@alexkroman1/aai-cli 13.3.0 → 14.0.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (134) hide show
  1. package/dist/{_agent-Clbq3a4h.mjs → _agent-BzUeqOdj.mjs} +32 -9
  2. package/dist/_agent.d.ts +23 -0
  3. package/dist/{_bundler-DEkjO1d8.mjs → _bundler-DolUCMxu.mjs} +8 -7
  4. package/dist/_cli-common.d.ts +45 -6
  5. package/dist/{_config-ZY2euhdh.mjs → _config-DYzC6WMD.mjs} +2 -2
  6. package/dist/_dev-env.d.ts +1 -1
  7. package/dist/{_dev-server-C13xuGB1.mjs → _dev-server-CSMqF8PN.mjs} +20 -22
  8. package/dist/_dev-vite-config.d.ts +3 -3
  9. package/dist/{_init-Ci4QGjp8.mjs → _init-CQ8idAwo.mjs} +9 -23
  10. package/dist/{_output-DbJt9gsl.mjs → _output-DBP9Op_d.mjs} +22 -13
  11. package/dist/_output.d.ts +10 -9
  12. package/dist/{_preflight-B2laWPdU.mjs → _preflight-CcAR577V.mjs} +2 -2
  13. package/dist/_resource-commands.d.ts +9 -9
  14. package/dist/{_server-common-BV4_CrP3.mjs → _server-common-vILJp3it.mjs} +1 -1
  15. package/dist/{_slug-api-DlKpj91X.mjs → _slug-api-CGvc2YSQ.mjs} +18 -9
  16. package/dist/_slug-api.d.ts +19 -0
  17. package/dist/{_studio-DTi3EdXQ.mjs → _studio-DnR_BqFp.mjs} +7 -3
  18. package/dist/_studio-commands.d.ts +13 -13
  19. package/dist/{_templates-V4l5sZJy.mjs → _templates-CK4oKoeX.mjs} +3 -3
  20. package/dist/{_typecheck-gate-sV4N5F93.mjs → _typecheck-gate-U3jcil8L.mjs} +3 -2
  21. package/dist/_typecheck-gate.d.ts +3 -1
  22. package/dist/{_utils-Dhd058D9.mjs → _utils-D5JGcjiW.mjs} +43 -1
  23. package/dist/_utils.d.ts +35 -0
  24. package/dist/{_workflow-determinism-C9l8uC2R.mjs → _workflow-determinism-CvpqMpdh.mjs} +18 -26
  25. package/dist/_workflow-determinism.d.ts +3 -3
  26. package/dist/{build-CWyFGe-W.mjs → build-Mxk8gWvX.mjs} +46 -7
  27. package/dist/build.d.ts +5 -0
  28. package/dist/cli.d.ts +1 -2
  29. package/dist/cli.mjs +117 -135
  30. package/dist/{client-bundler-CT-_26c5.mjs → client-bundler-BJgREAh6.mjs} +1 -1
  31. package/dist/client-bundler.mjs +1 -1
  32. package/dist/{delete-RqlQZ6Y_.mjs → delete-BdFnbnVR.mjs} +4 -4
  33. package/dist/{deploy-B7f3CK7v.mjs → deploy-uAJ4NukN.mjs} +11 -11
  34. package/dist/{dev-DamXieW_.mjs → dev-DApPSaE_.mjs} +3 -3
  35. package/dist/{eval-h3DTVfRw.mjs → eval-BK47A_K5.mjs} +3 -3
  36. package/dist/{init-D-vFrV8h.mjs → init-DukDxECd.mjs} +11 -19
  37. package/dist/{login-CDiemNmC.mjs → login-DaLrPOFg.mjs} +4 -6
  38. package/dist/{logs-DGwPNCzB.mjs → logs-7BRwq0cG.mjs} +21 -14
  39. package/dist/project-config.mjs +1 -1
  40. package/dist/scaffold/CLAUDE.md +84 -56
  41. package/dist/scaffold/package.json +4 -4
  42. package/dist/scaffold/server.mjs +129 -25
  43. package/dist/{secret-V1w7y9-j.mjs → secret-Cjp912yF.mjs} +5 -5
  44. package/dist/secret.d.ts +8 -6
  45. package/dist/{studio-BpMCfUNx.mjs → studio-CpHlNHUZ.mjs} +13 -15
  46. package/dist/templates/call-audit/agent.test.ts +11 -6
  47. package/dist/templates/call-audit/agent.ts +2 -2
  48. package/dist/templates/call-audit/client.tsx +27 -42
  49. package/dist/templates/call-audit/workflows/audit.ts +7 -5
  50. package/dist/templates/call-audit/workflows/ingest.ts +7 -7
  51. package/dist/templates/call-audit/workflows/media.ts +12 -37
  52. package/dist/templates/call-audit/workflows/summarize.ts +8 -8
  53. package/dist/templates/call-audit/workflows/sync-api.ts +3 -3
  54. package/dist/templates/dispatch-center/agent.test.ts +7 -7
  55. package/dist/templates/dispatch-center/client.tsx +53 -61
  56. package/dist/templates/dispatch-center/shared.ts +0 -8
  57. package/dist/templates/dispatch-center/tools/ops_dashboard.ts +4 -1
  58. package/dist/templates/embedded-assets/agent.eval.test.ts +4 -2
  59. package/dist/templates/health-assistant/agent.eval.test.ts +4 -2
  60. package/dist/templates/infocom-adventure/client.tsx +40 -36
  61. package/dist/templates/link-digest/agent.test.ts +3 -3
  62. package/dist/templates/link-digest/agent.ts +2 -2
  63. package/dist/templates/link-digest/client.tsx +27 -17
  64. package/dist/templates/link-digest/workflows/digest.ts +13 -16
  65. package/dist/templates/night-owl/agent.eval.test.ts +4 -2
  66. package/dist/templates/night-owl/client.tsx +8 -2
  67. package/dist/templates/pipeline-simple/agent.test.ts +2 -2
  68. package/dist/templates/pizza-ordering/client.tsx +5 -4
  69. package/dist/templates/pizza-ordering/shared.ts +3 -8
  70. package/dist/templates/pizza-ordering/tools/add_pizza.ts +3 -2
  71. package/dist/templates/pizza-ordering/tools/place_order.ts +3 -2
  72. package/dist/templates/pizza-ordering/tools/remove_pizza.ts +3 -2
  73. package/dist/templates/pizza-ordering/tools/update_pizza.ts +3 -2
  74. package/dist/templates/pizza-ordering/tools/view_order.ts +3 -2
  75. package/dist/templates/plan-and-execute/agent.test.ts +10 -5
