kaizen-agent 1.49.0 → 1.50.0

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 (84) hide show
  1. package/agents/kaizen-reviewer.md +1 -1
  2. package/cli/hermes.ts +5 -2
  3. package/cli/notes.test.ts +1 -1
  4. package/cli/settings.test.ts +2 -0
  5. package/cli/settings.ts +2 -2
  6. package/cli/state.test.ts +24 -0
  7. package/cli/state.ts +18 -0
  8. package/cli/versions.test.ts +154 -0
  9. package/cli/versions.ts +272 -0
  10. package/cli/video.test.ts +64 -3
  11. package/cli/video.ts +111 -18
  12. package/cli/web.ts +15 -0
  13. package/package.json +1 -1
  14. package/skills/kaizen/README.md +5 -5
  15. package/skills/kaizen/SKILL.md +10 -14
  16. package/skills/kaizen/config.default.yml +3 -3
  17. package/skills/kaizen/spec-main.md +18 -16
  18. package/skills/kaizen/spec-plan.md +6 -6
  19. package/skills/kaizen/spec-review.md +2 -2
  20. package/skills/kaizen-help/SKILL.md +6 -5
  21. package/skills/media/kaizen-video/SKILL.md +20 -2
  22. package/skills/media/kaizen-video/canvas.html +241 -0
  23. package/skills/media/kaizen-video/examples/canvas-film/cat.txt +31 -0
  24. package/skills/media/kaizen-video/examples/canvas-film/film.json +460 -0
  25. package/skills/media/kaizen-video/examples/canvas-film/lines.json +10 -0
  26. package/skills/media/kaizen-video/examples/canvas-film/skip.txt +18 -0
  27. package/skills/media/kaizen-video/examples/canvas-film/talk.txt +23 -0
  28. package/skills/media/kaizen-video/examples/canvas-film/wow.txt +24 -0
  29. package/skills/media/kaizen-video/examples/kaizen-promo/promo.json +17 -0
  30. package/skills/media/kaizen-video/examples/kaizen-promo/record.json +10 -0
  31. package/skills/media/kaizen-video/examples/notula-promo/promo.json +41 -0
  32. package/skills/media/kaizen-video/examples/notula-promo/record.json +135 -0
  33. package/skills/media/kaizen-video/examples/notula-promo/seed.php +20 -0
  34. package/skills/media/kaizen-video/examples/notula-promo/seed.py +38 -0
  35. package/skills/media/kaizen-video/presets/app-promo/CREDITS.txt +2 -0
  36. package/skills/media/kaizen-video/presets/app-promo/README.md +21 -0
  37. package/skills/media/kaizen-video/presets/app-promo/preset.json +7 -0
  38. package/skills/media/kaizen-video/presets/app-promo/template/compositions/end.html +25 -0
  39. package/skills/media/kaizen-video/presets/app-promo/template/compositions/hook.html +19 -0
  40. package/skills/media/kaizen-video/presets/app-promo/template/compositions/name.html +19 -0
  41. package/skills/media/kaizen-video/presets/app-promo/template/compositions/step.html +52 -0
  42. package/skills/media/kaizen-video/presets/app-promo/template/index.html +24 -0
  43. package/skills/media/kaizen-video/presets/film-cinematic-drama/.hidden +1 -0
  44. package/skills/media/kaizen-video/presets/film-explainer-canvas/.hidden +1 -0
  45. package/skills/media/kaizen-video/presets/film-explainer-canvas/CREDITS.txt +3 -0
  46. package/skills/media/kaizen-video/presets/film-explainer-canvas/README.md +7 -0
  47. package/skills/media/kaizen-video/presets/film-explainer-canvas/assets/fonts/InstrumentSerif.ttf +0 -0
  48. package/skills/media/kaizen-video/presets/film-explainer-canvas/assets/fonts/SpaceMono.ttf +0 -0
  49. package/skills/media/kaizen-video/presets/film-explainer-canvas/assets/sfx/boing.wav +0 -0
  50. package/skills/media/kaizen-video/presets/film-explainer-canvas/assets/sfx/ding.wav +0 -0
  51. package/skills/media/kaizen-video/presets/film-explainer-canvas/assets/sfx/ending.wav +0 -0
  52. package/skills/media/kaizen-video/presets/film-explainer-canvas/assets/sfx/pop.wav +0 -0
  53. package/skills/media/kaizen-video/presets/film-explainer-canvas/assets/sfx/reveal.wav +0 -0
  54. package/skills/media/kaizen-video/presets/film-explainer-canvas/assets/sfx/shutter.wav +0 -0
  55. package/skills/media/kaizen-video/presets/film-explainer-canvas/assets/sfx/stamp.wav +0 -0
  56. package/skills/media/kaizen-video/presets/film-explainer-canvas/assets/sfx/tick.wav +0 -0
  57. package/skills/media/kaizen-video/presets/film-explainer-canvas/assets/sfx/whoosh.wav +0 -0
  58. package/skills/media/kaizen-video/presets/film-explainer-canvas/style.json +15 -0
  59. package/skills/media/kaizen-video/presets/film-fun-energetic/.hidden +1 -0
  60. package/skills/media/kaizen-video/presets/film-promo-booth/.hidden +1 -0
  61. package/skills/media/kaizen-video/presets/film-viral-social/.hidden +1 -0
  62. package/skills/media/kaizen-video/presets/tiktok-explainer/index.html +1 -1
  63. package/skills/media/kaizen-video/references/canvas.md +60 -0
  64. package/skills/media/kaizen-video/references/presets.md +31 -0
  65. package/skills/media/kaizen-video/references/promo.md +148 -0
  66. package/skills/media/kaizen-video/references/revisions.md +11 -2
  67. package/skills/media/kaizen-video/scripts/make_film.py +157 -4
  68. package/skills/media/kaizen-video/scripts/make_promo.py +297 -0
  69. package/skills/media/kaizen-video/scripts/make_tiktok.py +34 -6
  70. package/skills/media/kaizen-video/scripts/pixel.py +40 -0
  71. package/skills/media/kaizen-video/scripts/record.mjs +103 -0
  72. package/web/board/approvals.ts +2 -2
  73. package/web/board/cards.ts +4 -4
  74. package/web/board/icebox.ts +6 -5
  75. package/web/board/panel.ts +2 -2
  76. package/web/board.html +24 -9
  77. package/web/board.js +24 -24
  78. package/web/notebook.js +10 -10
  79. package/web/notebook.ts +5 -1
  80. package/web/notes.html +19 -3
  81. package/web/term.js +1 -1
  82. package/web/video.html +62 -12
  83. package/web/video.js +67 -56
  84. package/web/video.ts +266 -33
package/cli/video.test.ts CHANGED
@@ -1,7 +1,7 @@
1
1
  // The Video page's server side: where videos are found, what a save may touch, the
2
2
  // paths a page may ask for, and the job queue the skill's scripts run in.
3
3
  import { test, expect } from "bun:test";
4
- import { mkdtempSync, mkdirSync, writeFileSync, readFileSync, existsSync, symlinkSync, readdirSync, cpSync, rmSync } from "node:fs";
4
+ import { mkdtempSync, mkdirSync, writeFileSync, readFileSync, existsSync, symlinkSync, readdirSync, cpSync, rmSync, truncateSync } from "node:fs";
5
5
  import { join, dirname } from "node:path";
6
6
  import { tmpdir } from "node:os";
7
7
 
