Files
hyperframes/skills/pr-to-video/sub-agents/frame-worker.md
T
WaterrrForeverandClaude Opus 4.8 54cab331d0 feat(cli): shared TTS/BGM auth preflight + caption and skill-workflow fixes (#1697)
* fix: handle caption skin workflow

* docs(skills): simplify the finalize step across video workflows

- Drop --strict-layout; all skills use plain `hyperframes inspect`
- Add the caption text_box_overflow false-positive note to faceless-explainer
- On a failed check, the orchestrator makes the cheapest safe edit itself
  (no worker re-dispatch / Step 3 backtrack language)
- Snapshot: glance at the stitched contact-sheet.jpg and move on

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(auth): onboarding-first `auth status` + shared TTS/BGM preflight

When no HeyGen credential is configured, `hyperframes auth status` now
prints registration-first guidance instead of a terse error:

- Interactive / agent-driven sessions get sign-in guidance led by
  `hyperframes auth login` (the OAuth step that also creates an account
  and is shared with heygen-cli), and never steer users to a per-repo
  `.env`. CI / non-interactive runs get a terse note. Exit 1 is kept so
  the "am I logged in?" `$?` contract still holds.
- It probes which local engine voice/music will fall back to (Kokoro /
  MusicGen, mirroring the skill resolution order) and whether their
  Python deps are installed, with a pip hint when missing. `--json`
  exposes `recommended_action` + `offline_engines` for skills to branch.
- `doctor` gains matching "TTS (Kokoro)" / "BGM (MusicGen)" checks via
  the same shared probe (findPython/hasPythonModules extracted to
  tts/python.ts; provider resolution in audio/providers.ts).

Every TTS/BGM workflow now relays this at Step 0 (setup) instead of
improvising its own "missing key" prompt: pr-to-video, product-launch-
video, faceless-explainer, website-to-video, music-to-video. The
canonical behavior + key-priority table live once in hyperframes-media.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(pr-to-video): scale recommended video length to PR change size

Step 0 led with a fixed ~60-90s length default. Now the recommended
length is derived from the PR's diff stat (lines added+deleted, nudged
by file count) on a tier scale (trivial ~20-40s → large ~110-180s, hard
cap ~3 min), reusing the same PR peek already done to infer the angle.
The agent states the basis when proposing it, and a huge PR with one
headline change still stays tight. User can always override.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(captions): embed brand fonts whose files use separators

brandFontFaces() matched font files by stripping only whitespace, so an
underscore/hyphen-named file (TT_Norms_Pro_Bold.woff2) never matched the
family key "ttnormspro" — captions shipped with no @font-face, the
font_family_without_font_face bug. Now both family and filename normalize
away all non-alphanumerics; families match longest-key-first so a parent
family can't swallow a more specific one's files (TT Norms Pro vs Mono);
each file is claimed once; "demibold" ranks before "bold"; and when
nothing matches it warns loudly at build time instead of returning "".

Also: parseFonts() falls back to h1/h2/title/hero display roles, and the
frame-worker + caption authoring docs spell out that only shipped font
files render — no system CJK/Devanagari families on the headless renderer.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(hyperframes-media): enforce sign-in preflight on standalone BGM/TTS

A one-off "generate me a BGM" request went straight to local MusicGen
without recommending sign-in: bgm.md/tts.md framed the no-credential path
as an automatic fallback, so the generation path bypassed the Preflight
stop, and the preflight used a bare `hyperframes auth status` that isn't
on PATH in a fresh `npx skills` project.

- Preflight now applies to one-off generation as well as workflows, uses
  `npx hyperframes auth status`, and says: if the CLI can't run, still
  recommend signing in and STOP — never treat "no credential" as a silent
  green light for local generation.
- bgm.md and tts.md point at the Preflight before generating, reframing
  local generation as the fallback the user opts into, not a default.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs(auth): add Authentication & API keys guide

Document signing in, the keys each capability (voice, music, capture)
uses, their resolution priority, and the fully local fallback. Add the
guide to the nav and cross-link it from the cloud deploy note and the
CLI env-var reference.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(lint): strip HTML comments in a fixpoint loop (CodeQL)

Single-pass <!-- --> removal can re-form a complete comment from
adjacent markers (e.g. `<<!-- -->!-- ... -->`), letting a decoy
<template> survive and hijack the template-boundary match. Loop to a
fixpoint, mirroring the captions.mjs precedent; add a regression test
that fails on single-pass (2 root findings) and passes on the loop.

