "addPluginAssetModule", "asset", "pluginId", "path", "content", "pluginSpec", "mimeType", "make"]
+  "sources": ["../../../src/TranscriptionPlugin.tsx", "raw-loader:/__w/dxos/dxos/packages/plugins/plugin-transcription/PLUGIN.mdl"],
+  "sourcesContent": ["//\n// Copyright 2023 DXOS.org\n//\n\nimport { Plugin } from '@dxos/app-framework';\nimport { AppActivationEvents, AppPlugin } from '@dxos/app-toolkit';\nimport { MarkdownEvents } from '@dxos/plugin-markdown';\nimport { Transcript } from '@dxos/types';\n\nimport {\n  AppGraphBuilder,\n  EntityLookup,\n  MarkdownExtension,\n  OperationHandler,\n  PipelineStatus,\n  ReactSurface,\n  RecordingSession,\n  SkillDefinition,\n  TextContent,\n  Transcriber,\n  TranscriptionDriver,\n  TranscriptionSettings,\n} from '#capabilities';\nimport { meta } from '#meta';\nimport { translations } from '#translations';\n\n// eslint-disable-next-line import/no-relative-packages\nimport pluginSpec from '../PLUGIN.mdl?raw';\n\nexport const TranscriptionPlugin = Plugin.define(meta).pipe(\n  AppPlugin.addSkillDefinitionModule({ activate: SkillDefinition }),\n  AppPlugin.addTextContentModule({ activate: TextContent }),\n  AppPlugin.addOperationHandlerModule({ activate: OperationHandler }),\n  AppPlugin.addSchemaModule({ schema: [Transcript.Transcript] }),\n  AppPlugin.addSurfaceModule({ activate: ReactSurface }),\n  AppPlugin.addTranslationsModule({ translations }),\n  AppPlugin.addSettingsModule({ activate: TranscriptionSettings }),\n  AppPlugin.addAppGraphModule({ activate: AppGraphBuilder }),\n  AppPlugin.addReactContextModule({ activate: TranscriptionDriver }),\n  Plugin.addModule({\n    id: 'transcription',\n    activatesOn: AppActivationEvents.SetupAppGraph,\n    activate: Transcriber,\n  }),\n  Plugin.addModule({\n    id: 'entity-lookup',\n    activatesOn: AppActivationEvents.SetupAppGraph,\n    activate: EntityLookup,\n  }),\n  Plugin.addModule({\n    id: 'recording-session',\n    activatesOn: AppActivationEvents.SetupSettings,\n    activate: RecordingSession,\n  }),\n  Plugin.addModule({\n    id: 'pipeline-status',\n    activatesOn: AppActivationEvents.SetupSettings,\n    activate: PipelineStatus,\n  }),\n  Plugin.addModule({\n    id: 'markdown',\n    activatesOn: MarkdownEvents.SetupExtensions,\n    activate: MarkdownExtension,\n  }),\n  AppPlugin.addPluginAssetModule({\n    asset: { pluginId: meta.profile.key, path: 'PLUGIN.mdl', content: pluginSpec, mimeType: 'application/x-mdl' },\n  }),\n  Plugin.make,\n);\n\nexport default TranscriptionPlugin;\n", "---\nid: org.dxos.plugin.transcription\nname: TranscriptionPlugin\nversion: 0.1.0\n---\n\nA real-time voice-to-text transcription plugin for DXOS Composer.\nCaptures spoken audio from a microphone, converts it to timestamped text segments via the\nWhisper speech-recognition service, and stores the results in an ECHO `Transcript` object\nbacked by a `Feed`.\nThe plugin also exposes skill-compatible operations that let AI assistants open and\nsummarize transcripts inside agent workflows.\n\n## Extensions\n\nThe following extension dialects are used in this document.\nEach extension is defined in the Appendix or resolved via its URI.\n\n| Term        | URI                            |\n|-------------|--------------------------------|\n| `type`      | `org.dxos.mdl.type@1.0`       |\n| `feat`      | `org.dxos.mdl.feat@1.0`       |\n| `test`      | `org.dxos.mdl.test@1.0`       |\n| `component` | `org.dxos.mdl.component@1.0`  |\n| `op`        | `org.dxos.mdl.op@1.0`         |\n\n## Types\n\n```mdl\ntype TranscriptSegment\n  fields:\n    text: string              # transcribed text for this segment\n    started: string           # ISO-8601 wall-clock timestamp of segment start\n```\n\n```mdl\ntype Message\n  fields:\n    id: string\n    created: string           # ISO-8601 creation timestamp\n    blocks: TranscriptSegment[]\n    sender:\n      identityDid?: string    # DID of the speaker, if known\n```\n\n```mdl\ntype Transcript\n  fields:\n    id: string\n    feed: Ref<Feed>           # ECHO feed that stores Message objects\n    started?: string          # ISO-8601 timestamp when recording started\n```\n\n```mdl\ntype WhisperWord\n  fields:\n    word: string\n    start: number             # seconds from start of audio chunk\n    end: number\n```\n\n```mdl\ntype WhisperSegment\n  fields:\n    text: string\n    start: number\n    end: number\n    no_speech_prob: number    # probability of silence; used to filter noise\n    words: WhisperWord[]\n```\n\n```mdl\ntype Settings\n  fields:\n    entityExtraction: boolean    # default: true \u2014 annotate transcript with references to known objects\n```\n\n## Components\n\n```mdl\ncomponent Transcription\n  desc: Read-only CodeMirror editor that renders transcript messages as Markdown with byline headings.\n  props:\n    transcript?: Transcript\n    model: TranscriptModel<Message>\n  slots:\n    (none \u2014 self-contained editor viewport)\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502 [CodeMirror editor \u2014 read-only] \u2502\n    \u2502  ## Speaker \u00B7 HH:MM             \u2502\n    \u2502  Transcribed text\u2026              \u2502\n    \u2502  \u2026                              \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n```mdl\ncomponent Oracle\n  desc: |\n    Ambient display surface with an SVG graph canvas and a centered text overlay.\n    Used to show the latest transcribed utterance alongside a live visualization.\n  props:\n    text?: string             # latest utterance to display\n    delay?: number            # per-character reveal delay in ms (default 12)\n  slots:\n    canvas?: ReactNode        # replaces default SVG.Graph when provided\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502  [SVG canvas \u2014 absolutely fills]\u2502\n    \u2502         [Oracle.Text]           \u2502\n    \u2502   \"latest utterance\u2026\"           \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n```mdl\ncomponent TranscriptionArticle\n  desc: Full-page article panel combining the Transcription editor with a start/stop recording toolbar action.\n  props:\n    subject: Transcript\n    role: string\n    attendableId?: string\n  state:\n    recording: boolean        # whether microphone capture is active\n  actions:\n    toggleRecording()\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502 [Toolbar]  [\uD83C\uDFA4 Start/Stop]      \u2502\n    \u251C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2524\n    \u2502 [Transcription editor]          \u2502\n    \u2502  \u2026                              \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n## Operations\n\nPure operations \u2014 no side-effects.\n\n```mdl\nop openTranscript\n  desc: Loads a Transcript from ECHO, reads all Messages from its Feed, and returns the full text content rendered with speaker bylines.\n  input:\n    transcript: Ref<Transcript>\n  output: string\n  errors:\n    TranscriptNotFound: no Transcript object found for the given ref\n    FeedNotFound: the Transcript's feed ref cannot be resolved\n  requires: [Database, FeedService]\n```\n\nEffectful operations \u2014 produce output via AI or mutate persistent state.\n\n```mdl\nop createTranscript\n  desc: Creates a new Feed and a Transcript object linked to it, then writes both into the current ECHO space.\n  input:\n    name?: string\n    space: Space\n  output: Transcript\n  effects: [echo:write]\n```\n\n```mdl\nop summarizeTranscript\n  desc: Sends a transcript string and optional participant notes to an LLM and returns a Markdown summary with key points, verbatim quotes, and actionable tasks extracted from the conversation.\n  input:\n    transcript: string\n    notes?: string\n  output: string\n  effects: [http]\n  requires: [AiService]\n  note: |\n    Uses Claude Sonnet. Prompt instructs the model to bold speaker names,\n    include verbatim quotes, and emit only tasks that are directly actionable\n    and clearly assigned.\n```\n\n```mdl\nop enrichMessage\n  desc: Extracts proper nouns from each transcript block of a Message using a small LLM, full-text-searches the space for matching objects, and rewrites the text with inline `[noun](dxn)` reference links.\n  input:\n    message: Message\n  output:\n    message: Message\n  effects: [echo:read, ai:inference]\n  requires: [AiService, Database]\n  note: |\n    Uses Claude Haiku. Falls back to the raw message if extraction fails (e.g. the AI\n    service is unavailable). Gated by the entityExtraction setting.\n```\n\n```mdl\nop normalizeSentences\n  desc: Post-processes an array of raw transcript messages, merging word-level tokens into grammatically complete sentences.\n  input:\n    messages: Message[]\n  output: Message[]\n  note: pure transformation; no effects or service dependencies\n```\n\n## Features\n\n```mdl\nfeat F-1: Recording Lifecycle\n\n  req F-1.1: User can start microphone capture from TranscriptionArticle toolbar.\n  req F-1.2: User can stop microphone capture; final buffered chunks are flushed and transcribed.\n  req F-1.3: TranscriptionManager restarts the Transcriber automatically when the audio track changes.\n```\n\n```mdl\nfeat F-2: Audio Chunking and Transcription\n\n  req F-2.1:\n    when: user is speaking and chunk count reaches transcribeAfterChunksAmount\n    then: Transcriber schedules a deferred transcription task\n\n  req F-2.2:\n    when: transcription task runs\n    then: buffered PCM chunks are merged into a WAV file and POSTed to the Whisper endpoint\n\n  req F-2.3:\n    when: Whisper response is received\n    then: segments are filtered by no_speech_prob and aligned to absolute wall-clock timestamps\n\n  req F-2.4:\n    when: user is not speaking\n    then: only prefixBufferChunksAmount most-recent chunks are retained; older chunks discarded\n```\n\n```mdl\nfeat F-3: Transcript Persistence\n\n  req F-3.1:\n    when: aligned segments are produced\n    then: a Message object with the segments is appended to the Transcript's ECHO Feed\n\n  req F-3.2:\n    when: a messageEnricher is configured\n    then: Message is passed through the enricher before being written to the Feed\n\n  req F-3.3:\n    when: a peer's ECHO space receives the Feed update\n    then: TranscriptionArticle re-renders with the new messages\n    tags: [collaborative]\n```\n\n```mdl\nfeat F-4: Transcript Display\n\n  req F-4.1:\n    when: a Transcript object is opened\n    then: all Messages are rendered in the Transcription editor with speaker bylines and timestamps\n\n  req F-4.2:\n    when: new messages arrive during live recording\n    then: the TranscriptModel is updated and the editor scrolls to show the latest content\n```\n\n```mdl\nfeat F-5: AI Summary\n\n  req F-5.1:\n    when: op:summarizeTranscript is invoked\n    then: LLM returns a structured Markdown document with key points, quotes, and tasks\n\n  req F-5.2:\n    when: participant notes are provided\n    then: notes are correlated with transcript content in the summary\n\n  req F-5.3:\n    when: no actionable tasks are found\n    then: the tasks section is omitted from the summary\n```\n\n```mdl\nfeat F-6: Skill Integration\n\n  req F-6.1: Skill exposes op:openTranscript and op:summarizeTranscript as AI tools.\n  req F-6.2: An AI assistant can open a transcript by Ref and return formatted text to the user.\n  req F-6.3: An AI assistant can invoke the summary operation and present the result as a message.