docs(media): re-record queue pane demo in doom theme

[LeTuR]

Intent

The operator asked: "Redo a demo video of fleet queue custom pane with doom theme." The demo GIF in README.md was recorded while thurbox was on the Oxocarbon theme; the operator now runs thurbox's built-in doom theme, and the fleet lead answers in a VEGA register addressing the operator as SLAYER, so the picture should look like what fleet has become. Re-record media/fleet-queue-pane.gif through media/fleet-queue-pane.tape in that register.

Deliberate decisions, so the diff does not read as surprising:

1. 'Doom theme' was interpreted as the RECORDING's look, not a restyle of the pane. interface/fleet_queue.lua is deliberately UNTOUCHED: it draws through thurbox's theme roles so it matches whatever terminal theme its operator runs, and hard-coding colours there would break that for every user who is not recording a demo. That was an explicit constraint in the task.

2. A Set Theme was added to the tape with an inline JSON palette. vhs defaulted to a neutral #171717 background, which ringed the warm near-black TUI in ten pixels of grey padding. background and foreground are lifted from thurbox's own doom theme so the padding disappears into the pane; the sixteen ANSI slots are a dark red-accented set for the shell that runs before thurbox takes the screen, which the TUI itself never reaches because it addresses truecolour directly. The TUI's own theme is the operator's persisted active_theme, chosen with F4, and the tape CANNOT set it — vhs has no F-key command — so the tape header now says to read the theme name out of the frame's top-right corner before keeping a take. That limitation is documented rather than worked around.

3. The four next session steps were reduced to ONE. The tape's own header says the step count is part of the take and not a constant, and requires recounting against the live session list. Recounted: four workers in flight, two idle, one done, and the only one thurbox-cli session list called working was the lead's first child, which one step reaches. The requirement is that the centre pane ends on a worker that is actually working and whose topic the queue pane shows RUNNING, so the two halves of the frame are the same run of work. This is not a revert to the old flat-list count and the header records the recount.

4. The tape's honesty rule was followed rather than shortcut: it records the REAL queue and the REAL session list, no FLEET_QUEUE_DIR fixture. An earlier take used a fixture and produced a redacted column beside a status bar reading '3 session(s)', which reads as a rendering bug. The take was recorded while four tasks were genuinely in flight, and every frame was inspected: the session titles, the queue rows and the centre pane's transcript were read before committing, because this repo is public.

5. Nothing is typed into the centre pane once the agent tab is selected — an earlier take cleared its pane with clear, which would deliver clear to a live agent as a prompt. The centre stays read-only.

6. README.md's alt text was updated to describe the frame actually recorded: three running topics holding four tasks, not the six topics the old text described, plus the doom palette which is now visible in the picture.

7. The GIF was regenerated only through the tape plus the ffmpeg shrink step documented in the tape header (write to a temp file and mv, because ffmpeg reading and writing one path truncates its own input). It was not hand-edited. It is 873K, down from 1.0M, with no dwell trimmed and the F3 toggle — which is the demo — intact.

Taste constraints the operator stated: no emoji beyond the session glyphs the sessions already wear, no ASCII art, no title cards, no invented logo. The register lives in the palette and in the real session names on screen.

./scripts/check.sh passes locally.

What Changed

• Added a Set Theme block to media/fleet-queue-pane.tape with an inline JSON palette (doom's own background/foreground plus a dark red-accented ANSI set) so vhs's default grey padding no longer rings the near-black TUI, and documented in comments that the TUI's own active_theme still has to be set manually via F4 since vhs has no F-key command.
• Reduced the tape's next session step from four repetitions to one, with a comment recording the recount against a live session list (four workers in flight, one genuinely working).
• Regenerated media/fleet-queue-pane.gif from the updated tape (873K, down from 1.0M) against a real, unfixtured queue and session list.
• Updated README.md's alt text to describe the newly recorded frame: the doom red-on-near-black palette, four running tasks under three topics, and the centre pane showing the selected worker's transcript.

Risk Assessment

✅ Low: The change only touches README.md, the vhs tape script, and the regenerated GIF asset — no application code (interface/fleet_queue.lua confirmed untouched) — and every deliberate decision in the intent (single 'next session' step, real queue/session data, no typing into the centre pane, terminal-only theme palette, updated alt text) is verifiably reflected in the diff.

Testing

Baseline ./scripts/check.sh already passed; this is a docs/media-only change (README alt text, vhs tape, regenerated GIF) with no unit-testable code path, so validation was done by decoding the GIF into individual frames and visually confirming they match every claim in the intent: top-right corner reads 'Doom' theme throughout, the palette is the dark red-on-near-black doom set with no grey vhs padding ring, the session list shows the lead plus its 4 worker children each with a glyph, the queue pane shows exactly '4 running' tasks under 'RUNNING 3 topics' (ci-squash-coherence, demo-doom-theme, review-spscream-prs) with uncollected/unverified-artifact notes and a '54 archived' muted count, the centre pane holds the selected worker's real transcript (matching the 'demo-doom-theme' task the queue names running), and the F3 fleet-queue toggle visibly shows then hides the pane across the clip (frames 0/109 hidden, 40/70 shown) with nothing typed into the centre pane. All of this lines up exactly with README.md's rewritten alt text and the tape's own recount/theme/honesty commentary, and interface/fleet_queue.lua was confirmed untouched.

• Evidence: Frame 0 — boot, queue pane hidden, theme reads Doom (local file: /home/magicletur/.no-mistakes/evidence/01M2AJW5EMMKFGNC7VWZ7VHC60/frame-000-start.png)
• Evidence: Frame 40 — queue pane shown: 4 running tasks / 3 topics, session list with lead+4 workers (local file: /home/magicletur/.no-mistakes/evidence/01M2AJW5EMMKFGNC7VWZ7VHC60/frame-040-mid.png)
• Evidence: Frame 70 — queue pane still shown, mid-dwell (local file: /home/magicletur/.no-mistakes/evidence/01M2AJW5EMMKFGNC7VWZ7VHC60/frame-070-toggle.png)
• Evidence: Frame 109 — final frame, queue pane toggled hidden again (F3 round-trip) (local file: /home/magicletur/.no-mistakes/evidence/01M2AJW5EMMKFGNC7VWZ7VHC60/frame-109-end.png)

Pipeline

Updates from git push no-mistakes

✅ **intent** - passed

✅ No issues found.

✅ **Rebase** - passed

✅ No issues found.

✅ **Review** - passed

✅ No issues found.

✅ **Test** - passed

✅ No issues found.

• ./scripts/check.sh
• python3 json.loads() on the tape's Set Theme inline JSON — parses cleanly
• identify media/fleet-queue-pane.gif — 110 frames, 1000x679, confirms regenerated GIF is well-formed
• magick media/fleet-queue-pane.gif -coalesce to extract composited frames 0, 40, 70, 109 for visual inspection
• Visual inspection of extracted frames against README.md's updated alt text and the tape's own commentary
• git diff 285199c..5408816 --stat / -- README.md media/fleet-queue-pane.tape — confirmed no source/lua files touched
• git status --short post-testing — worktree left clean, no stray artifacts

✅ **Document** - passed

✅ No issues found.

✅ **Lint** - passed

✅ No issues found.

✅ **Push** - passed

✅ No issues found.