  76. package/dist/templates/plan-and-execute/client.tsx +2 -2
  77. package/dist/templates/podcast-digest/agent.test.ts +20 -11
  78. package/dist/templates/podcast-digest/client.tsx +16 -22
  79. package/dist/templates/podcast-digest/workflows/digest.ts +21 -19
  80. package/dist/templates/podcast-digest/workflows/feeds.ts +61 -33
  81. package/dist/templates/podcast-digest/workflows/slack.ts +7 -7
  82. package/dist/templates/recap-workflow/agent.eval.test.ts +21 -13
  83. package/dist/templates/recap-workflow/agent.test.ts +27 -23
  84. package/dist/templates/recap-workflow/workflows/recap.ts +24 -28
  85. package/dist/templates/redline/agent.eval.test.ts +5 -8
  86. package/dist/templates/redline/agent.test.ts +1 -1
  87. package/dist/templates/redline/client.tsx +36 -34
  88. package/dist/templates/redline/workflows/redline.ts +17 -17
  89. package/dist/templates/research-workflow/agent.eval.test.ts +5 -5
  90. package/dist/templates/research-workflow/agent.test.ts +10 -7
  91. package/dist/templates/research-workflow/workflows/research.ts +28 -30
  92. package/dist/templates/retail/agent.test.ts +17 -15
  93. package/dist/templates/retail/cancel.ts +7 -6
  94. package/dist/templates/retail/client.tsx +39 -58
  95. package/dist/templates/retail/payment.ts +4 -3
  96. package/dist/templates/retail/pending.test.ts +19 -19
  97. package/dist/templates/retail/store.test.ts +5 -5
  98. package/dist/templates/retail/swap.ts +8 -11
  99. package/dist/templates/solo-rpg/agent.eval.test.ts +2 -16
  100. package/dist/templates/solo-rpg/agent.test.ts +12 -12
  101. package/dist/templates/solo-rpg/client.tsx +3 -3
  102. package/dist/templates/spoken-summary/agent.test.ts +9 -9
  103. package/dist/templates/spoken-summary/agent.ts +3 -3
  104. package/dist/templates/spoken-summary/client.tsx +23 -30
  105. package/dist/templates/spoken-summary/workflows/summarize.ts +11 -11
  106. package/dist/templates/spoken-summary/workflows/transcribe.ts +14 -14
  107. package/dist/templates/support-line/client.tsx +4 -3
  108. package/dist/templates/support-line/prompts.ts +0 -2
  109. package/dist/templates/transcription-workflow/agent.test.ts +20 -12
  110. package/dist/templates/transcription-workflow/agent.ts +2 -2
  111. package/dist/templates/transcription-workflow/client.tsx +15 -14
  112. package/dist/templates/transcription-workflow/recover.ts +1 -1
  113. package/dist/templates/transcription-workflow/run-panel.tsx +34 -14
  114. package/dist/templates/transcription-workflow/total-latency.tsx +46 -20
  115. package/dist/templates/transcription-workflow/workflows/batch.ts +15 -15
  116. package/dist/templates/transcription-workflow/workflows/downsample.ts +1 -1
  117. package/dist/templates/transcription-workflow/workflows/normalize.ts +15 -19
  118. package/dist/templates/transcription-workflow/workflows/stream.ts +12 -12
  119. package/dist/templates/transcription-workflow/workflows/sync-api.ts +3 -3
  120. package/dist/templates/transcription-workflow/workflows/transcribe.ts +23 -34
  121. package/dist/templates/travel-concierge/agent.eval.test.ts +2 -2
  122. package/dist/templates/travel-concierge/agent.test.ts +15 -7
  123. package/dist/templates/travel-concierge/client.tsx +2 -2
  124. package/dist/templates/travel-concierge/shared.ts +10 -14
  125. package/dist/templates/travel-concierge/tools/lookup_booking.ts +4 -3
  126. package/dist/templates/travel-concierge/tools/search_car_rentals.ts +3 -2
  127. package/dist/templates/travel-concierge/tools/search_excursions.ts +3 -2
  128. package/dist/templates/travel-concierge/tools/search_flights.ts +3 -2
  129. package/dist/templates/travel-concierge/tools/search_hotels.ts +3 -2
  130. package/dist/{test-guenJUxi.mjs → test-D032V2cA.mjs} +4 -17
  131. package/dist/typecheck.mjs +1 -1
  132. package/dist/worker-bundler.mjs +1 -1
  133. package/dist/{workflow-BZyKLkhV.mjs → workflow-CVioyO3V.mjs} +6 -5
  134. package/package.json +4 -4
@@ -18,7 +18,7 @@
18
18
  * - **`stepSpeak`** synthesizes from inside a step. The session TTS surface
19
19
  * cannot: a `TtsSession` is an event stream wired into a live pipeline's
20
20
  * playback, and a step has no turn to be part of and has to return a VALUE.
21
- * - **`writeUpload`** puts that value somewhere. A run's OUTPUT is read back as
21
+ * - **`stepWriteUpload`** puts that value somewhere. A run's OUTPUT is read back as
22
22
  * JSON, so audio cannot travel in one — the same rule that keeps a
23
23
  * recording's bytes out of a run's INPUT, arriving at the other end of the
24
24
  * run. The bytes go to the store, the output carries the id, and the page
@@ -49,9 +49,9 @@
49
49
  * expensive one.
50
50
  */
51
51
 