\n```\n\n## Acceptance\n\n```mdl\ntest T-1: Create transcript\n  given: a DXOS space is available\n  when: op:createTranscript is invoked\n  then:\n    - a new Feed is created in the space\n    - a Transcript object linked to the Feed is created in the space\n    - the returned Transcript.feed.target is not null\n```\n\n```mdl\ntest T-2: Record and persist segment\n  given: a Transcript with a Feed, microphone capture active\n  when: the user speaks and silence is detected\n  then:\n    - Whisper returns at least one segment\n    - a Message is appended to the Feed\n    - Message.blocks contains at least one TranscriptSegment with non-empty text\n```\n\n```mdl\ntest T-3: Deduplicate segments across chunk boundaries\n  given: two consecutive transcription tasks with overlapping Whisper output\n  when: the second task produces segments whose end timestamps fall before lastTimestamp\n  then:\n    - duplicate segments are filtered out\n    - only novel words (start >= lastTimestamp) are included\n```\n\n```mdl\ntest T-4: Open transcript\n  given: a Transcript with three Messages in its Feed\n  when: op:openTranscript is invoked\n  then:\n    - content string contains three byline sections\n    - each section starts with \"## <speaker> \u00B7 <time>\"\n```\n\n```mdl\ntest T-5: Summarize transcript\n  given: a transcript string and optional notes string\n  when: op:summarizeTranscript is invoked\n  then:\n    - summary is a non-empty Markdown string\n    - speaker names appear in bold\n    - a tasks section is present if actionable tasks were detected\n```\n\n```mdl\ntest T-6: Sentence normalization is pure\n  given: an array of raw word-level Messages\n  when: op:normalizeSentences is called\n  then:\n    - output messages contain complete sentences\n    - no ECHO writes occur\n    - input array is not mutated\n```\n\n```mdl\ntest T-7: Toggle recording in UI\n  given: TranscriptionArticle is mounted with a Transcript\n  when: user clicks the Start Recording toolbar button\n  then:\n    - recording state becomes true\n    - toolbar button label changes to \"Stop Recording\"\n    - microphone capture begins\n```\n\n---\n\n## Appendix: Extension Definitions\n\nExtension block types used in this document are defined below using\nthe core `ext` primitive \u2014 the only construct the base language provides.\n\n```mdl\next type\n  uri: org.dxos.mdl.type@1.0\n  desc: A named data structure with typed fields and optional literals.\n  fields:\n    desc?: Prose\n    fields?: FieldMap        # name[?]: TypeExpr  (# inline comment)\n    literals?: UnionList     # a | b | c\n    extends?: TypeRef[]\n```\n\n```mdl\next feat\n  uri: org.dxos.mdl.feat@1.0\n  desc: A named feature grouping one or more requirements.\n  fields:\n    desc?: Prose\n    req: RequirementList\n  nesting: self              # feat blocks may contain feat blocks\n```\n\n```mdl\next test\n  uri: org.dxos.mdl.test@1.0\n  desc: An acceptance scenario expressed as given / when / then steps.\n  fields:\n    given?: Step | Step[]\n    when?: Step | Step[]\n    then: Step | Step[]\n    tags?: TagList\n```\n\n```mdl\next component\n  uri: org.dxos.mdl.component@1.0\n  desc: A UI component with props, internal state, slots, actions, and events.\n  fields:\n    desc?: Prose\n    props?: FieldMap            # external inputs (immutable inside component)\n    state?: FieldMap            # internal reactive state\n    slots?: FieldMap            # named ReactNode injection points\n    actions?: ActionMap         # methods the component exposes or handles\n    emits?: EventMap            # events the component raises to its parent\n    layout?: CodeBlock          # ASCII sketch of visual structure (non-normative)\n```\n\n```mdl\next op\n  uri: org.dxos.mdl.op@1.0\n  desc: |\n    A named operation with typed inputs, outputs, and declared errors.\n    Pure ops have no effects or requires. Effectful ops declare both.\n  fields:\n    desc?: Prose\n    input?: FieldMap            # named input parameters\n    output?: TypeExpr           # return type\n    errors?: ErrorMap           # name: Prose  (when this error occurs)\n    effects?: EffectList        # echo:read | echo:write | http | fs | ...\n    requires?: ServiceList      # injected service dependencies\n    note?: Prose                # implementation guidance (non-normative)\n```\n"],
+  "mappings": ";;;AAIA,SAASA,cAAc;AACvB,SAASC,qBAAqBC,iBAAiB;AAC/C,SAASC,sBAAsB;AAC/B,SAASC,kBAAkB;AAE3B,SACEC,iBACAC,cACAC,mBACAC,kBACAC,gBACAC,cACAC,kBACAC,iBACAC,aACAC,aACAC,qBACAC,6BACK;AACP,SAASC,YAAY;AACrB,SAASC,oBAAoB;;;ACxB7B;;;AD6BO,IAAMC,sBAAsBC,OAAOC,OAAOC,IAAAA,EAAMC,KACrDC,UAAUC,yBAAyB;EAAEC,UAAUC;AAAgB,CAAA,GAC/DH,UAAUI,qBAAqB;EAAEF,UAAUG;AAAY,CAAA,GACvDL,UAAUM,0BAA0B;EAAEJ,UAAUK;AAAiB,CAAA,GACjEP,UAAUQ,gBAAgB;EAAEC,QAAQ;IAACC,WAAWA;;AAAY,CAAA,GAC5DV,UAAUW,iBAAiB;EAAET,UAAUU;AAAa,CAAA,GACpDZ,UAAUa,sBAAsB;EAAEC;AAAa,CAAA,GAC/Cd,UAAUe,kBAAkB;EAAEb,UAAUc;AAAsB,CAAA,GAC9DhB,UAAUiB,kBAAkB;EAAEf,UAAUgB;AAAgB,CAAA,GACxDlB,UAAUmB,sBAAsB;EAAEjB,UAAUkB;AAAoB,CAAA,GAChExB,OAAOyB,UAAU;EACfC,IAAI;EACJC,aAAaC,oBAAoBC;EACjCvB,UAAUwB;AACZ,CAAA,GACA9B,OAAOyB,UAAU;EACfC,IAAI;EACJC,aAAaC,oBAAoBC;EACjCvB,UAAUyB;AACZ,CAAA,GACA/B,OAAOyB,UAAU;EACfC,IAAI;EACJC,aAAaC,oBAAoBI;EACjC1B,UAAU2B;AACZ,CAAA,GACAjC,OAAOyB,UAAU;EACfC,IAAI;EACJC,aAAaC,oBAAoBI;EACjC1B,UAAU4B;AACZ,CAAA,GACAlC,OAAOyB,UAAU;EACfC,IAAI;EACJC,aAAaQ,eAAeC;EAC5B9B,UAAU+B;AACZ,CAAA,GACAjC,UAAUkC,qBAAqB;EAC7BC,OAAO;IAAEC,UAAUtC,KAAKuC,QAAQC;IAAKC,MAAM;IAAcC,SAASC;IAAYC,UAAU;EAAoB;AAC9G,CAAA,GACA9C,OAAO+C,IAAI;AAGb,IAAA,8BAAehD;",
+  "names": ["Plugin", "AppActivationEvents", "AppPlugin", "MarkdownEvents", "Transcript", "AppGraphBuilder", "EntityLookup", "MarkdownExtension", "OperationHandler", "PipelineStatus", "ReactSurface", "RecordingSession", "SkillDefinition", "TextContent", "Transcriber", "TranscriptionDriver", "TranscriptionSettings", "meta", "translations", "TranscriptionPlugin", "Plugin", "define", "meta", "pipe", "AppPlugin", "addSkillDefinitionModule", "activate", "SkillDefinition", "addTextContentModule", "TextContent", "addOperationHandlerModule", "OperationHandler", "addSchemaModule", "schema", "Transcript", "addSurfaceModule", "ReactSurface", "addTranslationsModule", "translations", "addSettingsModule", "TranscriptionSettings", "addAppGraphModule", "AppGraphBuilder", "addReactContextModule", "TranscriptionDriver", "addModule", "id", "activatesOn", "AppActivationEvents", "SetupAppGraph", "Transcriber", "EntityLookup", "SetupSettings", "RecordingSession", "PipelineStatus", "MarkdownEvents", "SetupExtensions", "MarkdownExtension", "addPluginAssetModule", "asset", "pluginId", "profile", "key", "path", "content", "pluginSpec", "mimeType", "make"]
 }