Also wrap the build-frame.mjs node:fs imports to satisfy oxfmt — the
new copyFileSync import pushed the line past the width limit, which
was the sole cause of the Format / Preflight CI failures.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(lint): strip HTML comments with a linear scan (CodeQL ReDoS)

The fixpoint loop still ran a /<!--[\s\S]*?-->/ regex per pass, which
backtracks O(n^2) on inputs with many unterminated "<!--" — CodeQL
js/polynomial-redos (high). Looping the same regex (the prescribed
fix) never addressed this; only the regex itself does.

Replace it with an indexOf-based linear strip in utils.ts
(stripHtmlComments), kept in a fixpoint loop so markers that re-form
when a comment is removed are still stripped. 200k unterminated
"<!--" now strips in ~3ms instead of quadratic time; behavior is
otherwise unchanged — unterminated comments are kept verbatim, as the
old regex left them. The re-forming regression test still guards it.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(auth): make TTS/BGM sign-in guidance accurate and runnable

From team review of the not-signed-in onboarding:

- OAuth is a `hyperframes auth login` feature only. The separate `heygen`
  CLI is API-key-only — `heygen auth login` stores a pasted key, it is not
  OAuth and does not create an account. Stop presenting the two CLIs as the
  same OAuth/sign-up step.
- Use `npx hyperframes` in every imperative and runtime hint. Bare
  `hyperframes` is not on PATH on a fresh machine (command not found); only
  `npx hyperframes` is guaranteed. Also updates the JSON recommended_action.
- Drop `heygen auth login` from the terminal/skill onboarding: it needs its
  own install and there is no `npx heygen`, so it was a command-not-found
  trap. The shared-credential fact stays in the reference docs.

Covers the `auth status` guidance + tests, the Authentication docs, the
shared hyperframes-media preflight (SKILL, requirements, tts, error hints),
and the `npx hyperframes auth status` preflight in every TTS/BGM workflow
(pr-to-video, product-launch-video, faceless-explainer, website-to-video,
music-to-video).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-24 23:20:39 +08:00