52
- import type { WorkflowCtx } from "@alexkroman1/aai";
53
- import { report, stepSpeak, TRANSCRIBE_API, writeUpload } from "@alexkroman1/aai/step";
54
- import { stepGenerateJsonClassified } from "@alexkroman1/aai/step-errors";
52
+ import type { WorkflowContext } from "@alexkroman1/aai";
53
+ import { stepReport, stepSpeak, stepWriteUpload, TRANSCRIBE_API } from "@alexkroman1/aai/step";
54
+ import { stepGenerateJsonOrFail } from "@alexkroman1/aai/step-errors";
55
55
  import { countWords, omitUndefined } from "@alexkroman1/aai/utils";
56
56
  // ERASED at build time, so the body can name the schema's own output type without
57
57
  // a runtime cycle back through `agent.ts` — the same mechanism `client.tsx` uses
@@ -128,7 +128,7 @@ export type SpokenSummary = {
128
128
  /** Transcribe a recording, summarize it, and read the summary back. */
129
129
  export async function spokenSummaryFlow(
130
130
  input: WorkflowInputOf<typeof spokenSummary>,
131
- ctx: WorkflowCtx,
131
+ ctx: WorkflowContext,
132
132
  ): Promise<SpokenSummary> {
133
133
  const transcript = await transcribe(input.recording, ctx);
134
134
  const summary = await ctx.step("summarize", () => summarize(transcript.text));
@@ -157,7 +157,7 @@ export async function spokenSummaryFlow(
157
157
  * re-derives exactly the same sequence. It takes the `ctx` for that reason: a
158
158
  * helper that reaches the journal has to be handed the handle.
159
159
  */
160
- async function transcribe(recording: string, ctx: WorkflowCtx): Promise<Transcript> {
160
+ async function transcribe(recording: string, ctx: WorkflowContext): Promise<Transcript> {
161
161
  // `maxAttempts: 6` was `uploadToProvider.maxRetries = 5` — five retries after
162
162
  // the first attempt. It is the one step here worth extra patience: it streams
163
163
  // the whole recording, so a transient failure is expensive to reach again.
@@ -187,8 +187,8 @@ async function transcribe(recording: string, ctx: WorkflowCtx): Promise<Transcri
187
187
  export async function summarize(
188
188
  text: string,
189
189
  ): Promise<{ headline: string; points: string[]; spoken: string }> {
190
- await report("Summarizing the transcript.");
191
- const reply = await stepGenerateJsonClassified(
190
+ await stepReport("Summarizing the transcript.");
191
+ const reply = await stepGenerateJsonOrFail(
192
192
  "Summarize this transcript of a recording.\n\n" +
193
193
  "Answer with JSON only, in this shape:\n" +
194
194
  `{"headline": "...", "points": ["..."], "spoken": "..."}\n\n` +
@@ -204,7 +204,7 @@ export async function summarize(
204
204
  system: "You summarize recordings. You answer with JSON and nothing else.",
205
205
  schema: SummaryReply,
206
206
  },
207
- // The `Classified` caller is `stepGenerateJson` plus `throwStepError`, which
207
+ // The `OrFail` caller is `stepGenerateJson` plus `throwStepError`, which
208
208
  // reads the gateway's own status: a 429 is worth another attempt and a 400
209
209
  // is not, and that is what tells the DevKit which.
210
210
  );
@@ -226,7 +226,7 @@ export async function speak(
226
226
  voice?: string,
227
227
  ): Promise<{ audio: string; durationMs: number }> {
228
228
  const spoken = await stepSpeak(script, omitUndefined({ voice }));
229
- const stored = await writeUpload(spoken.audio, {
229
+ const stored = await stepWriteUpload(spoken.audio, {
230
230
  // Named, because this is what a person sees on the download link rather
231
231
  // than an opaque id — and typed, because the byte route serves the type it
232
232
  // was given and a browser will not play a file it was handed as bytes.
@@ -234,7 +234,7 @@ export async function speak(
234
234
  type: "audio/wav",
235
235
  });
236
236
 