package/dist/lib/neutral/TranscriptionPlugin.node.mjs

@@ -4,10 +4,10 @@ import "./chunk-J5LGTIGS.mjs";
 import { Plugin } from "@dxos/app-framework";
 import { AppPlugin } from "@dxos/app-toolkit";
 import { Transcript } from "@dxos/types";
-import { BlueprintDefinition, OperationHandler, TextContent } from "#capabilities";
+import { OperationHandler, SkillDefinition, TextContent } from "#capabilities";
 import { meta } from "#meta";
-var TranscriptionPlugin = Plugin.define(meta).pipe(AppPlugin.addBlueprintDefinitionModule({
-  activate: BlueprintDefinition
+var TranscriptionPlugin = Plugin.define(meta).pipe(AppPlugin.addSkillDefinitionModule({
+  activate: SkillDefinition
 }), AppPlugin.addTextContentModule({
   activate: TextContent
 }), AppPlugin.addOperationHandlerModule({

package/dist/lib/neutral/TranscriptionPlugin.node.mjs.map

@@ -1,7 +1,7 @@
 {
   "version": 3,
   "sources": ["../../../src/TranscriptionPlugin.node.ts"],
-  "sourcesContent": ["//\n// Copyright 2023 DXOS.org\n//\n\nimport { Plugin } from '@dxos/app-framework';\nimport { AppPlugin } from '@dxos/app-toolkit';\nimport { Transcript } from '@dxos/types';\n\nimport { BlueprintDefinition, OperationHandler, TextContent } from '#capabilities';\nimport { meta } from '#meta';\n\nexport const TranscriptionPlugin = Plugin.define(meta).pipe(\n  AppPlugin.addBlueprintDefinitionModule({ activate: BlueprintDefinition }),\n  AppPlugin.addTextContentModule({ activate: TextContent }),\n  AppPlugin.addOperationHandlerModule({ activate: OperationHandler }),\n  AppPlugin.addSchemaModule({ schema: [Transcript.Transcript] }),\n  Plugin.make,\n);\n\nexport default TranscriptionPlugin;\n"],
-  "mappings": ";;;AAIA,SAASA,cAAc;AACvB,SAASC,iBAAiB;AAC1B,SAASC,kBAAkB;AAE3B,SAASC,qBAAqBC,kBAAkBC,mBAAmB;AACnE,SAASC,YAAY;AAEd,IAAMC,sBAAsBP,OAAOQ,OAAOF,IAAAA,EAAMG,KACrDR,UAAUS,6BAA6B;EAAEC,UAAUR;AAAoB,CAAA,GACvEF,UAAUW,qBAAqB;EAAED,UAAUN;AAAY,CAAA,GACvDJ,UAAUY,0BAA0B;EAAEF,UAAUP;AAAiB,CAAA,GACjEH,UAAUa,gBAAgB;EAAEC,QAAQ;IAACb,WAAWA;;AAAY,CAAA,GAC5DF,OAAOgB,IAAI;AAGb,IAAA,mCAAeT;",
-  "names": ["Plugin", "AppPlugin", "Transcript", "BlueprintDefinition", "OperationHandler", "TextContent", "meta", "TranscriptionPlugin", "define", "pipe", "addBlueprintDefinitionModule", "activate", "addTextContentModule", "addOperationHandlerModule", "addSchemaModule", "schema", "make"]
+  "sourcesContent": ["//\n// Copyright 2023 DXOS.org\n//\n\nimport { Plugin } from '@dxos/app-framework';\nimport { AppPlugin } from '@dxos/app-toolkit';\nimport { Transcript } from '@dxos/types';\n\nimport { OperationHandler, SkillDefinition, TextContent } from '#capabilities';\nimport { meta } from '#meta';\n\nexport const TranscriptionPlugin = Plugin.define(meta).pipe(\n  AppPlugin.addSkillDefinitionModule({ activate: SkillDefinition }),\n  AppPlugin.addTextContentModule({ activate: TextContent }),\n  AppPlugin.addOperationHandlerModule({ activate: OperationHandler }),\n  AppPlugin.addSchemaModule({ schema: [Transcript.Transcript] }),\n  Plugin.make,\n);\n\nexport default TranscriptionPlugin;\n"],
+  "mappings": ";;;AAIA,SAASA,cAAc;AACvB,SAASC,iBAAiB;AAC1B,SAASC,kBAAkB;AAE3B,SAASC,kBAAkBC,iBAAiBC,mBAAmB;AAC/D,SAASC,YAAY;AAEd,IAAMC,sBAAsBP,OAAOQ,OAAOF,IAAAA,EAAMG,KACrDR,UAAUS,yBAAyB;EAAEC,UAAUP;AAAgB,CAAA,GAC/DH,UAAUW,qBAAqB;EAAED,UAAUN;AAAY,CAAA,GACvDJ,UAAUY,0BAA0B;EAAEF,UAAUR;AAAiB,CAAA,GACjEF,UAAUa,gBAAgB;EAAEC,QAAQ;IAACb,WAAWA;;AAAY,CAAA,GAC5DF,OAAOgB,IAAI;AAGb,IAAA,mCAAeT;",
+  "names": ["Plugin", "AppPlugin", "Transcript", "OperationHandler", "SkillDefinition", "TextContent", "meta", "TranscriptionPlugin", "define", "pipe", "addSkillDefinitionModule", "activate", "addTextContentModule", "addOperationHandlerModule", "addSchemaModule", "schema", "make"]
 }