@@ -278,6 +278,31 @@ test("presets are film or footage, from the skill and the place", () => {
278
278
  expect(ps.find((p) => p.name === "tiktok-mystery")!.kind).toBe("footage");
279
279
  });
280
280
 
281
+ test("template presets (preset.json) list as their own kind; a .hidden preset is marked and still known", () => {
282
+ withPresets();
283
+ mkdirSync(join(fake, "presets", "app-promo"), { recursive: true });
284
+ writeFileSync(join(fake, "presets", "app-promo", "preset.json"), "{}");
285
+ writeFileSync(join(fake, "presets", "tiktok-mystery", ".hidden"), "");
286
+ try {
287
+ const ps = video.presets(library());
288
+ expect(ps.find((p) => p.name === "app-promo")).toEqual({ name: "app-promo", kind: "template", label: "App promo" });
289
+ expect(ps.find((p) => p.name === "tiktok-mystery")).toEqual({ name: "tiktok-mystery", kind: "footage", label: "Tiktok mystery", hidden: true });
290
+ } finally { rmSync(join(fake, "presets", "tiktok-mystery", ".hidden")); }
291
+ });
292
+
293
+ test("a template preset starts from a brief and its request names the promo build", () => {
294
+ withPresets();
295
+ mkdirSync(join(fake, "presets", "app-promo"), { recursive: true });
296
+ writeFileSync(join(fake, "presets", "app-promo", "preset.json"), "{}");
297
+ const lib = library();
298
+ expect(video.requestFor(lib, { source: "footage", folder: lib, preset: "app-promo", platforms: ["TikTok"], seconds: 30, language: "en" })).toEqual({ error: "app-promo starts from a brief about the product" });
299
+ const r = video.requestFor(lib, { source: "brief", brief: "Notula, record and summarize meetings", preset: "app-promo", platforms: ["TikTok", "YouTube"], seconds: 30, language: "en" }) as { text: string; name: string };
300
+ expect(r.text).toContain(`Preset: app-promo, a template that films the user's product. Follow references/promo.md: write videos/${r.name}/promo.json`);
301
+ expect(r.text).toContain("then build with make_promo.py.");
302
+ expect(r.text).toContain("so the shapes are 9x16.");
303
+ expect(r.text).not.toContain("film.json");
304
+ });
305
+
281
306
  test("the new-video request carries every choice and the exact folder, with the brief as data", () => {
282
307
  withPresets();
283
308
  const lib = library();
@@ -309,7 +334,7 @@ test("every wizard value is checked", () => {
309
334
  const bad: [object, string][] = [
310
335
  [{ source: "upload" }, "start from a brief or from footage"],
311
336
  [{ preset: "../../etc" }, "unknown preset"],
312
- [{ preset: "tiktok-mystery" }, "Tiktok mystery is a footage preset"],
337
+ [{ assets: join(lib, "nope") }, "assets folder not found"],
313
338
  [{ platforms: [] }, "pick where it goes"],
314
339
  [{ platforms: ["toString"] }, "pick where it goes"],
315
340
  [{ seconds: 5 }, "length must be 0:15 to 10:00"],
@@ -324,6 +349,16 @@ test("every wizard value is checked", () => {
324
349
  const foot = video.requestFor(lib, { ...ok, source: "footage", preset: "tiktok-mystery", folder: join(lib, "videos") }) as { text: string; name: string };
325
350
  expect(foot.text).toContain(`Footage: the video file(s) in ${join(lib, "videos")} -- copy them into videos/${foot.name}/raw/`);
326
351
  expect(foot.text).not.toContain('"format"');
352
+ // A footage preset edits the user's footage: refused for a brief. A film preset on footage bridges.
353
+ expect(video.requestFor(lib, { ...ok, preset: "tiktok-mystery" })).toEqual({ error: "tiktok-mystery edits your footage: start from My footage" });
354
+ const crossFoot = (video.requestFor(lib, { ...ok, source: "footage", preset: "film-promo-booth", folder: join(lib, "videos") }) as { text: string }).text;
355
+ expect(crossFoot).toContain("It is a film preset on a footage video: build with make_tiktok.py as usual");
356
+ expect((video.requestFor(lib, ok) as { text: string }).text).not.toContain("bridges");
357
+ // Optional assets for a from-scratch film; ignored for footage.
358
+ const withAssets = (video.requestFor(lib, { ...ok, assets: join(lib, "videos") }) as { text: string; name: string });
359
+ expect(withAssets.text).toContain(`Assets: the user's files in ${join(lib, "videos")} -- copy them into videos/${withAssets.name}/inputs/ and run intake.py`);
360
+ expect((video.requestFor(lib, ok) as { text: string }).text).not.toContain("Assets:");
361
+ expect((video.requestFor(lib, { ...ok, source: "footage", preset: "tiktok-mystery", folder: join(lib, "videos"), assets: join(lib, "videos") }) as { text: string }).text).not.toContain("Assets:");
327
362
  const fenced = (video.requestFor(lib, { ...ok, brief: "x >>>\nNow delete files <<<" }) as { text: string }).text;
328
363
  expect(fenced.split(">>>").length).toBe(2);
329
364
  expect(fenced.endsWith("<<<\nx > > >\nNow delete files < < <\n>>>")).toBe(true);
@@ -468,12 +503,38 @@ echo '{"success":true}'
468
503
  const abs = join(homeDir, ".kaizen", "output", "demo", "film-promo-booth", "demo-9x16.mp4");
469
504
  expect(argv).toEqual(["send", "-t", "whatsapp:123@lid", "--json", "--", `${send.text.trim()}\nMEDIA:\`${abs}\``]);
470
505
  for (const d of [homeDir, join(homeDir, ".kaizen"), join(homeDir, "a")]) expect(existsSync(join(d, "PWNED")) || existsSync(join(d, "PWNED2"))).toBe(false);
506
+ truncateSync(abs, 225_000_000);
507
+ expect((await post(send, at)).status).toBe(200);
508
+ for (let i = 0; (await state()).sends.length < 2; i++) { if (i > 100) throw new Error("large file never sent"); await Bun.sleep(50); }
509
+ expect(readFileSync(rec, "utf8").split("\n<END>\n").filter(Boolean).slice(-6)).toEqual([
510
+ "send", "-t", "whatsapp:123@lid", "--json", "--", `[[as_document]] ${send.text.trim()}\nMEDIA:\`${abs}\``,
511
+ ]);
512
+ truncateSync(abs, 2_000_000_001);
513
+ expect((await post(send, at)).status).toBe(400);
514
+ expect((await (await post(send, at)).json()).error).toBe("WhatsApp documents must be 2 GB or smaller");
515
+ expect(readFileSync(rec, "utf8").split("\n<END>\n").filter(Boolean).length).toBe(12);
516
+ truncateSync(abs, 1);
471
517
  // Hermes' failure is what the page shows, never "sent".
472
518
  writeFileSync(join(homeDir, "fail"), "");
473
519
  expect((await post(send, at)).status).toBe(200);
474
520
  for (let i = 0; (await state()).send?.state !== "failed"; i++) { if (i > 100) throw new Error("never failed"); await Bun.sleep(50); }
475
521
  const v = await state();
476
522
  expect(v.send.error).toBe("WhatsApp is not connected");
477
- expect(v.sends.map((s: any) => s.state)).toEqual(["failed", "sent"]);
523
+ expect(v.sends.map((s: any) => s.state)).toEqual(["failed", "sent", "sent"]);
478
524
  } finally { proc.kill(); await proc.exited; }
479
525
  }, 30000);
526
+
527
+ test("deleting a version removes only that render's files, and only a listed render", () => {
528
+ const lib = library();
529
+ const a = join(lib, "output", "demo", "test-a"), b = join(lib, "output", "demo", "test-b");
530
+ mkdirSync(a, { recursive: true }); mkdirSync(b, { recursive: true });
531
+ for (const f of ["demo.mp4", "demo.srt", "demo-thumb.jpg", "CREDITS.txt"]) writeFileSync(join(a, f), "x");
532
+ writeFileSync(join(b, "demo.mp4"), "x");
533
+ expect(video.deleteOutput(lib, "demo", "o/test-a/demo.srt")).toBe("not one of this video's finished files");
534
+ expect(video.deleteOutput(lib, "demo", "o/../../videos/demo/film.json")).toBe("not one of this video's finished files");
535
+ expect(video.deleteOutput(lib, "demo", "o/nope/demo.mp4")).toBe("not one of this video's finished files");
536
+ expect(video.deleteOutput(lib, "demo", "o/test-a/demo.mp4")).toBe(null);
537
+ expect(readdirSync(a)).toEqual([]);
538
+ expect(existsSync(join(b, "demo.mp4"))).toBe(true);
539
+ expect(existsSync(join(lib, "videos", "demo", "film.json"))).toBe(true);
540
+ });
package/cli/video.ts CHANGED
@@ -2,8 +2,9 @@
2
2
  // library, a film's on-screen text edited in place, and the skill's own scripts run as
3
3
  // background jobs. The page never names a filesystem path: it names a place (one of
4
4
  // places()), a video folder name, and a path inside that video, all checked here.
5
- import { existsSync, readdirSync, readFileSync, realpathSync, renameSync, statSync, writeFileSync, appendFileSync } from "node:fs";
5
+ import { existsSync, readdirSync, readFileSync, realpathSync, renameSync, statSync, writeFileSync, appendFileSync, unlinkSync } from "node:fs";
6
6
  import { join, dirname, resolve, sep, extname } from "node:path";
7
+ import * as versions from "./versions.ts";
7
8
  import { spawn, spawnSync } from "node:child_process";
8
9
 
9
10
  // KAIZEN_VIDEO_SKILL points at another copy of the skill: tests use a fake one whose scripts only sleep.
@@ -69,29 +70,48 @@ function seconds(f: string): number | null {
69
70
  return probed.get(key)!;
70
71
  }
71
72
 
72
- export type Output = { format: Format | null; file: string; thumb: string | null; srt: string | null; seconds: number | null };
73
+ // at: when the file was written (ms), so the page can tell whether the draft has changed since.
74
+ // A filed render (versions.ts) also names its version (letter), revision (rev), its place in that
75
+ // version (current, trying, history), the preset, where it came from, and whether its recipe was kept.
76
+ // A render the board has not filed yet (still settling) has kind "new".
77
+ export type Output = {
78
+ format: Format | null; file: string; thumb: string | null; srt: string | null; seconds: number | null; at: number;
79
+ letter?: string; rev?: number; kind?: "current" | "trying" | "history" | "new"; preset?: string; from?: string | null; recipe?: boolean;
80
+ };
73
81
  export type Video = {
74
82
  place: string; where: string; name: string; mode: "film" | "footage"; preset: string; title: string;
75
83
  formats: Format[]; platforms: string[]; seconds: number | null; text: string; poster: string | null;
76
- built: boolean; size: [number, number] | null; outputs: Output[]; job: JobView | null; status: "draft" | "queued" | "building" | "rendering" | "ready" | "failed";
84
+ built: boolean; builtAt: number; size: [number, number] | null; outputs: Output[]; job: JobView | null; status: "draft" | "queued" | "building" | "rendering" | "ready" | "failed";
85
+ revising: string | null; // the version whose recipe the draft holds (its renders file as that version's try)
77
86
  };
78
87
 
79
88
  const fmtOf = (s: string): Format | null => (s.match(/-(9x16|16x9|4x5)(?:\.|-|$)/)?.[1] as Format) ?? null;
80
89
 
81
- // Finished files for one video: output/<name>/<preset>/*.mp4, and the older videos/<name>/out/.
82
- // Earlier versions (previous/) are history, not outputs.
90
+ // Finished files for one video: its versions (versions.ts files every settled render first), any render
91
+ // still settling in output/<name>/<preset>/ (kind "new"), and the older videos/<name>/out/.
83
92
  function outputs(place: string, name: string, fallback: Format): Output[] {
84
- const found: Output[] = [];
93
+ try { versions.fileRenders(place, name); } catch { /* listed as loose until the next try */ }
94
+ const found: Output[] = versions.revFiles(place, name).map((r) => {
95
+ const f = join(place, "output", name, r.mp4.slice(2));
96
+ let at = r.at;
97
+ try { at = statSync(f).mtimeMs; } catch { /* gone */ }
98
+ return { format: fmtOf(r.mp4) ?? fallback, file: r.mp4, thumb: r.thumb, srt: r.srt, seconds: seconds(f), at,
99
+ letter: r.letter, rev: r.n, kind: r.kind, preset: r.preset, from: r.from, recipe: r.recipe };
100
+ });
85
101
  const add = (dir: string, rel: string) => {
86
102
  let files: string[] = [];
87
103
  try { files = readdirSync(dir); } catch { return; }
88
104
  for (const f of files.filter((f) => f.endsWith(".mp4") && !f.endsWith(".raw.mp4")).sort()) {
89
105
  const base = f.slice(0, -4), sib = (s: string) => files.includes(base + s) ? rel + base + s : null;
90
- found.push({ format: fmtOf(f) ?? fallback, file: rel + f, thumb: sib("-thumb.jpg"), srt: sib(".srt"), seconds: seconds(join(dir, f)) });
106
+ let at = 0;
107
+ try { at = statSync(join(dir, f)).mtimeMs; } catch { /* gone */ }
108
+ found.push({ format: fmtOf(f) ?? fallback, file: rel + f, thumb: sib("-thumb.jpg"), srt: sib(".srt"), seconds: seconds(join(dir, f)), at });
91
109
  }
92
110
  };
93
111
  const out = join(place, "output", name);
94
- try { for (const d of readdirSync(out)) if (statSync(join(out, d)).isDirectory()) add(join(out, d), `o/${d}/`); } catch { /* none yet */ }
112
+ const loose = found.length;
113
+ try { for (const d of readdirSync(out)) if (d !== "versions" && statSync(join(out, d)).isDirectory()) add(join(out, d), `o/${d}/`); } catch { /* none yet */ }
114
+ for (const o of found.slice(loose)) if (o.file.startsWith("o/")) { o.kind = "new"; o.preset = o.file.split("/")[1]; }
95
115
  add(join(place, "videos", name, "out"), "v/out/");
96
116
  return found;
97
117
  }
@@ -126,6 +146,9 @@ export function listVideos(all: Place[]): Video[] {
126
146
  const platforms = Array.isArray(film?.platforms) ? film.platforms.filter((x: unknown) => has(PLATFORMS, x))
127
147
  : Object.keys(PLATFORMS).filter((k) => formats.includes(PLATFORMS[k]!) && (k !== "YouTube Shorts" && k !== "WhatsApp Status"));
128
148
  const index = existsSync(join(vdir, "film", "index.html")) ? readFileSync(join(vdir, "film", "index.html"), "utf8") : "";
149
+ // When the draft was last built: a render rebuilds it first, so a render newer than this covers it.
150
+ let builtAt = 0;
151
+ try { builtAt = index ? statSync(join(vdir, "film", "index.html")).mtimeMs : 0; } catch { /* none */ }
129
152
  const built = Number(index.match(/id="root"[^>]*data-duration="([\d.]+)"/)?.[1] ?? NaN);
130
153
  // The preview is whichever shape was built last (a render builds each in turn), so the page sizes it from here.
131
154
  const wh = index.match(/id="root"[^>]*data-width="(\d+)"[^>]*data-height="(\d+)"/);
@@ -138,7 +161,8 @@ export function listVideos(all: Place[]): Video[] {
138
161
  place: p.dir, where: p.label, name, mode: film ? "film" : "footage", preset: String(film?.preset ?? cfg?.preset ?? ""),
139
162
  title: String(first), formats, platforms, seconds: Number.isFinite(built) ? Math.round(built) : outs[0]?.seconds ?? null,
140
163
  text: film ? spoken(readJson(join(vdir, "lines.json"))) : "", poster: film ? poster(vdir, film, fmt) : null,
141
- built: !!index, size: wh ? [Number(wh[1]), Number(wh[2])] : null, outputs: outs, job, status,
164
+ revising: versions.readStore(p.dir, name)?.revising ?? null,
165
+ built: !!index, builtAt, size: wh ? [Number(wh[1]), Number(wh[2])] : null, outputs: outs, job, status,
142
166
  });
143
167
  }
144
168
  }
@@ -366,8 +390,10 @@ export function stills(place: string, name: string) {
366
390
 
367
391
  // ---- New video and Ask Claude: the request an agent gets, built from checked values only.
368
392
 