237
- await report(
237
+ await stepReport(
238
238
  `Recorded a ${Math.round(spoken.durationMs / 1000)}s summary in ${spoken.voice}'s voice.`,
239
239
  );
240
240
  return { audio: stored.id, durationMs: spoken.durationMs };
@@ -12,7 +12,7 @@
12
12
  *
13
13
  * **Every one of them is four lines, because the SDK owns the endpoint.**
14
14
  * `stepTranscribeUpload` / `stepTranscribeSubmit` / `stepTranscribePoll` on
15
- * `@alexkroman1/aai/step` — reached here through their `*Classified` callers on
15
+ * `@alexkroman1/aai/step` — reached here through their `*OrFail` callers on
16
16
  * `@alexkroman1/aai/step-errors` — carry the URL, the raw-key auth, the windowed
17
17
  * streaming upload, the PLURAL `speech_models` field and the failure
18
18
  * classification — all of which this file used to spell out, and all of which
@@ -40,11 +40,11 @@
40
40
  * the one leg that should be as boring as possible.
41
41
  */
42
42
 
43
- import { report, uploadInfo } from "@alexkroman1/aai/step";
43
+ import { stepReport, stepUploadInfo } from "@alexkroman1/aai/step";
44
44
  import {
45
- stepTranscribePollClassified,
46
- stepTranscribeSubmitClassified,
47
- stepTranscribeUploadClassified,
45
+ stepTranscribePollOrFail,
46
+ stepTranscribeSubmitOrFail,
47
+ stepTranscribeUploadOrFail,
48
48
  } from "@alexkroman1/aai/step-errors";
49
49
  import { countWords, formatBytes } from "@alexkroman1/aai/utils";
50
50
 
@@ -85,24 +85,24 @@ export type Transcript = {
85
85
  * expires before the next step runs; that costs one fresh upload, once, instead
86
86
  * of five.
87
87
  *
88
- * The `Classified` callers on `@alexkroman1/aai/step-errors` are the SDK's own
88
+ * The `OrFail` callers on `@alexkroman1/aai/step-errors` are the SDK's own
89
89
  * `stepTranscribe*` plus `throwStepError` and nothing else, which is what turns
90
90
  * the SDK's `TranscribeError` into the DevKit's verdict — a missing key and a
91
91
  * 400 stop, a 429 waits as long as the service asked. Every step here ends the
92
92
  * same way for the same reason.
93
93
  */
94
94
  export async function uploadToProvider(uploadId: string): Promise<{ audioUrl: string }> {
95
- const stored = await uploadInfo(uploadId);
96
- await report(
95
+ const stored = await stepUploadInfo(uploadId);
96
+ await stepReport(
97
97
  `Uploading ${stored.name || uploadId} (${formatBytes(stored.size)}) for transcription.`,
98
98
  );
99
- return await stepTranscribeUploadClassified(uploadId);
99
+ return await stepTranscribeUploadOrFail(uploadId);
100
100
  }
101
101
 
102
102
  /** Create the transcription job, and answer with the id that outlives this run. */
103
103
  export async function createJob(audioUrl: string): Promise<{ id: string }> {
104
- const job = await stepTranscribeSubmitClassified(audioUrl);
105
- await report(`Transcribing — job ${job.id}.`);
104
+ const job = await stepTranscribeSubmitOrFail(audioUrl);
105
+ await stepReport(`Transcribing — job ${job.id}.`);
106
106
  return job;
107
107
  }
108
108
 
@@ -119,11 +119,11 @@ export async function pollTranscript(
119
119
  uploadId: string,
120
120
  id: string,
121
121
  ): Promise<{ done: false } | { done: true; transcript: Transcript }> {
122
- const progress = await stepTranscribePollClassified(id);
122
+ const progress = await stepTranscribePollOrFail(id);
123
123
  if (!progress.done) return { done: false };
124
124
 
125
- const stored = await uploadInfo(uploadId);
126
- await report(`Transcribed ${countWords(progress.transcript.text)} words.`);
125
+ const stored = await stepUploadInfo(uploadId);
126
+ await stepReport(`Transcribed ${countWords(progress.transcript.text)} words.`);
127
127
  return {
128
128
  done: true,
129
129
  transcript: {
@@ -1,5 +1,6 @@
1
+ import { plural } from "@alexkroman1/aai/utils";
1
2
  import "@alexkroman1/aai-ui/styles.css";
2
- import { AutoScroll, client, useAgentState } from "@alexkroman1/aai-ui";
3
+ import { AutoScroll, mountClient, useAgentState } from "@alexkroman1/aai-ui";
3
4
  import { PRODUCT, supportProjection } from "./shared.ts";
4
5
 
5
6
  /**
@@ -19,7 +20,7 @@ function TraceSidebar() {
19
20
  <div>
20
21
  <h3 className="text-sm font-bold uppercase tracking-wide opacity-60">{support.product}</h3>
21
22
  <p className="text-xs opacity-50">
22
- {support.asked.length} question{support.asked.length === 1 ? "" : "s"} this call
23
+ {support.asked.length} {plural(support.asked.length, "question")} this call
23
24
  {support.ticket ? ` · ticket ${support.ticket}` : ""}
24
25
  </p>
25
26
  </div>
@@ -97,7 +98,7 @@ function TraceSidebar() {
97
98
  );
98
99
  }
99
100
 
100
- client({
101
+ mountClient({
101
102
  // Derived, not typed twice: `PRODUCT` comes off `knowledge.json`, which is
102
103
  // what `agent.ts` names the agent and greets with. A knowledge base swapped
103
104
  // for another product otherwise leaves the browser tab advertising the old
@@ -40,8 +40,6 @@ export const binaryScore = z.object({
40
40
  reason: z.string().max(200).describe("One short sentence of justification"),
41
41
  });
42
42
 
43
- export type BinaryScore = z.infer<typeof binaryScore>;
44
-
45
43
  /** Their `retrieval_grader`: deliberately NOT a stringent test. */
46
44
  export const DOC_GRADER_SYSTEM = [
47
45
  "You are a grader assessing whether a retrieved support document is relevant",
@@ -20,9 +20,9 @@
20
20
 