package/dist/lib/neutral/app-graph-builder-GQRMI22E.mjs

@@ -0,0 +1,48 @@
+import "./chunk-J5LGTIGS.mjs";
+
+// src/capabilities/app-graph-builder.tsx
+import * as Effect from "effect/Effect";
+import React from "react";
+import { Capability } from "@dxos/app-framework";
+import { AppCapabilities } from "@dxos/app-toolkit";
+import { GraphBuilder, Node, NodeMatcher } from "@dxos/plugin-graph";
+import { Markdown } from "@dxos/plugin-markdown/types";
+import { Mic } from "#components";
+import { meta } from "#meta";
+var whenMarkdownDocument = NodeMatcher.whenAll(NodeMatcher.whenEchoObjectMatches, NodeMatcher.whenEchoTypeMatches(Markdown.Document));
+var app_graph_builder_default = Capability.makeModule(Effect.fnUntraced(function* () {
+  const extensions = yield* GraphBuilder.createExtension({
+    id: "transcriptionToolbar",
+    match: (node, get) => whenMarkdownDocument(node, get),
+    // The control owns recording state, mode, device selection, and entity-extraction —
+    // interactions the action model cannot express (press-and-hold, an embedded dropdown, a
+    // live device list) — so it renders via the custom toolbar variant rather than a plain
+    // toggle action. The node is static (state lives in the control), keeping the toolbar's
+    // action atom stable across recording changes.
+    actions: (matched) => Effect.succeed([
+      Node.makeAction({
+        id: "transcription",
+        data: () => Effect.void,
+        properties: {
+          label: [
+            "start-recording.label",
+            {
+              ns: meta.profile.key
+            }
+          ],
+          icon: "ph--microphone--regular",
+          disposition: "toolbar",
+          variant: "custom",
+          render: () => /* @__PURE__ */ React.createElement(Mic, {
+            docId: matched.id
+          })
+        }
+      })
+    ])
+  });
+  return Capability.contributes(AppCapabilities.AppGraphBuilder, extensions);
+}));
+export {
+  app_graph_builder_default as default
+};
+//# sourceMappingURL=app-graph-builder-GQRMI22E.mjs.map

package/dist/lib/neutral/app-graph-builder-GQRMI22E.mjs.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../../src/capabilities/app-graph-builder.tsx"],
+  "sourcesContent": ["//\n// Copyright 2026 DXOS.org\n//\n\nimport * as Effect from 'effect/Effect';\nimport React from 'react';\n\nimport { Capability } from '@dxos/app-framework';\nimport { AppCapabilities } from '@dxos/app-toolkit';\nimport { GraphBuilder, Node, NodeMatcher } from '@dxos/plugin-graph';\nimport { Markdown } from '@dxos/plugin-markdown/types';\n\nimport { Mic } from '#components';\nimport { meta } from '#meta';\n\nconst whenMarkdownDocument = NodeMatcher.whenAll(\n  NodeMatcher.whenEchoObjectMatches,\n  NodeMatcher.whenEchoTypeMatches(Markdown.Document),\n);\n\nexport default Capability.makeModule(\n  Effect.fnUntraced(function* () {\n    const extensions = yield* GraphBuilder.createExtension({\n      id: 'transcriptionToolbar',\n      match: (node, get) => whenMarkdownDocument(node, get),\n      // The control owns recording state, mode, device selection, and entity-extraction —\n      // interactions the action model cannot express (press-and-hold, an embedded dropdown, a\n      // live device list) — so it renders via the custom toolbar variant rather than a plain\n      // toggle action. The node is static (state lives in the control), keeping the toolbar's\n      // action atom stable across recording changes.\n      actions: (matched) =>\n        Effect.succeed([\n          Node.makeAction({\n            id: 'transcription',\n            data: () => Effect.void,\n            properties: {\n              label: ['start-recording.label', { ns: meta.profile.key }],\n              icon: 'ph--microphone--regular',\n              disposition: 'toolbar',\n              variant: 'custom',\n              render: () => <Mic docId={matched.id} />,\n            },\n          }),\n        ]),\n    });\n\n    return Capability.contributes(AppCapabilities.AppGraphBuilder, extensions);\n  }),\n);\n"],
+  "mappings": ";;;AAIA,YAAYA,YAAY;AACxB,OAAOC,WAAW;AAElB,SAASC,kBAAkB;AAC3B,SAASC,uBAAuB;AAChC,SAASC,cAAcC,MAAMC,mBAAmB;AAChD,SAASC,gBAAgB;AAEzB,SAASC,WAAW;AACpB,SAASC,YAAY;AAErB,IAAMC,uBAAuBJ,YAAYK,QACvCL,YAAYM,uBACZN,YAAYO,oBAAoBN,SAASO,QAAQ,CAAA;AAGnD,IAAA,4BAAeZ,WAAWa,WACjBC,kBAAW,aAAA;AAChB,QAAMC,aAAa,OAAOb,aAAac,gBAAgB;IACrDC,IAAI;IACJC,OAAO,CAACC,MAAMC,QAAQZ,qBAAqBW,MAAMC,GAAAA;;;;;;IAMjDC,SAAS,CAACC,YACDC,eAAQ;MACbpB,KAAKqB,WAAW;QACdP,IAAI;QACJQ,MAAM,MAAaC;QACnBC,YAAY;UACVC,OAAO;YAAC;YAAyB;cAAEC,IAAItB,KAAKuB,QAAQC;YAAI;;UACxDC,MAAM;UACNC,aAAa;UACbC,SAAS;UACTC,QAAQ,MAAM,sBAAA,cAAC7B,KAAAA;YAAI8B,OAAOd,QAAQL;;QACpC;MACF,CAAA;KACD;EACL,CAAA;AAEA,SAAOjB,WAAWqC,YAAYpC,gBAAgBqC,iBAAiBvB,UAAAA;AACjE,CAAA,CAAA;",
+  "names": ["Effect", "React", "Capability", "AppCapabilities", "GraphBuilder", "Node", "NodeMatcher", "Markdown", "Mic", "meta", "whenMarkdownDocument", "whenAll", "whenEchoObjectMatches", "whenEchoTypeMatches", "Document", "makeModule", "fnUntraced", "extensions", "createExtension", "id", "match", "node", "get", "actions", "matched", "succeed", "makeAction", "data", "void", "properties", "label", "ns", "profile", "key", "icon", "disposition", "variant", "render", "docId", "contributes", "AppGraphBuilder"]
+}