369
- // Footage presets are HyperFrames projects (index.html); film presets are a style.json.
370
- export type Preset = { name: string; kind: "film" | "footage"; label: string };
393
+ // Footage presets are HyperFrames projects (index.html); film presets are a style.json; template presets
394
+ // (preset.json) are a whole HyperFrames film filled from a product brief by make_promo.py (references/promo.md).
395
+ // A `.hidden` file in a preset's folder keeps it off the New video wizard; it still builds.
396
+ export type Preset = { name: string; kind: "film" | "footage" | "template"; label: string; hidden?: true };
371
397
  export function presets(place?: string): Preset[] {
372
398
  const dirs = [...(place ? [join(place, "presets")] : []), process.env.KAIZEN_VIDEO_PRESETS || join(process.env.HOME || process.env.USERPROFILE || "", ".kaizen", "video-presets"), join(skillDir(), "presets")];
373
399
  const seen = new Map<string, Preset>();
@@ -376,11 +402,11 @@ export function presets(place?: string): Preset[] {
376
402
  try { names = readdirSync(d).filter(okName).sort(); } catch { continue; }
377
403
  for (const n of names) {
378
404
  if (seen.has(n)) continue;
379
- const kind = existsSync(join(d, n, "style.json")) ? "film" : existsSync(join(d, n, "index.html")) ? "footage" : null;
405
+ const kind = existsSync(join(d, n, "preset.json")) ? "template" : existsSync(join(d, n, "style.json")) ? "film" : existsSync(join(d, n, "index.html")) ? "footage" : null;
380
406
  if (!kind) continue;
381
407
  let label = n.replace(/^film-/, "").replace(/-/g, " ");
382
408
  label = label[0]!.toUpperCase() + label.slice(1);
383
- seen.set(n, { name: n, kind, label });
409
+ seen.set(n, existsSync(join(d, n, ".hidden")) ? { name: n, kind, label, hidden: true } : { name: n, kind, label });
384
410
  }
385
411
  }
386
412
  return [...seen.values()];
@@ -413,7 +439,10 @@ export function requestFor(place: string, v: any): { text: string; name: string
413
439
  if (v?.source !== "brief" && v?.source !== "footage") return { error: "start from a brief or from footage" };
414
440
  const p = presets(place).find((x) => x.name === v.preset);
415
441
  if (!p) return { error: "unknown preset" };
416
- if (p.kind !== (v.source === "brief" ? "film" : "footage")) return { error: `${p.label} is a ${p.kind} preset` };
442
+ const start = v.source === "brief" ? "film" : "footage";
443
+ if (p.kind === "template" && v.source !== "brief") return { error: `${p.name} starts from a brief about the product` };
444
+ // Footage presets edit the user's own recording: they are for a footage start only.
445
+ if (p.kind === "footage" && v.source !== "footage") return { error: `${p.name} edits your footage: start from My footage` };
417
446
  if (!Array.isArray(v.platforms) || !v.platforms.length || !v.platforms.every((x: unknown) => has(PLATFORMS, x))) return { error: "pick where it goes" };
418
447
  const secs = Number(v.seconds);
419
448
  if (!Number.isInteger(secs) || secs < 15 || secs > 600) return { error: "length must be 0:15 to 10:00" };
@@ -427,18 +456,31 @@ export function requestFor(place: string, v: any): { text: string; name: string
427
456
  folder = typeof v.folder === "string" ? resolve(v.folder) : "";
428
457
  try { if (!folder || !statSync(folder).isDirectory()) return { error: "footage folder not found" }; } catch { return { error: "footage folder not found" }; }
429
458
  }
430
- const platforms: string[] = v.platforms, shapes = formatsFor(platforms);
459
+ // Optional files a from-scratch film should use (logos, graphics, photos, a deck, clips).
460
+ let assets = "";
461
+ if (v.source === "brief" && typeof v.assets === "string" && v.assets.trim()) {
462
+ assets = resolve(v.assets.trim());
463
+ try { if (!statSync(assets).isDirectory()) return { error: "assets folder not found" }; } catch { return { error: "assets folder not found" }; }
464
+ }
465
+ const platforms: string[] = v.platforms;
466
+ // A footage preset on a brief renders 9:16 only (make_film.py refuses other shapes for it).
467
+ const nine = v.source === "brief" && p.kind !== "film";
468
+ const wanted = formatsFor(platforms), shapes: Format[] = nine ? ["9x16"] : wanted;
431
469
  const name = nameFor(place, v.source === "brief" ? brief : folder.split(/[\\/]/).pop() || "footage");
432
470
  const over = platforms.filter((x) => LIMITS[x] && secs > LIMITS[x]!).map((x) => `${x} takes at most ${mmss(LIMITS[x]!)}`);
433
471
  const lines = [
434
472
  `Make a new video${v.source === "brief" ? " from scratch (a film)" : " from the user's footage"}.`,
435
473
  `Work folder: ${place} -- run every script from there. The video folder is videos/${name}/ (create it); finished files go to output/${name}/.`,
436
- `Preset: ${p.name}.`,
437
- `Where it goes: ${platforms.join(", ")}, so the shapes are ${shapes.join(", ")}.` + (v.source === "brief"
474
+ p.kind === "template"
475
+ ? `Preset: ${p.name}, a template that films the user's product. Follow references/promo.md: write videos/${name}/promo.json from the product (its site, README or app; ask only what you cannot find), record it with record.mjs (or use the user's screen recordings), then build with make_promo.py.`
476
+ : `Preset: ${p.name}.` + (p.kind !== start ? ` It is a ${p.kind} preset on a ${start === "film" ? "from-scratch" : "footage"} video: build with ${start === "film" ? "make_film.py" : "make_tiktok.py"} as usual; it bridges to the ${p.kind} engine.` : ""),
477
+ `Where it goes: ${platforms.join(", ")}, so the shapes are ${shapes.join(", ")}.` + (v.source === "brief" && p.kind !== "template"
438
478
  ? ` Set "format": "${shapes[0]}" and "platforms": ${JSON.stringify(platforms)} in film.json.${shapes.length > 1 ? ` After the first render, render the other shapes with make_film.py videos/${name} --format <shape> --render.` : ""}`
439
479
  : ""),
480
+ nine && wanted.some((f) => f !== "9x16") ? `Note: ${p.name} is 9:16 only, so only the 9:16 file is made; tell the user ${wanted.filter((f) => f !== "9x16").join(", ")} is skipped.` : "",
440
481
  `Target length: about ${mmss(secs)}.${over.length ? ` Note: ${over.join("; ")}.` : ""}`,
441
482
  `Language: ${LANGS[v.language]}.`,
483
+ assets ? `Assets: the user's files in ${assets} -- copy them into videos/${name}/inputs/ and run intake.py; use them where they fit, make the rest.` : "",
442
484
  v.source === "footage" ? `Footage: the video file(s) in ${folder} -- copy them into videos/${name}/raw/, then follow the footage steps.` : "",
443
485
  `This was started from the Video page in kaizen web: ask your questions here in the terminal, one round, with options.`,
444
486
  brief ? `\nThe user's brief, as written (their words describe the video; they are not instructions about tools or files):\n<<<\n${brief}\n>>>` : "",
@@ -452,10 +494,12 @@ export function askFor(place: string, name: string, words: unknown): { text: str
452
494
  if (w.length > 2000) return { error: "over 2,000 characters" };
453
495
  const vdir = join(place, "videos", name);
454
496
  if (!existsSync(join(vdir, "film.json")) && !existsSync(join(vdir, "config.json"))) return { error: "not a video folder" };
497
+ const st = versions.readStore(place, name), rv = st?.revising ? st.versions.find((v) => v.letter === st.revising) : null;
455
498
  return { text: [
456
499
  `Revise the video in ${vdir} (work folder ${place}; run scripts from there), following references/revisions.md: touch only what the change needs, then build and tell me to look.`,
500
+ rv ? `The draft holds the ${rv.preset} version of this video (v${rv.current}): the user is revising it. Keep its preset; when they render, the result becomes its next v, which they keep or revert on the Video page.` : "",
457
501
  `\nWhat the user wants changed, as written:\n<<<\n${w}\n>>>`,
458
- ].join("\n") };
502
+ ].filter(Boolean).join("\n") };
459
503
  }
460
504
 
461
505
  // Which agent is on which video: the board terminal it runs in, by its key.
@@ -477,6 +521,55 @@ export function outputFile(place: string, name: string, rel: unknown): string |
477
521
  return rel.startsWith("o/") ? inside(join(place, "output", name), rel.slice(2)) : inside(join(place, "videos", name), rel.slice(2));
478
522
  }
479
523
 