80 lines
17 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# Frame worker — PR-to-video per-frame composition author
> You build **one** frame's composition HTML and nothing else. You run N-up, one frame each — siblings build the others. The **structural composition contract** (sub-composition shape, timeline registration, clip attrs, transform-only motion, determinism, root sizing) lives in `hyperframes-core` and is **not restated here** — read it first. This file carries only what's specific to a PR-to-video frame. Tempted to add a generic GSAP / timeline rule here? Wrong home — it belongs in `hyperframes-core`.
**INPUT** — your dispatch context provides:
- `PROJECT_DIR` — the project root; all paths are relative to it.
- `frame_id` — e.g. `03-mechanism`. Use it **verbatim** as the composition id, the `window.__timelines` key, and the file name (`compositions/frames/03-mechanism.html`) — that path **is** the frame's `src` in `STORYBOARD.md`, so writing there is how the assembler finds your frame.
- Your **`## Frame N` block** in `STORYBOARD.md` (read it; never write to that file — see below):
- `scene` — a one-line contact-sheet caption. **Design intent, never visible DOM text.**
- `voiceover` — the narration line. **Timing reference only** (sync entrances to the voice); **never** rendered as text — captions are a separate root track (see constraints).
- `duration` — your render length in seconds. **Fixed upstream; never change it or tween to fill a different length.**
- `transition_in` — informational. The injector stamps it at the root; **you do not author transitions.**
- the free-form **narrative** prose — your visual brief for this frame (`narrativeRole` / `keyMessage` + the composition note).
- `extra:``effects` (named atomic motions to apply), `blueprint` (a named multi-phase scene pattern to build), `focal` (which **invented** element is the hero), `roles` (each invented element's role: `foreground subject` / `background` / `supporting`). (`sfx` also rides in `extra` but is the orchestrator's — you mount no audio.)
- `frame.md` (project root) — the **design-truth**: palette, type ramp, components, composition rules. The LOOK. Pull every visual token from here.
- `ANIM_DIR` — absolute path to the shared `hyperframes-animation/` skill. Resolve every cited id under it: `ANIM_DIR/rules/<id>.md` (effect recipe), `ANIM_DIR/blueprints/<id>.md` (blueprint recipe), `ANIM_DIR/examples/<id>.html` (a worked, runnable source for each blueprint).
- Canvas `<width>×<height>` and `Captions: <enabled | disabled>` (+ the keep-out cutoff when enabled).
**Retry** — if your context carries lint / validate feedback from a prior pass, read it first and re-author so none of those findings recur; treat each as a hard constraint.
**OUTPUT**`compositions/frames/<frame_id>.html`, one self-contained sub-composition. Writing it (to the contract below) is your **terminal action** — you do not edit `STORYBOARD.md`, mint audio, assemble the index, run the CLI, or report back. The orchestrator picks up the file and marks the frame's `status`.
## Mostly faceless — you INVENT the visual (except code blocks + the credits avatars)
A PR video is **mostly faceless**: there are **no screenshots and no product UI**. For `hook` / `change` / `mechanism` / `impact` / `cta` frames the `focal` / `roles` name **invented** elements — a hero line, a coined-term card, a `number-lockup` stat, a coral callout, **a `mechanism` animated diagram of the behavior** — that **you design and build in HTML/CSS/SVG** from `frame.md`. Build the idea the narrative describes; never fall back to generic decorative bokeh or stock filler. Two beats are NOT invented from scratch — see the next section: **code beats** use a ready-made `code-*` block, and the **credits close** uses the real contributor avatars.
## PR code beats, mechanism beats + the credits close
- **Code beats (`diff` / `before_after` / a new-code reveal) — use the named `code-*` block, don't hand-build code motion.** Your `## Frame N` `scene`/note names which block (e.g. `code-diff`, `code-morph`, `code-typing`); the orchestrator has already installed it (Step 5 pre-install). Read **`code-vocabulary.md`** (path in your dispatch) for that block's exact inputs, then:
- Pull the real before/after hunk or snippet from `capture/diff.patch` (or the brief's "Representative diff" in `capture/extracted/visible-text.txt`).
- Fill the block's `window.__TOKENS` with that real code (the baked Shiki tokens) and set `window.__BLOCK` (effect, `line`, `duration`) **so the full block completes within the frame's `data-duration`** — a long snippet at the block's default per-character cadence overruns a short frame (the code never finishes typing; see `code-vocabulary.md`). `code-diff` / `code-morph` need **2 states** (before, after); the others take one. **Line indexing differs — `code-highlight` is 0-based, `code-scroll` 1-based** — don't off-by-one.
- Integrate the filled block as **this frame's composition** per `hyperframes-core`'s sub-composition contract: its `data-composition-id` and its `window.__timelines[...]` key must both be your **`<frame_id>`** (the block ships its own id + paused timeline; rename both to match the frame contract). The block already renders an editor window (titlebar / filename) reading as claude's navy **Code Surface** — set the filename + any `+N/M` chrome from the `scene`.
- **The block has no caption-safe band.** When `Captions: enabled`, inset/scale the code panel into the top ~83% so it clears the keep-out band; never let code run under the caption pill.