21
21
  import { readdir } from "node:fs/promises";
22
22
  import { tmpdir } from "node:os";
23
- import { readUpload, type UploadRange } from "@alexkroman1/aai/step";
23
+ import { stepReadUpload, type UploadRange } from "@alexkroman1/aai/step";
24
24
  import { FatalError, RetryableError } from "@alexkroman1/aai/step-errors";
25
- import { createWorkflowCtx } from "@alexkroman1/aai/testing";
25
+ import { createWorkflowContext } from "@alexkroman1/aai/testing";
26
26
  import {
27
27
  installStubReporter,
28
28
  installStubStepFetch,
@@ -41,6 +41,7 @@ import {
41
41
  requestFormat,
42
42
  } from "./workflows/downsample.ts";
43
43
  import { cuttable, heavierThanNormalized, normalizeRecording } from "./workflows/normalize.ts";
44
+ import { stitchChunks, stitchTranscript, TRANSCRIPT_STREAM } from "./workflows/stitch.ts";
44
45
  import {
45
46
  expectedSegments,
46
47
  nextPollDelay,
@@ -54,9 +55,6 @@ import {
54
55
  import {
55
56
  mergeTranscript,
56
57
  splitRecording,
57
- stitchChunks,
58
- stitchTranscript,
59
- TRANSCRIPT_STREAM,
60
58
  type Transcript,
61
59
  transcribeFlow,
62
60
  transcribeSegment,
@@ -88,9 +86,9 @@ const UPLOAD_ID = "upl_test";
88
86
  const STARTED_AT = 1_000_000;
89
87
 
90
88
  /**
91
- * Publish one in-memory upload, the way `createServer` publishes a real store.
89
+ * Publish one in-memory upload, the way `createRuntimeServer` publishes a real store.
92
90
  *
93
- * This is the seam that makes a step testable at all: `readUpload` reads a
91
+ * This is the seam that makes a step testable at all: `stepReadUpload` reads a
94
92
  * process-wide slot rather than dialling anything, so a spec supplies its own
95
93
  * bytes with no server, no database and no HTTP.
96
94
  */
@@ -898,7 +896,7 @@ describe("transcribeSegment", () => {
898
896
  // What makes the run's answer streamable rather than only its narration: the
899
897
  // page stitches whatever has arrived, so the transcript renders growing
900
898
  // instead of appearing when the last segment does. The reporter is the SDK's
901
- // published slot, which is the same seam `report()` goes through.
899
+ // published slot, which is the same seam `stepReport()` goes through.
902
900
  const reported = installStubReporter();
903
901
  stubProvider();
904
902
 
@@ -964,7 +962,7 @@ describe("transcribeSegment", () => {
964
962
  // `transcribeStreamFlow`'s own `transcribeSegment` call carries the same
965
963
  // budget and is asserted with that flow, not here — this drives only
966
964
  // `transcribeFlow`.
967
- const ctx = createWorkflowCtx({
965
+ const ctx = createWorkflowContext({
968
966
  runSteps: false,
969
967
  results: {
970
968
  normalizeRecording: { recording: UPLOAD_ID, converted: false },
@@ -1116,7 +1114,7 @@ describe("the streaming flow", () => {
1116
1114
 
1117
1115
  test("segmentStored refuses a window that STRADDLES a hole", () => {
1118
1116
  // A run is contiguous, so containment in one is the whole test — and it has to
1119
- // be, because `readUpload` clamps to the run a read starts in. A segment
1117
+ // be, because `stepReadUpload` clamps to the run a read starts in. A segment
1120
1118
  // spanning two runs would come back short and be transcribed as a fragment,
1121
1119
  // which is a wrong transcript rather than a failed one.
1122
1120
  const at = poll(
@@ -1191,12 +1189,12 @@ describe("the streaming flow", () => {
1191
1189
  });
1192
1190
 