package/dist/lib/neutral/capabilities/index.mjs

@@ -2,16 +2,30 @@ import "../chunk-J5LGTIGS.mjs";
 
 // src/capabilities/index.ts
 import { Capability } from "@dxos/app-framework";
-var BlueprintDefinition = Capability.lazy("BlueprintDefinition", () => import("../blueprint-definition-DA4QFRYJ.mjs"));
-var TextContent = Capability.lazy("TextContent", () => import("../text-content-KBB37HSM.mjs"));
+var AppGraphBuilder = Capability.lazy("AppGraphBuilder", () => import("../app-graph-builder-GQRMI22E.mjs"));
+var EntityLookup = Capability.lazy("EntityLookup", () => import("../entity-lookup-PCDW7NC4.mjs"));
+var MarkdownExtension = Capability.lazy("MarkdownExtension", () => import("../markdown-extension-CCDXXY7E.mjs"));
+var PipelineStatus = Capability.lazy("PipelineStatus", () => import("../pipeline-status-XMWMAUJY.mjs"));
+var RecordingSession = Capability.lazy("RecordingSession", () => import("../recording-session-IADO7ATU.mjs"));
+var TranscriptionDriver = Capability.lazy("TranscriptionDriver", () => import("../transcription-driver-DN6ELQXZ.mjs"));
+var SkillDefinition = Capability.lazy("SkillDefinition", () => import("../skill-definition-DXNK324V.mjs"));
+var TextContent = Capability.lazy("TextContent", () => import("../text-content-NOWR642C.mjs"));
 var OperationHandler = Capability.lazy("OperationHandler", () => import("../operation-handler-NXMOAXFP.mjs"));
 var ReactSurface = Capability.lazy("ReactSurface", () => import("../react-surface-YMZRO242.mjs"));
-var Transcriber = Capability.lazy("Transcriber", () => import("../transcriber-OGSF3XO5.mjs"));
+var Transcriber = Capability.lazy("Transcriber", () => import("../transcriber-6XZHUUHK.mjs"));
+var TranscriptionSettings = Capability.lazy("TranscriptionSettings", () => import("../settings-7KE4MABW.mjs"));
 export {
-  BlueprintDefinition,
+  AppGraphBuilder,
+  EntityLookup,
+  MarkdownExtension,
   OperationHandler,
+  PipelineStatus,
   ReactSurface,
+  RecordingSession,
+  SkillDefinition,
   TextContent,
-  Transcriber
+  Transcriber,
+  TranscriptionDriver,
+  TranscriptionSettings
 };
 //# sourceMappingURL=index.mjs.map

