🦞🎬 StoryFlow — Studio Shell on omp

D-202 bounded continuation — browser phase complete, acceptance incomplete

D-202 now admits loca-hosted headless Chrome only through the public vhosts. A separately owned NI10 Chrome has completed genuine pxl sign-in to the public application, with public-IP resolver mapping to 178.18.250.170, TLS 1.3, authenticated API responses and WebSocket HTTP101. Loopback CDP is browser control only; no TLS bypass or loopback application route is accepted. The first browser-tooling probe is quarantined because its separate ownership, NI10 and launch arguments were not established.

The bounded browser phase is complete, not full-parity acceptance. Receipts cover populated imports and keyboard mapping, four compatible existing/local-derived images, independent compare slots, a promoted draft with exact Undo, three-deep resize lineage, real historical-video playback, downloads and literal naming. D-202 progress preserves the limits: no compared-winner/sibling proof, no selection-independent filename stability, installed-background WSS302 and narrow Queue layout failures, and no real credential lifecycle. The sole native invocation refused admission with checkpoint_clone_changed before provider activity; its allowance is consumed with zero provider POSTs and no replay authorized. All 41 exact scenarios and the immutable D-198 BLOCKED result remain unchanged.

An actual 4K download escaped to Chrome’s default Downloads outside the authorized evidence roots. It remains untouched and is an explicit scope violation; a later scoped download did not erase it. Bounded cleanup soft-deleted the owned batches/shots but retains protected media, stopped jobs, productions and audit records. Original application rows remain unchanged. Main owns final checks, publication, rendered-route verification and closure; these sources do not claim those later outcomes.

D-198 dated release boundary — 2026-09-28

D-198’s canonical face and gateway source are now deployed, with owner-host local proof: both health routes return 200 and all 13 served static files match canonical build hashes. This is not authorized external-browser acceptance; the mission verdict remains BLOCKED. Genuine existing-auth evidence remains four images, three videos and one terminal HTTP403 across eight native submission POSTs, with no retry or account/provider switch after refusal. Canonical daisyUI 5.7.0/TypeScript 5.9.3 builds/compiler-API smoke and the separate 55-test/12-HTTP-case duration guard pass, not a full-project typecheck. All 41 acceptance scenarios remain in scope; Public progress distinguishes commit delivery, deployment, native proof and missing authorized non-loca/non-Mac Chrome/realpxl UI/WS101 acceptance.

The diagram and studio loop below describe the production contract, not complete implemented integration. Its provider list is historical breadth, not the admitted pinned D-198 routes. Current architecture identifies the single RPC bridge, missing PWA/plugin/agent surfaces and other implementation gaps.

Thesis: omp is the body; StoryFlow is the always-on shell that probes, casts, produces, molts — and its face is one daemon, one port, three surfaces.

flowchart LR
  subgraph FACE["Face — three surfaces, one bundle"]
    W["Web app<br/>SvelteKit SPA"]
    P["Side panel<br/>WXT · Chrome"]
    M["Responsive mobile web<br/>PWA remains a design target"]
  end
  G["Gateway · Bun<br/>WS · REST · SQLite · queue · registry"]
  W --> G
  P --> G
  M --> G
  subgraph HOOD["Under the hood — the molty loop"]
    SH["Shell<br/>profile · channels · heartbeat"] --> RG["Registry<br/>capabilities.yml"]
    RG --> CA["Cast<br/>director · writers · refuters · crew · media"]
    CA --> GR["Grammar<br/>brief to bible to shots to assets to QA to ship"]
    GR --> MO["Molt<br/>learn · skills · extensions"]
    MO -.-> RG
  end
  G -- "JSONL stdio · rpc v2" --> O["omp --mode rpc<br/>--profile storyflow"]
  O --> SH
  G --> PR["Providers<br/>veo · grok-imagine · codex-images · xai-tts · ffmpeg · flow-relay"]

Section 0: One Breath

StoryFlow is a studio project built on omp; the D-198 owner is verified on omp 18.3.2. The intended studio loop is:

  1. Probes what it can do right now through live host probes.
  2. Casts models like a film crew across different provider families.
  3. Produces from a one-line logline to a shipped deliverable through one unified grammar.
  4. Molts (grows its own shell) after every run by harvesting lessons and skills.

The deployed gateway uses one local daemon on port 7540 for a static web SPA and shared Chrome-panel components:

  • Web app: SvelteKit adapter-static with fallback, Svelte 5 runes, Tailwind and daisyUI; canonical static bytes are deployed and locally hash-verified, not externally UI-accepted.
  • Chrome side panel: WXT shell sharing workspace-source components, with WebSocket in its background worker; installed-extension acceptance is outstanding.
  • Responsive mobile web: source layouts exist, but an installable PWA manifest/service worker is not implemented in the inspected face. Mobile interaction acceptance is also outstanding.

The frozen FlowTurbine archive defines the Queue workflow target, not completed parity. All 41 KEEP, ADD and DROP requirements remain in scope. Per-item commits and receipts are delivered, but they do not equal 41 accepted workflows. The deployed source contains authoring/import, Library, mapping, queue, compare/promotion, logs and credential surfaces. The parity table retains each acceptance boundary, including incomplete real credential-UI lifecycle, text-only native video and successful unlocked/reference-removal output. Captions remain unavailable.

The Two Human Gates

StoryFlow’s production contract specifies an automated swarm pipeline with exactly two human gates; current D-198 evidence does not establish that this complete production journey has run:

  1. Bible Veto Gate (Pre-Production): The Director model compiles the project bible, character sheets, and scene specs. Production halts until human sign-off via extension_ui_request / Approvals tab.
  2. Ship Gate (Post-Production): Independent QA agents grade the rendered artifacts against blockbuster metrics. The final release and market dispatch require explicit human authorization.

Core Wiki Navigation