1193
1191
  test("a segment reads SHORT rather than failing when its bytes have not landed", async () => {
1194
- // The property the whole flow rests on, and it predates streaming: `readUpload`
1192
+ // The property the whole flow rests on, and it predates streaming: `stepReadUpload`
1195
1193
  // clamps its window to what is stored. So a body that asks slightly early gets
1196
1194
  // what exists — which is why the body checks `end <= size` and can trust the
1197
1195
  // clamp for the final segment of a file that came up short.
1198
1196
  publishPartial(1000, 320_000);
1199
- const slice = await readUpload(UPLOAD_ID, { start: 44, end: 44 + 320_000 });
1197
+ const slice = await stepReadUpload(UPLOAD_ID, { start: 44, end: 44 + 320_000 });
1200
1198
  expect(slice.bytes.length).toBe(1000);
1201
1199
  expect(slice.end).toBe(44 + 1000);
1202
1200
  });
@@ -1356,6 +1354,16 @@ describe("normalizing the recording", () => {
1356
1354
  * later in `splitRecording` with a message about a header; a `timeout`
1357
1355
  * classified as fatal is a run that gives up on work that would have finished.
1358
1356
  */
1357
+ // Point the toolchain at nothing, so this block really is what its doc says:
1358
+ // a unit test that spawns no binary. Without it a developer with ffmpeg
1359
+ // installed runs a different test from CI, which has none — and the one
1360
+ // assertion below that reaches `normalizeRecording` is about a reporter line
1361
+ // emitted BEFORE the spawn. `unstubEnvs` is repo-wide, so there is no teardown.
1362
+ beforeEach(() => {
1363
+ vi.stubEnv("AAI_FFMPEG_PATH", "/nonexistent/ffmpeg");
1364
+ vi.stubEnv("AAI_FFPROBE_PATH", "/nonexistent/ffprobe");
1365
+ });
1366
+
1359
1367
  test("a canonical WAV is cuttable, so the desk converts nothing", () => {
1360
1368
  expect(cuttable(wavFile(MONO_16K, 32_000), 44 + 32_000)).toBe(true);
1361
1369
  });
@@ -40,7 +40,7 @@
40
40
  * `POST /workflows/uploads` (the browser does this for you: `uploads` below is
41
41
  * what makes `<WorkflowFields>` render a file picker, and `useWorkflowSubmit`
42
42
  * stores the file before starting the run), the input carries the returned id,
43
- * and each step reads exactly the window it needs with `readUpload` — which is
43
+ * and each step reads exactly the window it needs with `stepReadUpload` — which is
44
44
  * what keeps sixty steps from moving the same recording sixty times.
45
45
  *
46
46
  * None of that is this template's code. Uploads are the SDK's, for the reason
@@ -118,7 +118,7 @@ export const transcribe = workflow({
118
118
  }),
119
119
  // The one line that makes the form take a file: `<WorkflowFields>` renders a
120
120
  // picker for this property, `useWorkflowSubmit` stores the chosen file, and
121
- // the steps read it back with `readUpload`.
121
+ // the steps read it back with `stepReadUpload`.
122
122
  uploads: ["recording"],
123
123
  run: transcribeFlow,
124
124
  });
@@ -56,7 +56,7 @@
56
56
  * rather than a fourth option. A single request moves a file at one connection's
57
57
  * throughput, which over any distance is a fraction of the link — so the SDK cuts
58
58
  * the file into megabyte-aligned parts and sends four at once. Nothing about the
59
- * workflow changes: the agent reassembles them, `readUpload` reads the same
59
+ * workflow changes: the agent reassembles them, `stepReadUpload` reads the same
60
60
  * windows, and the streaming flow still watches the file grow (what it polls is the
61
61
  * CONTIGUOUS prefix, which is honest whether one connection or four are filling
62
62
  * it).
@@ -83,7 +83,7 @@
83
83
  * - **The page stitches with the RUN's own function.** `stitchChunks` is
84
84
  * `workflows/stitch.ts`, imported by both, so the live text and the stored one
85
85
  * cannot drift into two different transcripts of one recording.
86
- * - **It is a SEPARATE stream from the progress log.** `report()`'s lines go to
86
+ * - **It is a SEPARATE stream from the progress log.** `stepReport()`'s lines go to
87
87
  * the default one, which `<WorkflowProgress>` renders verbatim; objects in
88
88
  * there would come out as `[object Object]` between the sentences.
89
89
  * - **The finished run wins.** Once `output` exists the panel renders that
@@ -193,7 +193,7 @@ import "@alexkroman1/aai-ui/styles.css";
193
193
  import {
194
194
  Form,
195
195
  isTerminal,
196
- page,
196
+ mountPage,
197
197
  SubmitButton,
198
198
  UploadProgressBar,
199
199
  useWorkflowRuns,
@@ -270,10 +270,6 @@ function TranscriptionDesk() {
270
270
  // piece of state for all three hooks, because it describes the UPLOAD and every
271
271
  // mode has one — see the module doc.
272
272
  const [parallel, setParallel] = useState(true);
273
- // Did THIS load press Transcribe? A reload cannot have, and it is the only way
274
- // the page can tell "working on what you just sent" from "picking up where you
275
- // left off" — the hooks report the run, not who asked for it.
276
- const [startedHere, setStartedHere] = useState(false);
277
273
  // ALL THREE hooks are called every render, because a hook may not be conditional —
278
274
  // and that costs nothing here: none of them does anything until its `submit` is
279
275
  // called, and `useWorkflowRun` underneath them holds no id until then either.
@@ -292,7 +288,13 @@ function TranscriptionDesk() {
292
288
  // store — so it is the SAME hook against a different workflow. Only the streaming
293
289
  // mode needs the other one, because only it needs the id before the bytes.
294
290
  const active = mode === "streaming" ? streamed : mode === "batch" ? batched : stored;
295
- const { submitForm, run, upload, pending, error, reset, pauseUpload, resumeUpload } = active;
291
+ // `startedHere` comes off the ACTIVE hook rather than being one flag for the
292
+ // page, and that is a fix rather than a tidy-up: a single piece of page state
293
+ // survived a mode switch, so pressing Transcribe in one mode and then
294
+ // switching made the other mode's panel claim the reader had started its run.
295
+ // Per-hook, switching modes shows the truth about the run being shown.
296
+ const { submitForm, run, upload, pending, error, reset, pauseUpload, resumeUpload, startedHere } =
297
+ active;
296
298
  // History is per WORKFLOW, so the list follows the mode: two flows that produce
297
299
  // the same output are still two different things to have run, and merging them
298
300
  // would put a run under a heading that cannot explain it.
@@ -349,7 +351,6 @@ function TranscriptionDesk() {
349
351
  <Form
350
352
  onSubmit={(values) => {
351
353
  total.start();
352
- setStartedHere(true);
353
354
  // Written at SUBMIT rather than on the radio, so the remembered mode
354
355
  // is always the mode a run exists under — which is the only thing the
355
356
  // next load can use it for.
@@ -369,8 +370,8 @@ function TranscriptionDesk() {
369
370
  </Form>
370
371
 
371
372
  <TotalLatency
372
- elapsedMs={total.elapsedMs}
373
- running={total.running}
373
+ startedAt={total.startedAt}
374
+ frozenMs={total.frozenMs}
374
375
  runMs={run?.status === "completed" ? run.output.elapsedMs : undefined}
375
376
  />
376
377
 
@@ -394,8 +395,8 @@ function TranscriptionDesk() {
394
395
  onClear={() => {
395
396
  // A recovered run is dismissed as deliberately as one this load
396
397
  // started: the lookup is a mount-time act, so `reset()` is not
397
- // undone by a second one and Clear really does clear.
398
- setStartedHere(false);
398
+ // undone by a second one and Clear really does clear. `reset()`
399
+ // also clears `startedHere`, which this page used to mirror here.
399
400
  reset();
400
401
  total.clear();
401
402
  }}
@@ -500,4 +501,4 @@ function UploadPicker({
500
501
  );
501
502
  }
502
503
 