package/dist/lib/neutral/capabilities/index.mjs.map

@@ -1,7 +1,7 @@
 {
   "version": 3,
   "sources": ["../../../../src/capabilities/index.ts"],
-  "sourcesContent": ["//\n// Copyright 2025 DXOS.org\n//\n\nimport { Capability } from '@dxos/app-framework';\nimport { type AppCapabilities } from '@dxos/app-toolkit';\nimport type { OperationHandlerSet } from '@dxos/compute';\n\n// The contributed capability type references Blueprint types from @dxos/compute, so the lazy\n// wrapper needs an explicit annotation to keep the inferred type portable (TS2883).\nexport const BlueprintDefinition: Capability.LazyCapability<\n  void,\n  Capability.Capability<typeof AppCapabilities.BlueprintDefinition>[]\n> = Capability.lazy('BlueprintDefinition', () => import('./blueprint-definition'));\nexport const TextContent = Capability.lazy('TextContent', () => import('./text-content'));\nexport const OperationHandler = Capability.lazy<OperationHandlerSet.OperationHandlerSet>(\n  'OperationHandler',\n  () => import('./operation-handler'),\n);\nexport const ReactSurface = Capability.lazy('ReactSurface', () => import('./react-surface'));\nexport const Transcriber = Capability.lazy('Transcriber', () => import('./transcriber'));\n"],
-  "mappings": ";;;AAIA,SAASA,kBAAkB;AAMpB,IAAMC,sBAGTD,WAAWE,KAAK,uBAAuB,MAAM,OAAO,sCAAA,CAAA;AACjD,IAAMC,cAAcH,WAAWE,KAAK,eAAe,MAAM,OAAO,8BAAA,CAAA;AAChE,IAAME,mBAAmBJ,WAAWE,KACzC,oBACA,MAAM,OAAO,mCAAA,CAAA;AAER,IAAMG,eAAeL,WAAWE,KAAK,gBAAgB,MAAM,OAAO,+BAAA,CAAA;AAClE,IAAMI,cAAcN,WAAWE,KAAK,eAAe,MAAM,OAAO,6BAAA,CAAA;",
-  "names": ["Capability", "BlueprintDefinition", "lazy", "TextContent", "OperationHandler", "ReactSurface", "Transcriber"]
+  "sourcesContent": ["//\n// Copyright 2025 DXOS.org\n//\n\nimport { Capability } from '@dxos/app-framework';\nimport type { OperationHandlerSet } from '@dxos/compute';\n\nexport const AppGraphBuilder = Capability.lazy('AppGraphBuilder', () => import('./app-graph-builder'));\nexport const EntityLookup = Capability.lazy('EntityLookup', () => import('./entity-lookup'));\nexport const MarkdownExtension = Capability.lazy('MarkdownExtension', () => import('./markdown-extension'));\nexport const PipelineStatus = Capability.lazy('PipelineStatus', () => import('./pipeline-status'));\nexport const RecordingSession = Capability.lazy('RecordingSession', () => import('./recording-session'));\nexport const TranscriptionDriver = Capability.lazy('TranscriptionDriver', () => import('./transcription-driver'));\nexport const SkillDefinition = Capability.lazy('SkillDefinition', () => import('./skill-definition'));\nexport const TextContent = Capability.lazy('TextContent', () => import('./text-content'));\nexport const OperationHandler = Capability.lazy<OperationHandlerSet.OperationHandlerSet>(\n  'OperationHandler',\n  () => import('./operation-handler'),\n);\nexport const ReactSurface = Capability.lazy('ReactSurface', () => import('./react-surface'));\nexport const Transcriber = Capability.lazy('Transcriber', () => import('./transcriber'));\nexport const TranscriptionSettings = Capability.lazy('TranscriptionSettings', () => import('./settings'));\n"],
+  "mappings": ";;;AAIA,SAASA,kBAAkB;AAGpB,IAAMC,kBAAkBD,WAAWE,KAAK,mBAAmB,MAAM,OAAO,mCAAA,CAAA;AACxE,IAAMC,eAAeH,WAAWE,KAAK,gBAAgB,MAAM,OAAO,+BAAA,CAAA;AAClE,IAAME,oBAAoBJ,WAAWE,KAAK,qBAAqB,MAAM,OAAO,oCAAA,CAAA;AAC5E,IAAMG,iBAAiBL,WAAWE,KAAK,kBAAkB,MAAM,OAAO,iCAAA,CAAA;AACtE,IAAMI,mBAAmBN,WAAWE,KAAK,oBAAoB,MAAM,OAAO,mCAAA,CAAA;AAC1E,IAAMK,sBAAsBP,WAAWE,KAAK,uBAAuB,MAAM,OAAO,sCAAA,CAAA;AAChF,IAAMM,kBAAkBR,WAAWE,KAAK,mBAAmB,MAAM,OAAO,kCAAA,CAAA;AACxE,IAAMO,cAAcT,WAAWE,KAAK,eAAe,MAAM,OAAO,8BAAA,CAAA;AAChE,IAAMQ,mBAAmBV,WAAWE,KACzC,oBACA,MAAM,OAAO,mCAAA,CAAA;AAER,IAAMS,eAAeX,WAAWE,KAAK,gBAAgB,MAAM,OAAO,+BAAA,CAAA;AAClE,IAAMU,cAAcZ,WAAWE,KAAK,eAAe,MAAM,OAAO,6BAAA,CAAA;AAChE,IAAMW,wBAAwBb,WAAWE,KAAK,yBAAyB,MAAM,OAAO,0BAAA,CAAA;",
+  "names": ["Capability", "AppGraphBuilder", "lazy", "EntityLookup", "MarkdownExtension", "PipelineStatus", "RecordingSession", "TranscriptionDriver", "SkillDefinition", "TextContent", "OperationHandler", "ReactSurface", "Transcriber", "TranscriptionSettings"]
 }

package/dist/lib/neutral/chunk-5BVHY75B.mjs

@@ -0,0 +1,54 @@
+// src/meta.ts
+import { Plugin } from "@dxos/app-framework";
+
+// dx.config.ts
+import { Config2 } from "@dxos/app-framework/config";
+import { trim } from "@dxos/util";
+var dx_config_default = Config2.make({
+  plugin: {
+    key: "org.dxos.plugin.transcription",
+    name: "Transcription",
+    author: "DXOS",
+    description: trim`
+      Real-time voice-to-text transcription for DXOS Composer.
+      Captures microphone audio in short PCM chunks, batches them into WAV payloads, and streams
+      the results to a Whisper-compatible speech-recognition endpoint.
+      Transcribed segments are timestamped, deduplicated across chunk boundaries, and persisted
+      as Message objects in an ECHO Feed so that every participant in a shared space sees the
+      live transcript without any manual sync step.
+
+      Each recording session is represented as a Transcript object linked to its Feed.
+      The read-only editor renders messages with speaker bylines and wall-clock timestamps,
+      supporting markdown decoration and document preview.
+      An optional message enricher pipeline lets plugins annotate messages before they are
+      committed — for example, to attach speaker identity from a calls service.
+
+      The plugin ships two skill-compatible operations: Open (loads and formats a transcript
+      for reading by an AI assistant) and Summarize (sends the transcript text to an LLM and
+      returns a structured Markdown summary with key points, verbatim quotes, and actionable
+      tasks extracted from the conversation).
+
+      A sentence-normalization pass merges raw word-level tokens from the recognizer into
+      grammatically complete sentences, improving readability of the stored transcript without
+      altering the underlying audio timeline.
+    `,
+    source: "https://github.com/dxos/dxos/tree/main/packages/plugins/plugin-transcription",
+    icon: {
+      key: "ph--microphone--regular",
+      hue: "sky"
+    },
+    spec: "PLUGIN.mdl",
+    tags: [
+      "labs"
+    ],
+    screenshots: []
+  }
+});
+
+// src/meta.ts
+var meta = Plugin.getMetaFromConfig(dx_config_default);
+
+export {
+  meta
+};
+//# sourceMappingURL=chunk-5BVHY75B.mjs.map