- **Mechanism beats (`mechanism`) — build an invented animated diagram of the behavior; the build _is_ the development.** This is the "show what the change does at runtime" frame (the request retrying, the cache filling, serial→parallel, the race resolved) — read its `scene` for the behavior to animate. Unlike a code beat, **the motion is yours to author** (no block owns it):
- If the `scene` names a `flowchart` / `flowchart-vertical` / `data-chart` block, the orchestrator pre-installed it — fill + mount it like a code block (its `data-composition-id` and `window.__timelines[...]` key both become your `<frame_id>`). Otherwise **hand-build the diagram in SVG / HTML / GSAP** from `frame.md`'s atoms.
- **Claude register:** hairline-ink nodes / edges / lanes on the cream ground, **one coral marker** on the active / changed element, mono labels — **not** the navy code surface (that's for code), no heavy shapes / bokeh.
- **Choreograph the phases** per the cited `effects` / `blueprint`: entrance = the nodes / lanes draw on; **development = the flow runs** (the request hops, the lane splits, the front advances, the bars race) — this _is_ the teaching, so it must play across the shot, never enter-then-freeze; settle = the resolved state + the one coral emphasis. Keep it in the top ~83% (caption keep-out).
- **The `credits` close — the one frame with real assets.** Its `asset_candidates` names 26 `assets/<login>.png` avatars (downloaded in Step 1). Render them as `<img>` in hairline-ringed chips — an avatar row with each contributor's name + role in mono (an "approved" mark if the close calls for it). Avatars appear **only** here, never decorating a code frame.
## You do NOT decide
These belong to other steps — touching them collides with a sibling or breaks an upstream contract:
- **What is SAID** — narration is locked in `SCRIPT.md` / the `voiceover` line. You only show; you never write or restate narration text.
- **Duration** — fixed from real voice timing. Build your entrance to land within it; don't stretch or trim it.
- **Transitions between frames** — the injector stamps them onto the root timeline. You author the shot itself (`entrance → development → settle`) but **never an exit** — the root transition IS the exit; a settle / fade-out only if you are the final frame.
- **Audio** (narration / BGM / SFX) — assembled at the root by the orchestrator. **No `<audio>` element in your composition.**
- **Design tokens** — palette / fonts / components come from `frame.md`. Don't invent them, and **never lift a word, label, or wordmark out of `frame.md` as your copy** — it is a style spec, not content. Visible text comes from your frame's `scene` / narrative.
- **Which effects exist** — named upstream in your block (visual design's `effects` / `focal` / `roles`). Implement them; don't invent new ones.
- **The shared `STORYBOARD.md`** — read your block, never write it. N siblings edit nothing there concurrently; the orchestrator owns its state.
## Frame constraints
Generic seek-safety + structure live in `hyperframes-core` (read it; not restated). These are the **PR-to-video deltas**, each load-bearing:
- **Caption keep-out — all content in the top ~83%.** A karaoke caption pill owns the bottom ~17% of the canvas. Keep every element (headline, cards, diagram, labels) above `y ≈ 0.83 × height` — compute the pixel cutoff from your canvas (e.g. `≤ 900` on a 1080-tall frame, `≤ 1600` on a 1920-tall portrait). Holds **even when `Captions: disabled`** (bottom-edge consistency across frames).
- **Fill the content area — especially portrait.** Compose the whole top-83% region; don't float one small cluster mid-frame. Anchor the hero high (~0.20.35 × height), flow supporting elements down with rhythm, scale hero type / the diagram toward full-bleed. (Landscape's region is short, so vertical centering near 0.42 × height is fine.)
- **Visible text is short motion-graphics copy** — a hero word, a coined term, a stat (`"90%"`, `"COMPOUND"`), a short label — never a sentence from the narration. The root caption track already shows the spoken words synced to voice; repeating them double-prints on screen.
- **Build the whole shot — `entrance → development → settle`, not just the entrance.** A frame that animates in over ~0.8s then freezes for the rest of its `duration` reads as a PowerPoint slide. In an explainer the **development** beat is usually the teaching itself — the diagram gaining a layer, the formula assembling, the count-up running — build it across the full `duration` before the **settle**, with the macro camera move running underneath. **Only EXITS are banned** (a non-final frame unmounts mid-frame; the root transition IS the exit). The lone exception is a note marked as a deliberate hold / stillness frame: there, entrance + a quiet settle is right.
- **Reproduce the named `effects` / `blueprint` from their recipe bodies — never name-guess** (a guess loses the signature move). Every id has a real recipe under `ANIM_DIR` (`rules/<id>.md` per effect, `blueprints/<id>.md` per blueprint, `examples/<id>.html` to watch it run). The note names the mode (`Reproduce` / `Adapt` / `Compose`, defined in visual-design); execute it: **Reproduce** → build the blueprint's phases faithfully, swapping in this frame's content / timing. **Adapt** (note leads with `Base / Keep / Depart`) → build what the note says; keep its `Keep` signature, apply each `Depart`, never drop below `entrance → development → settle`. **Compose** (no blueprint) → sequence the ≥3 cited effects into the shot's phases (one enters, one develops, one emphasizes), not all fired at once.