503
- page({ name: "Transcription Desk", component: TranscriptionDesk });
504
+ mountPage({ name: "Transcription Desk", component: TranscriptionDesk });
@@ -14,7 +14,7 @@
14
14
  * `client.tsx` down to the page's shape:
15
15
  *
16
16
  * - **A template's spec is what makes its exemplar code true**, and none of this
17
- * can be tested from a `client.tsx`. That file ends in a `page()` call and
17
+ * can be tested from a `client.tsx`. That file ends in a `mountPage()` call and
18
18
  * imports a stylesheet, and this package's suites have no DOM at all: the
19
19
  * vitest `include` matches `.test.ts` and not `.test.tsx`, and the scaffold
20
20
  * declares no React testing library — so a `client.test.tsx` would be
@@ -21,6 +21,7 @@
21
21
  import { countWords, formatDuration, plural } from "@alexkroman1/aai/utils";
22
22
  import type { WorkflowOutputOf } from "@alexkroman1/aai/workflow-api";
23
23
  import {
24
+ Facts,
24
25
  isTerminal,
25
26
  useWorkflowProgress,
26
27
  WORKFLOW_STATUS_LABELS,
@@ -136,7 +137,7 @@ export function RunPanel({ run, onClear }: { run: WorkflowRun<Transcript>; onCle
136
137
  above, and the reason both exist: the status is `running` for the whole
137
138
  fan-out, so a sixty-segment recording and a one-segment recording look
138
139
  identical while they run. These lines come from the run itself
139
- (`report()` in `workflows/transcribe.ts`), and they REPLAY, so looking a
140
+ (`stepReport()` in `workflows/transcribe.ts`), and they REPLAY, so looking a
140
141
  finished run up in the panel below shows how it got there. */}
141
142
  <WorkflowProgress runId={run.runId} />
142
143
 
@@ -150,15 +151,25 @@ export function RunPanel({ run, onClear }: { run: WorkflowRun<Transcript>; onCle
150
151
  object with optional fields. */}
151
152
  {run.status === "completed" && (
152
153
  <>
153
- <p className="text-xs opacity-60">
154
- {run.output.segments} {plural(run.output.segments, "segment")} ·{" "}
155
- {formatDuration(run.output.durationMs)} of audio · took{" "}
156
- {formatDuration(run.output.elapsedMs)} · {run.output.words} words
157
- </p>
154
+ <Facts
155
+ size="xs"
156
+ items={[
157
+ `${run.output.segments} ${plural(run.output.segments, "segment")}`,
158
+ `${formatDuration(run.output.durationMs)} of audio`,
159
+ `took ${formatDuration(run.output.elapsedMs)}`,
160
+ `${run.output.words} words`,
161
+ ]}
162
+ />
158
163
  <pre className="whitespace-pre-wrap text-sm leading-relaxed">{run.output.transcript}</pre>
159
164
  </>
160
165
  )}
161
- {run.status === "failed" && <p className="text-red-600">{run.error}</p>}
166
+ {/* `role="alert"`, the same contract `<Form>` gives the submit error in
167
+ `client.tsx`: this is the outcome the reader waited minutes for. */}
168
+ {run.status === "failed" && (
169
+ <p role="alert" className="text-red-600">
170
+ {run.error}
171
+ </p>
172
+ )}
162
173
  </section>
163
174
  );
164
175
  }
@@ -186,17 +197,26 @@ function LiveTranscript({ runId }: { runId: string }) {
186
197
  // Memoized on the ARRAY, which the hook appends to per read: stitching is a
187
198
  // seam search per segment, and a fan-out re-renders this panel on every
188
199
  // progress poll whether or not anything arrived.
189
- const transcript = useMemo(() => stitchChunks(progress), [progress]);
200
+ //
201
+ // The word count and the coverage are derived in the SAME memo, not below
202
+ // it: both are functions of `progress` alone, and `countWords` rescans a
203
+ // transcript that reaches ~14k words on a feature-length recording.
204
+ //
205
+ // `covered` is the furthest point reached, not the count: segments land out
206
+ // of order, so "6 segments" says nothing about how much is covered.
207
+ const { transcript, words, covered } = useMemo(() => {
208
+ const stitched = stitchChunks(progress);
209
+ return {
210
+ transcript: stitched,
211
+ words: countWords(stitched),
212
+ covered: progress.reduce((furthest, chunk) => Math.max(furthest, chunk.endMs), 0),
213
+ };
214
+ }, [progress]);
190
215
  if (progress.length === 0) return null;
191
216
 
192
- // The furthest point reached, not the count: segments land out of order, so
193
- // "6 segments" says nothing about how much of the recording is covered.
194
- const covered = Math.max(...progress.map((chunk) => chunk.endMs));
195
217
  return (
196
218
  <div className="flex flex-col gap-2">
197
- <p className="text-xs opacity-60">
198
- {countWords(transcript)} words so far · through {formatDuration(covered)}
199
- </p>
219
+ <Facts size="xs" items={[`${words} words so far`, `through ${formatDuration(covered)}`]} />
200
220
  <pre className="whitespace-pre-wrap text-sm leading-relaxed opacity-80">{transcript}</pre>
201
221
  </div>
202
222
  );
@@ -15,6 +15,7 @@
15
15
  */
16
16
 
17
17
  import { formatDuration } from "@alexkroman1/aai/utils";
18
+ import { Facts } from "@alexkroman1/aai-ui";
18
19
  import { useCallback, useEffect, useRef, useState } from "react";
19
20
 