package/dist/lib/neutral/chunk-5BVHY75B.mjs.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../../src/meta.ts", "../../../dx.config.ts"],
+  "sourcesContent": ["//\n// Copyright 2023 DXOS.org\n//\n\nimport { Plugin } from '@dxos/app-framework';\n\nimport config from '../dx.config';\n\nexport const meta = Plugin.getMetaFromConfig(config);\n", "//\n// Copyright 2025 DXOS.org\n//\n\nimport { Config2 } from '@dxos/app-framework/config';\nimport { trim } from '@dxos/util';\n\nexport default Config2.make({\n  plugin: {\n    key: 'org.dxos.plugin.transcription',\n    name: 'Transcription',\n    author: 'DXOS',\n    description: trim`\n      Real-time voice-to-text transcription for DXOS Composer.\n      Captures microphone audio in short PCM chunks, batches them into WAV payloads, and streams\n      the results to a Whisper-compatible speech-recognition endpoint.\n      Transcribed segments are timestamped, deduplicated across chunk boundaries, and persisted\n      as Message objects in an ECHO Feed so that every participant in a shared space sees the\n      live transcript without any manual sync step.\n\n      Each recording session is represented as a Transcript object linked to its Feed.\n      The read-only editor renders messages with speaker bylines and wall-clock timestamps,\n      supporting markdown decoration and document preview.\n      An optional message enricher pipeline lets plugins annotate messages before they are\n      committed — for example, to attach speaker identity from a calls service.\n\n      The plugin ships two skill-compatible operations: Open (loads and formats a transcript\n      for reading by an AI assistant) and Summarize (sends the transcript text to an LLM and\n      returns a structured Markdown summary with key points, verbatim quotes, and actionable\n      tasks extracted from the conversation).\n\n      A sentence-normalization pass merges raw word-level tokens from the recognizer into\n      grammatically complete sentences, improving readability of the stored transcript without\n      altering the underlying audio timeline.\n    `,\n    source: 'https://github.com/dxos/dxos/tree/main/packages/plugins/plugin-transcription',\n    icon: { key: 'ph--microphone--regular', hue: 'sky' },\n    spec: 'PLUGIN.mdl',\n    tags: ['labs'],\n    screenshots: [],\n  },\n});\n"],
+  "mappings": ";AAIA,SAASA,cAAc;;;ACAvB,SAASC,eAAe;AACxB,SAASC,YAAY;AAErB,IAAA,oBAAeD,QAAQE,KAAK;EAC1BC,QAAQ;IACNC,KAAK;IACLC,MAAM;IACNC,QAAQ;IACRC,aAAaN;;;;;;;;;;;;;;;;;;;;;;;IAuBbO,QAAQ;IACRC,MAAM;MAAEL,KAAK;MAA2BM,KAAK;IAAM;IACnDC,MAAM;IACNC,MAAM;MAAC;;IACPC,aAAa,CAAA;EACf;AACF,CAAA;;;ADjCO,IAAMC,OAAOC,OAAOC,kBAAkBC,iBAAAA;",
+  "names": ["Plugin", "Config2", "trim", "make", "plugin", "key", "name", "author", "description", "source", "icon", "hue", "spec", "tags", "screenshots", "meta", "Plugin", "getMetaFromConfig", "config"]
+}

package/dist/lib/neutral/chunk-HNXSU7HJ.mjs

@@ -0,0 +1,8 @@
+// src/operations/index.ts
+import { OperationHandlerSet } from "@dxos/compute";
+var TranscriptionOperationHandlerSet = OperationHandlerSet.lazy(() => import("./create-HNHGCKZ4.mjs"), () => import("./open-SNMZXXYO.mjs"), () => import("./summarize-LBUZTOTH.mjs"), () => import("./enrich-message-DSJU4TXC.mjs"));
+
+export {
+  TranscriptionOperationHandlerSet
+};
+//# sourceMappingURL=chunk-HNXSU7HJ.mjs.map

package/dist/lib/neutral/{chunk-KXJZEYS7.mjs.map → chunk-HNXSU7HJ.mjs.map}

@@ -1,7 +1,7 @@
 {
   "version": 3,
   "sources": ["../../../src/operations/index.ts"],
-  "sourcesContent": ["//\n// Copyright 2025 DXOS.org\n//\n\nimport { OperationHandlerSet } from '@dxos/compute';\n\nexport const TranscriptionOperationHandlerSet = OperationHandlerSet.lazy(\n  () => import('./create'),\n  () => import('./open'),\n  () => import('./summarize'),\n);\n"],
-  "mappings": ";AAIA,SAASA,2BAA2B;AAE7B,IAAMC,mCAAmCD,oBAAoBE,KAClE,MAAM,OAAO,uBAAA,GACb,MAAM,OAAO,qBAAA,GACb,MAAM,OAAO,0BAAA,CAAA;",
+  "sourcesContent": ["//\n// Copyright 2025 DXOS.org\n//\n\nimport { OperationHandlerSet } from '@dxos/compute';\n\nexport const TranscriptionOperationHandlerSet = OperationHandlerSet.lazy(\n  () => import('./create'),\n  () => import('./open'),\n  () => import('./summarize'),\n  () => import('./enrich-message'),\n);\n"],
+  "mappings": ";AAIA,SAASA,2BAA2B;AAE7B,IAAMC,mCAAmCD,oBAAoBE,KAClE,MAAM,OAAO,uBAAA,GACb,MAAM,OAAO,qBAAA,GACb,MAAM,OAAO,0BAAA,GACb,MAAM,OAAO,+BAAA,CAAA;",
   "names": ["OperationHandlerSet", "TranscriptionOperationHandlerSet", "lazy"]
 }