Release 0.11.0

[dkackman]

Release 0.11.0: develop -> master.

Develop CI green at 66bfb05 (CI + CodeQL). Smoke and security regressions running separately.

After merge: scripts/release.sh 0.11.0 --next 0.12.0-alpha.1 from the Mac, then gh release edit v0.11.0 --notes-file with the notes below.

---

A short range. MiniMax H3's keyframe templates move to a 4-step turbo
LoRA that halves a long take's render time, music videos hold each shot's
audio slice by default and can give each entry its own singer, cost
estimates reach inline workflows, and torch.compile's kernel cache now
survives a reboot. Under the hood, the #790 refactor brought every module
and function back under the architecture ratchet.

MiniMax H3 (plugins/dw/skills/minimax-h3, workflows/templates)

• The nine T2VA/FL2VA templates default to lightx2v's 4-step v1.2 turbo
  (num_inference_steps 5, shift 6/3) at 960x544 and 1344x768: on a
  345-frame take, ~7.3 min against 14.6 min for the 8-step file it
  replaces. Ref2VA templates stay on 8 steps; there is no tested 4-step
  Ref2VA file.

Music video and cuts (docs/TASKS.md)

• music-video holds each shot's audio slice by default; measured 0.00 s
  sync offset on 8 of 8 shots against up to 1.37 s with the reference
  alone (music-video lip sync: 3-arm A/B (hold_audio + reference) on lem, then pick the default (split from #795) #808). hold_audio is declared (null) on it and the
  match_audio chains (music-video lip sync: test hold_audio + reference, expose hold_audio, correct the #619 conclusion #795).
• for_each items take an optional field with a default, so a
  music-video entry may carry its own singer (music-video: per-entry references falling back to singer_reference #801, validate_workflow: elision reason says 'singer_reference' was supplied when every entry's optional item field overrode it #809).
• plan_cuts divides a split sung scene's lyric across its pieces by
  aligned word times (music-video-cuts: word-timed vocal starts and per-piece lyric splitting (split from #801) #805).
• New measure_sync task cross-correlates a shot's audio against its
  source slice (measure_sync task: cross-correlate a shot's audio against its source slice #802); new relevel-clip utility template, named by the
  clip warning (Re-level a rendered clip without re-rendering: utility template + full-scale warning pointer #794).
• assess_output runs the whole-file level rules (full_scale,
  near_silent) (assess_output raises no full_scale finding on a clip that decodes above 0 dBFS #807); no level_spread warning when a later
  pair_audio replaces the track (Join level-spread warning fires when pair_audio replaces the track downstream #800).

Planning and validation

• An inline workflow is priced by pipeline identity, inheriting the
  catalog basis and warning where offload, quantization or frames differ
  (Inline workflows get cost basis unknown; price by pipeline identity like the VRAM projection #797); the field-shift check covers derived and per-entry estimates
  (Cost estimate: derived basis scales by entry count, not frames, and skips the #593 field-shift check #796).
• Validation refuses an image/last_image argument fed
  previous_result: of an earlier video step (validate passes a for_each entry's image: previous_result:<earlier member> (a video) again; run fails after GPU spent (2nd occurrence of #509) #813).

MCP and server

• list_assets takes limit (newest first), prefix, and reports
  total (list_assets: add limit, prefix filter and total (mirror list_jobs) #798).
• save_prompt accepts a plain string (save_prompt: a plain-string prompt fails as 'not valid JSON' and never names the {"text": ...} shape the #803 skills now ask for #806); wait_for_job falls back
  cleanly on a client timeout and a running job reports finished_steps
  (wait_for_job: skills give no fallback when the client times out before the server cap #792, get_job / wait_for_job: report finished steps' files mid-run (follow-up to #480) #799).

Acceleration (docs/ACCELERATION.md)

• TORCHINDUCTOR_CACHE_DIR defaults to ~/.cache/dw/torchinductor, so
  compiled kernels survive a reboot instead of living in /tmp.
• .env.example documents the environment knobs dw reads; copy it to .env
  (loaded at import, before torch) and restart the server after editing.

Web UI

• Per-card VRAM bars in the header, status popover and server page.
• Step and job durations on the job page, the log and the jobs list.
• The sidebar marks the open workspace; each workspace has its own hue.

Maintenance

• Architecture ratchet: bring the 0.10.0 re-baselined metrics back under the old baseline #790: task.py split by family, the H3 block classes moved to module
  level, task argument rules moved to dw/task_problems.py, and one
  iter_steps walk for every validation pass. CI runs the ratchet first
  so a test failure can no longer hide it.
• UI dependencies: vitest 5, monaco-editor 0.57; the lockfile update
  clears the source-map-js and postcss-selector-parser Dependabot
  alerts.

🤖 Generated with Claude Code