20
21
  /**
@@ -29,12 +30,17 @@ const STOPWATCH_TICK_MS = 250;
29
30
  /** What {@link useTotalLatency} reports. */
30
31
  export type TotalLatency = {
31
32
  /**
32
- * Milliseconds since the submit ticking while the submission is in flight,
33
- * frozen at the finish, and undefined before the first one.
33
+ * The mark the clock started at, or undefined before the first submit.
34
+ *
35
+ * The raw mark rather than an elapsed duration, and that is the load-bearing
36
+ * part: an elapsed number is stale the instant it is computed, so a page
37
+ * holding one has to re-render to keep it true — which is what used to drive
38
+ * the whole tree at 4 Hz. {@link TotalLatency} turns these two marks into a
39
+ * duration, and only that section re-renders.
34
40
  */
35
- elapsedMs: number | undefined;
36
- /** Whether the clock is still running, which is what makes the label honest. */
37
- running: boolean;
41
+ startedAt: number | undefined;
42
+ /** The elapsed time at the moment the run settled, once it has. */
43
+ frozenMs: number | undefined;
38
44
  /** Start (or restart) the clock. Called from the form's own submit handler. */
39
45
  start: () => void;
40
46
  /** Drop it, for a panel that no longer describes the submission it timed. */
@@ -53,7 +59,10 @@ export type TotalLatency = {
53
59
  *
54
60
  * - **The interval re-renders; it does not accumulate.** The elapsed time is read
55
61
  * from the clock at render, so a tick the tab throttled or dropped cannot make
56
- * the number lag behind real time.
62
+ * the number lag behind real time. The interval lives in {@link TotalLatency}
63
+ * rather than here, because a hook called from the page makes its ticks the
64
+ * PAGE's re-renders — the form, the upload bar, the run panel and the whole
65
+ * history list, four times a second, to move one duration string.
57
66
  * - **`performance.now()`, not `Date.now()`.** It is monotonic, so a clock
58
67
  * correction (NTP, a laptop waking up) cannot make a transcription look
59
68
  * instant — or negative.
@@ -61,8 +70,6 @@ export type TotalLatency = {
61
70
  export function useTotalLatency(inFlight: boolean): TotalLatency {
62
71
  const [startedAt, setStartedAt] = useState<number | undefined>(undefined);
63
72
  const [frozenMs, setFrozenMs] = useState<number | undefined>(undefined);
64
- // Re-render trigger only — see the doc above.
65
- const [, tick] = useState(0);
66
73
  // Whether `inFlight` has been seen true since the last `start()`. Without it,
67
74
  // a start that lands one render before the submission reports itself in flight
68
75
  // would freeze the clock at zero instead of running it.
@@ -72,8 +79,7 @@ export function useTotalLatency(inFlight: boolean): TotalLatency {
72
79
  if (startedAt === undefined || frozenMs !== undefined) return;
73
80
  if (inFlight) {
74
81
  began.current = true;
75
- const id = setInterval(() => tick((n) => n + 1), STOPWATCH_TICK_MS);
76
- return () => clearInterval(id);
82
+ return;
77
83
  }
78
84
  // Measured here rather than at render, so the frozen number is the one at the
79
85
  // moment the run settled rather than whenever this page next drew.
@@ -93,8 +99,8 @@ export function useTotalLatency(inFlight: boolean): TotalLatency {
93
99
  }, []);
94
100
 
95
101
  return {
96
- elapsedMs: frozenMs ?? (startedAt === undefined ? undefined : performance.now() - startedAt),
97
- running: startedAt !== undefined && frozenMs === undefined,
102
+ startedAt,
103
+ frozenMs,
98
104
  start,
99
105
  clear,
100
106
  };
@@ -114,17 +120,31 @@ export function useTotalLatency(inFlight: boolean): TotalLatency {
114
120
  * notices it finished. Clamped at zero, because the two numbers come from two
115
121
  * different clocks on two different machines and a few milliseconds the wrong way
116
122
  * would otherwise print a negative.
123
+ *
124
+ * The stopwatch interval lives HERE rather than in {@link useTotalLatency}, so a
125
+ * running clock re-renders this section and nothing else.
117
126
  */
118
127
  export function TotalLatency({
119
- elapsedMs,
120
- running,
128
+ startedAt,
129
+ frozenMs,
121
130
  runMs,
122
131
  }: {
123
- elapsedMs: number | undefined;
124
- running: boolean;
132
+ startedAt: number | undefined;
133
+ frozenMs: number | undefined;
125
134
  runMs: number | undefined;
126
135
  }) {
127
- if (elapsedMs === undefined) return null;
136
+ // Re-render trigger only: the elapsed time is read from the clock below, so a
137
+ // tick the tab throttled or dropped cannot make the number lag real time.
138
+ const [, tick] = useState(0);
139
+ const running = startedAt !== undefined && frozenMs === undefined;
140
+ useEffect(() => {
141
+ if (!running) return;
142
+ const id = setInterval(() => tick((n) => n + 1), STOPWATCH_TICK_MS);
143
+ return () => clearInterval(id);
144
+ }, [running]);
145
+
146
+ if (startedAt === undefined) return null;
147
+ const elapsedMs = frozenMs ?? performance.now() - startedAt;
128
148
  const outside = runMs === undefined ? undefined : Math.max(0, elapsedMs - runMs);
129
149
  return (
130
150
  <section className="flex flex-wrap items-baseline justify-between gap-x-4 gap-y-1 rounded-md border px-5 py-3">
@@ -134,9 +154,15 @@ export function TotalLatency({
134
154
  <span className="flex flex-wrap items-baseline gap-x-3 gap-y-1">
135
155
  <span className="text-sm tabular-nums">{formatDuration(elapsedMs)}</span>
136
156
  {runMs !== undefined && outside !== undefined && (
137
- <span className="text-xs tabular-nums opacity-60">
138
- {formatDuration(outside)} before the run · {formatDuration(runMs)} inside it
139
- </span>
157
+ <Facts
158
+ size="xs"
159
+ as="span"
160
+ className="tabular-nums"
161
+ items={[
162
+ `${formatDuration(outside)} before the run`,
163
+ `${formatDuration(runMs)} inside it`,
164
+ ]}
165
+ />
140
166
  )}
141
167
  </span>
142
168
  </section>