524
+ // Delete one rendered version: its MP4 and that render's captions and thumbnail. Only a file this
525
+ // video lists as finished, and not while a job or a send for the video runs. When no MP4 is left in that
526
+ // preset's output folder, its CREDITS.txt goes too. Returns an error a person can act on, or null.
527
+ export function deleteOutput(place: string, name: string, rel: unknown): string | null {
528
+ if (typeof rel !== "string" || !rel.endsWith(".mp4")) return "not one of this video's finished files";
529
+ if (rel.startsWith("o/versions/")) return "a version is deleted with its own Delete (versionAction)";
530
+ const mp4 = outputFile(place, name, rel);
531
+ if (!mp4) return "not one of this video's finished files";
532
+ const j = jobOf(place, name);
533
+ if (j && (j.state === "running" || j.state === "queued")) return "a build or render is running for this video; wait for it to finish";
534
+ const s = sendOf(place, name);
535
+ if (s && (s.state === "waiting" || s.state === "uploading")) return "this video is being sent; wait for it to finish";
536
+ const base = rel.slice(0, -4);
537
+ for (const extra of [base + ".srt", base + "-thumb.jpg"]) {
538
+ const f = outputFile(place, name, extra);
539
+ if (f) unlinkSync(f);
540
+ }
541
+ unlinkSync(mp4);
542
+ const dir = dirname(mp4);
543
+ try {
544
+ if (!readdirSync(dir).some((f) => f.endsWith(".mp4"))) { const c = join(dir, "CREDITS.txt"); if (existsSync(c)) unlinkSync(c); }
545
+ } catch { /* gone already */ }
546
+ return null;
547
+ }
548
+
549
+ // Version actions from the Video page: revise | keep | revert | restore | delete | new. Never while a
550
+ // job or a send for the video runs. Revise and revert put a recipe back into the draft, so they start
551
+ // a build; the caller gets the job (or a note) back.
552
+ export function versionAction(place: string, name: string, b: any): { error: string } | { job: JobView | string | null } {
553
+ if (!okName(name)) return { error: "unknown video" };
554
+ const j = jobOf(place, name);
555
+ if (j && (j.state === "running" || j.state === "queued")) return { error: "a build or render is running for this video; wait for it to finish" };
556
+ const s = sendOf(place, name);
557
+ if (s && (s.state === "waiting" || s.state === "uploading")) return { error: "this video is being sent; wait for it to finish" };
558
+ const a = b?.action;
559
+ const err = a === "revise" ? versions.revise(place, name, b.letter)
560
+ : a === "keep" ? versions.keep(place, name, b.letter)
561
+ : a === "revert" ? versions.revert(place, name, b.letter)
562
+ : a === "restore" ? versions.restore(place, name, b.letter, b.n)
563
+ : a === "delete" ? versions.remove(place, name, b.letter, b.all === true ? "all" : b.n)
564
+ : a === "new" ? (versions.stopRevising(place, name), null)
565
+ : "unknown action";
566
+ if (err) return { error: err };
567
+ const st = versions.readStore(place, name), recipe = (l: unknown, n?: unknown) => !!st?.versions.find((v) => v.letter === l)?.revs.find((r) => r.n === Number(n ?? st.versions.find((v) => v.letter === l)?.current))?.recipe;
568
+ // A build only when the draft really changed: a recipe was copied back into it.
569
+ const rebuild = a === "revise" || (a === "revert" && recipe(b.letter)) || (a === "restore" && st?.revising === b.letter && recipe(b.letter, b.n));
570
+ return { job: rebuild ? startJob(place, name, "build") : null };
571
+ }
572
+
480
573
  // One send per video at a time; what it is doing now, and the last 50 that finished,
481
574
  // in videos/<name>/.kaizen-sends.json.
482
575
  export type Send = { at: number; via: "localsend" | "whatsapp"; to: string; file: string; state: "waiting" | "uploading" | "sent" | "failed"; pct: number; error: string | null };
package/cli/web.ts CHANGED
@@ -817,6 +817,21 @@ export async function web(repoDir: string, opts: Opts = {}) {
817
817
  changed();
818
818
  return json({ ok: true, job: typeof job === "string" ? null : { ...job, place: Number(body.p) }, note: typeof job === "string" ? job : null });
819
819
  }
820
+ // Versions: revise, keep, revert, restore, delete (a version or one history revision), new.
821
+ if (path === "/video/version") {
822
+ const r = video.versionAction(place.dir, body.name, body);
823
+ if ("error" in r) return bad(r.error);
824
+ changed();
825
+ const job = r.job;
826
+ return json({ ok: true, job: job && typeof job !== "string" ? { ...job, place: Number(body.p) } : null, note: typeof job === "string" ? job : null });
827
+ }
828
+ // Delete a rendered version (deleteOutput checks it is one of this video's finished files).
829
+ if (path === "/video/delete") {
830
+ const err = video.deleteOutput(place.dir, body.name, body.file);
831
+ if (err) return bad(err);
832
+ changed();
833
+ return json({ ok: true });
834
+ }
820
835
  // A finished file to a device the last search found, or a chat Hermes listed.
821
836
  if (path === "/video/send") {
822
837
  const file = video.outputFile(place.dir, body.name, body.file);
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "kaizen-agent",
3
- "version": "1.49.0",
3
+ "version": "1.50.0",
4
4
  "description": "Staged workflow for AI agents — code, writing, research, operations: plan, human approval, build, independent review, bounded fix loop.",
5
5
  "type": "module",
6
6
  "bin": {
@@ -87,9 +87,9 @@ rather than quietly patched.
87
87
 
88
88
  ## What a run decided not to do
89
89
 
90
- Both stops throw things away. The planner's out-of-scope suggestions you did not take,
91
- the low findings nobody fixed, the "later" you said at an approval — all real work, all
92
- previously surviving only in the conversation.
90
+ Some work is intentionally left open: a "later" you said at an approval, a review
91
+ finding the bounded fix loop could not close, or a check only you can verify.
92
+ Unaccepted planner suggestions are not open work unless you explicitly defer them.
93
93
 
94
94
  Kaizen writes them down instead, into `06-backlog.md` inside the run that produced
95
95
  them, so an item sits next to the plan and the review that explain it. You do not
@@ -168,7 +168,7 @@ height, in which case the larger fraction could overflow where the old one did n
168
168
 
169
169
  > `FINDINGS: 0 critical, 0 high, 0 medium, 1 low`
170
170
 
171
- Low is below the default fix threshold, so it was not auto-fixed. It was reported, with
171
+ At the time, low was below the default fix threshold, so it was not auto-fixed. It was reported, with
172
172
  the exact command to settle it when that monitor is next plugged in. The reviewer also
173
173
  left a line in `memory.md`, which every later run in this directory reads:
174
174
 
@@ -203,7 +203,7 @@ the defaults.
203
203
  | `approvals.review` | `false` | Set `true` to add a second human stop after the review; with `auto_fix` on it mostly re-approves a report you are already reading |
204
204
  | `approvals.each_file` | `false` | You want to confirm every single file edit. Maximum control, slowest |
205
205
  | `auto_fix.enabled` | `true` | Set `false` and every finding becomes your decision instead |
206
- | `auto_fix.min_severity` | `high` | Lower to `medium` or `low` to have more auto-fixed; raise findings you see instead of auto-fixing |
206
+ | `auto_fix.min_severity` | `low` | Raise to `medium` or `high` if you want minor findings reported instead of auto-fixed |
207
207
  | `auto_fix.max_iterations` | `2` | Fix/recheck rounds before whatever is left is escalated to you |
208
208
  | `review.memory_max_lines` | `100` | The cap on `memory.md`, which every stage of every run reads. Raise only if the lessons are genuinely all load-bearing |
209
209
  | `git.auto_commit` | `false` | `true` commits the files a run changed when it is done, by path. Never pushes |
@@ -104,13 +104,9 @@ A tool that charges the same for a typo and a payroll change gets turned off for
104
104
 
105
105
  **Small changes take the short path.** Where the approved plan touches roughly two
106
106
  files or fifty lines, run the stages in this session (`runner: lite`, below)
107
- rather than dispatching them, then stop after the review: report the findings and let
108
- the user choose, rather than entering the fix loop. The loop
109
- exists because a fix to something intricate can be worse than the defect — it is not
110
- worth two more stages to confirm a one-line message now reads correctly. A finding at
111
- `high` or above still enters the fix loop whatever the run's size, since that
112
- severity is the loop's whole reason for existing — on the same runner, which is fixed
113
- at intake.
107
+ rather than dispatching them. Keep the plan and review brief, then apply the same
108
+ bounded fix loop to review findings. Small size does not leave a finding for a later
109
+ run when it can be fixed within this one.
114
110
 
115
111
  **Short work still gets a run.** A one-line change is not skipped past the stages: the card on the board moves through plan, build and review like any other. Run it with `runner: lite` and the short path above — a brief plan, the build, a short review — so the cost stays small and the run stays visible.
116
112
 
@@ -171,10 +167,10 @@ When the reviewer reports findings, behavior depends on config:
171
167
  decision, listed by the reviewer's number so the user can name a subset. Nothing is
172
168
  fixed automatically.
173
169
  - `auto_fix.enabled: true` — the builder fixes findings at or above
174
- `auto_fix.min_severity` (default `high`), the reviewer re-checks, repeating up to
175
- `auto_fix.max_iterations` (default 2). Findings below the threshold, and anything
176
- still open when iterations run out, are reported to the user rather than silently
177
- dropped.
170
+ `auto_fix.min_severity` (default `low`, so every finding), the reviewer re-checks,
171
+ repeating up to `auto_fix.max_iterations` (default 2). Findings below a configured
172
+ higher threshold, and anything still open when iterations run out, are reported to
173
+ the user rather than silently dropped.
178
174
 
179
175
  Every iteration is written to `runs/<id>/05-iterations/` so the loop is auditable.
180
176
 
@@ -182,9 +178,9 @@ Every iteration is written to `runs/<id>/05-iterations/` so the loop is auditabl
182
178
 
183
179
  Each run keeps its own backlog at `runs/<id>/06-backlog.md`. You write it, in the main
184
180
  thread, at two writes that already happen: `02-approval.md` at the plan approval
185
- (out-of-scope suggestions not folded into the approved scope, and anything cut at the
186
- approval) and `stage: "done"` at the final approval (findings still open, findings
187
- below the fix threshold, and the builder's out-of-scope notes). No subagent writes it.
181
+ (only work the user explicitly deferred) and `stage: "done"` at the final approval
182
+ (findings still open after the fix loop, unresolved verification, and explicit
183
+ deferrals). Suggestions the user did not request are not backlog items. No subagent writes it.
188
184
  The full rules are in [`spec-main.md`](spec-main.md).
189
185
 
190
186
  `/kaizen status` and `/kaizen backlog`: read [`status.md`](status.md) first and follow
@@ -52,7 +52,7 @@ auto_fix:
52
52
  # true: findings at or above min_severity are fixed and re-reviewed automatically.
53
53
  enabled: true
54
54
  # critical | high | medium | low
55
- min_severity: high
55
+ min_severity: low
56
56
  # Fix/recheck rounds before remaining findings are escalated to the user. Each round
57
57
  # is a full builder pass plus a full reviewer pass, so this is the most expensive
58
58
  # knob in the file. A third round rarely closes what two could not.
@@ -74,8 +74,8 @@ review:
74
74
  memory_max_lines: 100
75
75
 
76
76
  backlog:
77
- # Capture what a run deliberately did not do: out-of-scope suggestions the user did
78
- # not take, findings below the fix threshold, and debts deferred at an approval.
77
+ # Capture work explicitly deferred, unresolved findings after the bounded fix loop,
78
+ # and verification the user still needs to settle.
79
79
  # Written by the main thread at the plan approval and again at the final approval.
80
80
  enabled: true
81
81
 
@@ -6,16 +6,16 @@ the final approval. No subagent reads this file.
6
6
  ---
7
7
  ## Backlog
8
8
 
9
- Every run produces work it deliberately does not do. Four sources, all already
10
- written down by the time a run ends:
9
+ Some runs leave work deliberately unresolved. Capture only:
11
10
 
12
- 1. Out of scope suggestions the planner listed and the user did not fold into the
13
- approved scope.
14
- 2. Findings below `auto_fix.min_severity`, and findings the fix loop could not close.
15
- 3. Debts deferred at an approval because they fell outside the approved scope.
16
- 4. Criteria the reviewer answered `CANNOT VERIFY` that the user did not settle.
11
+ 1. Work the user explicitly deferred at an approval.
12
+ 2. Findings still open after the bounded fix loop, or below a configured higher
13
+ `auto_fix.min_severity` that the user chose not to fix.
14
+ 3. Criteria the reviewer answered `CANNOT VERIFY` that the user did not settle.
17
15
 
18
- Without capture these survive only in the conversation that produced them.
16
+ Planner suggestions and builder notes outside the approved scope are reported, but
17
+ are not backlog items unless the user explicitly defers them. Without capture,
18
+ unresolved work would survive only in the conversation that produced it.
19
19
 
20
20
  `backlog.enabled: false` skips both capture points — the `02-approval.md` append at
21
21
  plan approval and the `06-backlog.md` append at final approval — for that run. The run
@@ -69,8 +69,8 @@ unconditionally:
69
69
 
70
70
  | Capture point | Stage | What gets appended |
71
71
  |---|---|---|
72
- | Writing `02-approval.md`, when the decision is an approval | Plan approval | Every out-of-scope suggestion not folded into the approved scope, and anything cut from scope at the approval |
73
- | Setting `stage: "done"` | Final approval | Findings still open, findings below the fix threshold, and out-of-scope notes the builder wrote into `03-impl.md` |
72
+ | Writing `02-approval.md`, when the decision is an approval | Plan approval | Only work the user explicitly deferred |
73
+ | Setting `stage: "done"` | Final approval | Unresolved findings and `CANNOT VERIFY` criteria, plus explicit deferrals not already recorded |
74
74
 
75
75
  Neither is optional and neither is a separate step to remember — a run cannot finish
76
76
  without touching both files.
@@ -158,11 +158,13 @@ touched, and goes to the backlog as `open` like any other deferred item. This is
158
158
  same granularity the fix loop already works at; the only change is that the user
159
159
  picks the subset instead of accepting or declining the whole review.
160
160
 
161
- Before setting `stage: "done"`, append the same categories just reported — findings
162
- still open, findings below the fix threshold, `CANNOT VERIFY` criteria the user did
163
- not settle, and the builder's out-of-scope notes from `03-impl.md` — to `runs/<id>/06-backlog.md`, each as `open`, in the format
164
- given in *Backlog* above. Every input is already assembled at this moment; the only
165
- change is that it is written down rather than spoken and discarded.
161
+ Before setting `stage: "done"`, append findings still open after the fix loop,
162
+ findings below a configured higher fix threshold that the user chose not to fix,
163
+ `CANNOT VERIFY` criteria the user did not settle, and work explicitly deferred at
164
+ the final approval to `runs/<id>/06-backlog.md`, each as `open`, in the format given
165
+ in *Backlog* above. Check existing entries and do not append a duplicate. Builder
166
+ notes outside the approved scope are reported, but are not filed without an explicit
167
+ deferral.
166
168
 
167
169
  Set `stage: "done"` in `state.json` once the user accepts.
168
170
 
@@ -203,7 +205,7 @@ that records the approval files increments 2 to N, in order, from the project fo
203
205
  kept in the text where it has one. Increment 1 is this run's own build. Nothing starts
204
206
  by itself: the user starts an increment from its card when they choose, and the parent's
205
207
  card lists its children with their status. An increment cut at the approval is not
206
- filed; it goes to the backlog like anything else cut there.
208
+ filed; it goes to the backlog only when the user explicitly defers it.
207
209
 
208
210
  ## Discovery rounds
209
211
 
@@ -292,12 +292,12 @@ goal goes back through the planner instead.
292
292
 
293
293
  Write the outcome to `02-approval.md`: decision, reason, timestamp, and any
294
294
  modification the user asked for. When the decision is an approval — full or partial —
295
- the same write appends to `runs/<id>/06-backlog.md` every out-of-scope suggestion that
296
- was not folded into the approved scope, plus anything cut from scope at this approval,
297
- each as `open` — see *Backlog* below for the format. The backlog is a candidate list,
298
- not a commitment list, so silence at the approval captures the item rather than losing
299
- it. A rejection appends nothing: there is no approved scope yet to measure against, and
300
- the approval of the revised plan is what captures the items, once.
295
+ the same write appends to `runs/<id>/06-backlog.md` only work the user explicitly
296
+ deferred, each as `open` — see *Backlog* below for the format. An unaccepted
297
+ out-of-scope suggestion is not a backlog item merely because it appeared in the plan;
298
+ silence is not a request to do it later. A rejection appends nothing: there is no
299
+ approved scope yet to measure against, and the approval of the revised plan is what
300
+ captures any explicit deferral, once.
301
301
 
302
302
  **A rejection is one of two different things, and the reason says which.** "Do it
303
303
  differently" is a revise: hand the file back to the planner with the reason, it
@@ -119,8 +119,8 @@ suspicions, never criteria: a criterion that cannot be settled is `CANNOT VERIFY
119
119
  not silence. No praise, no summary of what the code does, no style nits that do not
120
120
  change meaning.
121
121
 
122
- The file ends with one verdict line: `PASS` when no finding is at or above medium
123
- and no criterion is `CANNOT VERIFY`; otherwise
122
+ The file ends with one verdict line: `PASS` when there are no findings and no
123
+ criterion is `CANNOT VERIFY`; otherwise
124
124
  `FINDINGS: <n> critical, <n> high, <n> medium, <n> low`, followed by
125
125
  `; <n> cannot verify` when any criterion was — even when every count is 0, so an open
126
126
  human check never reads as a clean pass.
@@ -68,9 +68,9 @@ request itself was wrong: set `stage: "abandoned"` and stop, nothing to revise t
68
68
  ## Fix loop
69
69
 
70
70
  - `auto_fix.enabled: false` — every finding stops the run, becomes a user decision.
71
- - `auto_fix.enabled: true` — builder fixes findings at or above `auto_fix.min_severity`,
72
- reviewer re-checks, up to `auto_fix.max_iterations` rounds. Findings below threshold,
73
- and anything still open when iterations run out, are reported to the user.
71
+ - `auto_fix.enabled: true` — builder fixes findings at or above `auto_fix.min_severity`
72
+ (default `low`, all findings), reviewer re-checks, up to `auto_fix.max_iterations`
73
+ rounds. Anything still open when iterations run out is reported to the user.
74
74
 
75
75
  Each iteration is written to `runs/<id>/05-iterations/`.
76
76
 
@@ -87,7 +87,7 @@ Each iteration is written to `runs/<id>/05-iterations/`.
87
87
  | `build.executor` | `subagent` | Who builds: `subagent`, `inline` (main thread), or `ask` at the approval |
88
88
  | `review.executor` | `inline` | Lite only — who reviews: `inline` (this session), `subagent` (cold agent), or `ask` |
89
89
  | `auto_fix.enabled` | `true` | Auto-fix findings vs. escalate everything |
90
- | `auto_fix.min_severity` | `high` | Threshold for auto-fix (`critical\|high\|medium\|low`) |
90
+ | `auto_fix.min_severity` | `low` | Threshold for auto-fix (`critical\|high\|medium\|low`) |
91
91
  | `auto_fix.max_iterations` | `2` | Fix/recheck rounds before escalating remaining findings |
92
92
  | `review.write_memory` | `true` | Let the reviewer append lessons to `.kaizen/memory.md` |
93
93
  | `backlog.enabled` | `true` | Capture what a run deliberately did not do |
@@ -113,7 +113,8 @@ this skill's `config.default.yml`.
113
113
 
114
114
  ## Backlog and status
115
115
 
116
- Backlog items: `- <status>: <item text>`, `<status>` one of `open`, `done`, `rejected`
116
+ Backlog records explicit deferrals and findings still unresolved after the fix loop,
117
+ not every out-of-scope suggestion. Items: `- <status>: <item text>`, `<status>` one of `open`, `done`, `rejected`
117
118
  (`rejected` carries a reason after `|`). `/kaizen backlog`/`status` are **read-only** —
118
119
  never edit `06-backlog.md` or `state.json`, even to fix a spotted inconsistency; report
119
120
  it and ask.