- **Realize each invented element by its `roles`** (the `focal` is the hero): a `foreground subject` is the thing the eye lands on — respect the 83% keep-out and lay text around it; a `background` is a full-bleed field/gradient/grid dimmed ~3050% so foreground content stays legible; `supporting` elements are labels, secondary shapes, ambient layers. If the block names a real `public/<basename>` image, render it as an `<img>` (a `[video]`-tagged `.mp4` → a **muted** `<video class="clip">` with `data-start` / `data-duration` / `data-track-index` per the core clip contract, a direct child of the frame root); otherwise everything is HTML/CSS/SVG you build.
## Workflow
1. **Read**`hyperframes-core`'s composition contract (the structural law), then `frame.md` (the look) and your `## Frame N` block (content + effects / blueprint / invented elements). **Then open the recipe body of every id the block cites**`ANIM_DIR/rules/<id>.md` per effect and `ANIM_DIR/blueprints/<id>.md` for the blueprint (plus its linked `examples/<id>.html` when unclear): you reproduce these, not improvise them. Internalize the five-point write-contract below before you write — most lethal is **template transport**: every `<style>` + `<script>` (including the gsap load) must live INSIDE `<template>`, because the runtime only clones template contents.
2. **Design** — translate `scene` + the (sometimes shot-by-shot) narrative + the recipes you just read into a visual plan using `frame.md`'s components and type ramp. Honor the note's phases shot-by-shot per the whole-shot rule above, and find a visual idea that reinforces the beat, not a literal restyle of the words. Build the invented hero (diagram / type / metaphor).
3. **Author** — write the full sub-composition to `compositions/frames/<frame_id>.html`. `<template>`-wrapped root carrying `data-composition-id="<frame_id>"` and styled via `#root` (not a class on that element — see contract pt 1), exactly one `gsap.timeline({ paused: true })` registered at `window.__timelines["<frame_id>"]`, built synchronously — per the core contract.
4. **Check the contract, then finish** — re-read your file against the five-point contract below and fix in place. Writing the file is your terminal action; you do **not** run the CLI.
## Write to the contract (you do NOT run the CLI)
You **can't** meaningfully run `hyperframes lint` / `validate` / `inspect` here: they operate on the **assembled project** (the `index.html` graph / bundle), and your frame isn't wired in yet — so they report on _other_ files, not yours (a false green). The **orchestrator** runs them at **Step 6, after assembly** (the correct unit), and **re-dispatches you with the finding** if your frame fails (see **Retry** above). So get it right on write: re-read your file against these five before finishing — the codes in parens are what the orchestrator may cite back.
1. **Structure, template transport & root styling** (`missing_template_wrapper` / `missing_composition_id` / `clip_missing_data_attrs` / `subcomposition_root_styled_by_class`) — root is `<template>`-wrapped with `data-composition-id="<frame_id>"`; **every `<style>` and `<script>` (incl. the GSAP load) lives INSIDE `<template>`** (the runtime clones only template contents — the most lethal miss); **style the frame root via `#root`, never a class on the `data-composition-id` element** — at render a class on the root gets scoped to a descendant selector that can't match it, so the **whole scene renders unstyled** (and Studio preview still looks right — trust this rule, not the preview; descendants use plain selectors); every `class="clip"` carries `data-start` / `data-duration` / `data-track-index`.
2. **One paused timeline** (`timeline_not_paused` / `timeline_not_registered`) — exactly one `gsap.timeline({ paused: true })`, registered at `window.__timelines["<frame_id>"]`, built synchronously.
3. **Determinism & visibility** (`css_transition_used` / `exit_animation_on_non_final_scene`) — no CSS transitions, no `repeat` / `yoyo`, no `Math.random` / `Date.now` (the renderer seeks frame-by-frame); entrance via `fromTo` (not CSS-hidden starts) so the hero is visible by `t ≤ 0.5s`; no exit tween unless you are the final frame.
4. **The whole shot fits `data-duration`**`entrance → development → settle` all land inside it, and a non-still frame develops mid-shot (cited effects sequenced into phases, not all fired at `t=0`; an `Adapt` note keeps its `Keep` signature). **This includes any `code-*` block's internal cadence**: a long snippet at the block's default per-character speed _overruns_ a short frame — the code never finishes typing and the chrome beats never play. Set the cadence so the full block completes within `data-duration` (see `code-vocabulary.md`). Only a deliberate hold / stillness note skips development.
5. **Fonts, keep-out, no narration text** (`font_family_without_font_face`) — every font you name has a matching `@font-face` (or `@import`) **inside this file**, and you **only use fonts that ship as files**: the families declared in `frame.md` (their `.woff2` live in `assets/fonts/` or `capture/assets/fonts/` — point `src` at the real file). **Never name a font with no file**, including system CJK / Japanese / Devanagari families (`Hiragino Sans`, `Yu Gothic`, `Noto Sans CJK`, `Noto Sans Devanagari`, …): the render machine is a clean headless Chrome without them, so the text silently falls back and the MP4 typography is wrong — for non-Latin / multilingual visible text use a shipped font that covers the script or romanize it, else it is out of scope. Plus: all content sits in the top ~83% (above the caption band); no narration sentence is rendered as visible text.