Skip to content

The landing page's story shows the Row-Bot 5.0 React app - #382

Merged
siddsachar merged 3 commits into
mainfrom
chore/landing-react-media
Oct 2, 2026
Merged

siddsachar merged 3 commits into
mainfrom
chore/landing-react-media

Conversation

@siddsachar

Copy link
Copy Markdown
Owner

Summary

The landing page's four story scenes (Research, Create, Automate, Ship) showed the retired NiceGUI app. They now show the Row-Bot 5.0 React app, reviewed and approved by the owner.

  • Stills docs/media/landing-story/screenshots/{research,create,automate,ship}.webp: 1440×810 WebP, 62–70 KB.
  • Clips clips/*.webm (VP9) and *.mp4 (H.264, faststart): 1440×810, 30 fps, no audio. Research 9.4 s, Create 9.4 s, Automate 8.8 s, Ship 2.9 s (the page turns Buddy to success 2.25 s in).
  • Scenes:
    • Research: a local-model conversation (qwen3.8:27b) with its source trace, then the Knowledge graph settling.
    • Create: GPT-5.6 Sol with an editable landing page in the Design panel.
    • Automate: delivery kept in the app, the workflow saved with qwen3.8:27b, its run history.
    • Ship: a "Send an email?" approval naming the PDF and recipient, then Approve.
  • manifest.json and recording-receipt.json: new hashes, durations and alt texts, a production block that says how the footage was made, and review_status: approved.
  • docs/index.html: clip cache keys and alt texts only; the "Real app capture · Recorded in-product" framing is kept, as the owner chose.
  • scripts/docs/capture_landing_media.py (new) reproduces the footage. It reuses the docs capture's launcher and display-only local runtime. It runs the real app and its React client in Edge on an isolated profile seeded with fictional data (a neighbourhood solar co-op). No model runs; the frontier model shows as connected only because the harness says so, and the harness answers Ship's Approve itself, so nothing is sent. Stable relation ids make the graph layout deterministic.

Type of change

  • Docs

Risk area

  • Other (marketing site)

Testing

  • uv run python scripts/run_test_matrix.py docs: 109 passed (landing story hashes, sizes, durations, frame rate, runtime behaviour, approved manifest)
  • Every still and 3–4 frames per clip checked by eye, and the clips seen in place on the landing page
  • Not applicable, docs-only change

Release notes

  • Internal-only change, no release note needed

Checklist

  • Branch is based on latest main
  • No direct secrets, API keys, local paths, or private data included
  • The change is focused and does not include unrelated cleanup
  • Windows/macOS behavior considered where relevant

🤖 Generated with Claude Code

siddsachar and others added 3 commits October 2, 2026 02:54
The four story stills and clips (Research, Create, Automate, Ship) were
recorded in the retired NiceGUI interface. They are now scripted captures
of the real 5.0.0 React client, on an isolated temporary profile seeded
with a fictional neighbourhood solar co-op: no real user data, accounts
or model calls, and nothing is sent.

- Research: a local qwen3:8b conversation with its source trace, then
  the Knowledge graph (103 memories, 170 links) settling.
- Create: GPT-5.6 Sol (display-only) in the model picker beside an
  editable landing page in the Design panel.
- Automate: workflow delivery kept in the app, the local model saved,
  then the retained run history.
- Ship: an email to the co-op board waiting for approval, then Approve
  (answered by the harness), on the page's 2.25 s Buddy cue.

scripts/docs/capture_landing_media.py seeds the profile, drives the app
with Playwright and encodes with ffmpeg from imageio-ffmpeg (uv run
--with; not a project dependency). The manifest and recording receipt
carry the new hashes and say how the media was made; the manifest is
pending_review until the owner approves it. The landing page bumps the
clip cache keys and describes the new stills.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ph labels stand clear

Automate is captured again on top of the workflow profile fix, so the
Workflows list names the agent profile "Default", as the run drawer does.

Research is captured again with a knowledge graph whose labels no longer
overlap: the seeded relations now have fixed ids as well as the memories
(the graph reads its edges in id order, which steered the layout and made
it differ between captures), and the id seed was chosen by measuring label
overlaps across seeded profiles. The manifest and recording receipt carry
the new hashes; the manifest stays pending_review.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…s media are approved

The display-only local runtime in the landing capture now lists
qwen3.8:27b, the local model the owner runs, in place of the docs
capture's demo models (overridden in the landing script, so docs
screenshots are unchanged). Every local conversation, the workflows and
the Create picker's Ollama Local group show it. All four scenes were
captured again from the current client build with the same layout seed,
timings and Buddy-safe framing; the Research and Create stills come out
byte-identical, as they show no local model.

The owner reviewed and approved the footage: the manifest records
review_status "approved" with its review time (capture_landing_media.py
--records --approve), and the receipt names the model it shows.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@siddsachar
siddsachar merged commit dc1fad8 into main Oct 2, 2026
16 of 19 checks passed
@siddsachar
siddsachar deleted the chore/landing-react-media branch October 2, 2026 02:34
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant