Ghost Hand
Public Unreviewedby John Lauer
A small battery gadget that keeps an automatic sink faucet running while you use the sink: it learns the faucet's timeout, re-arms with the shortest blip, and stops after 10 minutes or when picked up.
| Name | Last updated |
|---|---|
| 3d-1-sheet.png | 1d ago |
| 3d-1.mp4 | 1d ago |
| design-1-sheet.png | 1d ago |
| design-1.mp4 | 1d ago |
| export-2-sheet.png | 1d ago |
| export-2.mp4 | 1d ago |
| fields-1-sheet.png | 1d ago |
| fields-1.mp4 | 1d ago |
| fields-2-sheet.png | 1d ago |
| fields-2.mp4 | 1d ago |
| final-1794192182.mp4 | 1d ago |
| README.md | 1d ago |
| rules-1-sheet.png | 1d ago |
| rules-1.mp4 | 1d ago |
| shot-design-1-end.png | 1d ago |
| shot-design-1-start.png | 1d ago |
| shot-rules-1-end.png | 1d ago |
| shot-rules-1-start.png | 1d ago |
| shot-sourcing-1-end.png | 1d ago |
| shot-sourcing-1-start.png | 1d ago |
| shot-test-plan-1-start.png | 1d ago |
| sourcing-1-sheet.png | 1d ago |
| sourcing-1.mp4 | 1d ago |
| test-plan-1.mp4 | 1d ago |
AI Flow run: claude-opus-5-5 on board-0
Delivered at 2026-11-09T02:52:09Z: 528.7 min of AI time from the prompt to "done, here is your video" (632.1 min on the wall clock).
Prompt at 2026-11-08T16:20:00Z (UTC). Flow: board. Every number here is from the run's append-only ledger (run.jsonl). AI time is the time the AI was working on a prompt: from the human's prompt to the AI's done, with any gap over 15 min where no command ran (unobserved, not proven idle) cut out.
The video
The final cut of this run, 144 s: only the moments something moved, every step in order, the AI's analysis drawings fullscreen, the run and step clocks in the corner, the words cut to the picture.
Sessions
One row per human prompt: the AI's time on it, unobserved gaps excluded by the heuristic, which may include external AI work.
| Session | Prompt at | Done at | AI time min | Unobserved gap min | What |
|---|---|---|---|---|---|
| 1 | 2026-11-08T16:20:00Z | 2026-11-09T02:52:10Z | 528.7 | 103 | the first prompt |
| all | 528.7 |
Per step
AI time only, the idle cut out.
| Step | AI min | AI thinking min | Tool min | Turns | Visits | Returns | Longest think |
|---|---|---|---|---|---|---|---|
| intake | 13.2 | 13.2 | 0.0 | 1 | 1 | 0 | 13.1 min |
| placement | 41.5 | 26.5 | 15.0 | 31 | 3 | 2 | 7.6 min |
| routing | 82.3 | 40.5 | 41.8 | 34 | 4 | 1 | 14.9 min |
| pours | 2.9 | 0.3 | 2.6 | 16 | 3 | 1 | 0.1 min |
| current | 4.2 | 3.3 | 0.9 | 10 | 2 | 0 | 1.9 min |
| thermal | 0.2 | 0.1 | 0.0 | 2 | 1 | 0 | 0.1 min |
| finish | 2.9 | 1.2 | 1.7 | 11 | 2 | 0 | 0.4 min |
| 3d | 10.8 | 9.1 | 1.7 | 5 | 2 | 0 | 8.0 min |
| components | 0.1 | 0.1 | 0.0 | 2 | 1 | 0 | 0.1 min |
| design | 1.1 | 0.1 | 1.1 | 5 | 1 | 0 | 0.1 min |
| export | 20.0 | 16.7 | 3.3 | 13 | 2 | 0 | 10.1 min |
| fields | 47.7 | 1.4 | 46.3 | 6 | 2 | 1 | 0.9 min |
| library | 0.0 | 0.0 | 0.0 | 1 | 1 | 0 | 0.0 min |
| models | 0.6 | 0.1 | 0.5 | 2 | 1 | 0 | 0.1 min |
| nets | 10.6 | 9.0 | 1.5 | 4 | 2 | 0 | 8.8 min |
| outline | 1.0 | 0.4 | 0.6 | 10 | 3 | 0 | 0.2 min |
| parts | 11.5 | 7.5 | 4.1 | 12 | 1 | 0 | 5.2 min |
| project-page | 1.4 | 0.5 | 0.9 | 5 | 2 | 0 | 0.1 min |
| rules | 0.9 | 0.8 | 0.1 | 2 | 1 | 0 | 0.8 min |
| schematic | 0.3 | 0.1 | 0.2 | 2 | 1 | 0 | 0.1 min |
| silkscreen | 239.4 | 188.8 | 50.6 | 174 | 3 | 1 | 12.7 min |
| simulation | 0.1 | 0.0 | 0.1 | 3 | 1 | 0 | 0.0 min |
| sourcing | 9.4 | 3.2 | 6.2 | 2 | 1 | 0 | 3.2 min |
| test-plan | 26.4 | 25.8 | 0.6 | 5 | 1 | 0 | 13.6 min |
| thermal-vias | 0.2 | 0.1 | 0.1 | 6 | 2 | 0 | 0.1 min |
| total | 528.7 | 348.9 | 179.8 |
Returns to earlier steps
- 2026-11-08T19:00:43Z back to placement (visit 2): U1's bottom-row pins (RX_SIG, LED_DIN) and right-column pins (SDA, VBUS_SENSE) could not escape: U5's top pads sit 1 mm below U1; move U5 1 mm toward the front edge
- 2026-11-08T19:29:50Z back to placement (visit 3): three links boxed in after routing: move test pad TP4 next to U2's PROG pin and R8 (VBAT sense divider) next to the VBAT copper at U3
- 2026-11-08T20:11:00Z back to pours (visit 2): the native save after landing the B.Cu pour voided the pour record; land the same plan again to re-record it
- 2026-11-08T20:15:05Z back to routing (visit 4): current screen: Freerouting drew the power nets at the 0.12 mm default; widened VBAT, the emitter nets, VBUS, 3V3, VLED and GND wherever clearance allows (rev2/tools/widen.py, 585 segments, DRC clean)
- 2026-11-09T00:56:47Z back to fields (visit 2):
- 2026-11-09T02:30:11Z back to silkscreen (visit 3): director 19:55: light-row silkscreen fix (gauge names centred over their LEDs, D5..D8 in a row, D4 with STATUS); silk only, copper unchanged
Decisions along the way
- 2026-11-08T16:33:15Z clip-start:rules-1
- 2026-11-08T16:35:08Z clip-start:design-1
- 2026-11-08T16:35:46Z clip-start:test-plan-1
- 2026-11-08T17:01:38Z clip guard: the test-plan clip ran 25 min, past its 10 min budget twice over; stopped by the guard
- 2026-11-08T17:07:46Z clip-start:sourcing-1
- 2026-11-08T20:41:54Z net GND: 628 items highlighted
- 2026-11-08T20:41:58Z net VBAT: 71 items highlighted
- 2026-11-08T20:42:02Z net VBUS: 62 items highlighted
- 2026-11-08T20:42:07Z net 3V3: 89 items highlighted
- 2026-11-08T20:42:11Z net IR940_D: 14 items highlighted
- 2026-11-08T20:42:15Z net IR850_D: 14 items highlighted
- 2026-11-08T20:42:19Z net RX_SIG: 32 items highlighted
- 2026-11-08T20:42:23Z net TIA_IN: 22 items highlighted
- 2026-11-09T01:20:45Z net GND: 628 items highlighted
- 2026-11-09T01:20:50Z net 3V3: 89 items highlighted
- 2026-11-09T01:20:55Z net IR850_D: 14 items highlighted
- 2026-11-09T01:21:00Z net IR940_D: 14 items highlighted
- 2026-11-09T01:21:05Z net VBAT: 71 items highlighted
- 2026-11-09T01:21:11Z net VBUS: 62 items highlighted
- 2026-11-09T01:21:16Z net IR850_K: 17 items highlighted
- 2026-11-09T01:21:21Z net IR940_KA: 8 items highlighted
Where the board stands
- routing: 0 unrouted after null passes
- gate: 0 new DRC errors, 0 unconnected, null inherited
- landed live: 0 of 0 route entries
- pours (landed): 0 groups, failed []
- copper kept (mm2 per layer): {"B.Cu":348.49,"F.Cu":292.5}
- analyze-current: pass
- analyze-thermal: pass
- acknowledged current peak: GND 88.2 A/mm2 at (117.50, 89.30) ([email protected],89.3): The spec load here is the IR echo pulse peak (0.83 A for 25 us at most 20 Hz, 0.05 % duty); the pulse is supplied mostly by C15 (100 uF) 2 to 6 mm from the emitters, and this neck is on the battery-side path, hemmed in by other nets (left edge: MC3/3V3 and the IR850 gate tracks; beside U1: SDA, SCL, VBUS_SENSE and 3V3), so it cannot widen without moving the molecule contacts. Physics: RMS current is about 19 mA; adiabatic rise over one 25 us pulse at 186 A/mm2 is J^2 rho t / c_v = 0.004 C, and the Fields thermal solve (24 C peak, from U2 charging) is unaffected. (by Ghost Hand rev 2 board)
- acknowledged current peak: VBAT 185.9 A/mm2 at (100.30, 93.10) ([email protected],93.1): The spec load here is the IR echo pulse peak (0.83 A for 25 us at most 20 Hz, 0.05 % duty); the pulse is supplied mostly by C15 (100 uF) 2 to 6 mm from the emitters, and this neck is on the battery-side path, hemmed in by other nets (left edge: MC3/3V3 and the IR850 gate tracks; beside U1: SDA, SCL, VBUS_SENSE and 3V3), so it cannot widen without moving the molecule contacts. Physics: RMS current is about 19 mA; adiabatic rise over one 25 us pulse at 186 A/mm2 is J^2 rho t / c_v = 0.004 C, and the Fields thermal solve (24 C peak, from U2 charging) is unaffected. (by Ghost Hand rev 2 board)
The steps on camera
One clip per step visit, sped up 10x, with the editor at the start and the end of the visit. A caption above every clip and screenshot says what it shows.
rules
What this step does:
Clip: rules, the PCB editor from 2026-11-08T16:33:15Z to 2026-11-08T16:34:12Z. Motion only: 0.9 min of real time, 41 s of it had something moving; the clock in the corner is the real elapsed time.
Contact sheet: nine frames spread through the rules clip, what the camera saw.
Screenshot: the board when rules began.
Screenshot: the board when rules ended.
design
What this step does:
Clip: design, the PCB editor from 2026-11-08T16:35:08Z to 2026-11-08T16:35:19Z. Motion only: 0.2 min of real time, 8 s of it had something moving; the clock in the corner is the real elapsed time.
Contact sheet: nine frames spread through the design clip, what the camera saw.
Screenshot: the board when design began.
Screenshot: the board when design ended.
test-plan
What this step does:
Clip: test-plan, the PCB editor from 2026-11-08T16:35:46Z to 2026-11-08T17:02:59Z, 25.9 min of real time shown at 10x (0 s).
Screenshot: the board when test-plan began.
sourcing
What this step does:
Clip: sourcing, the PCB editor from 2026-11-08T17:07:46Z to 2026-11-08T17:11:15Z. Motion only: 3.3 min of real time, 132 s of it had something moving; the clock in the corner is the real elapsed time.
Contact sheet: nine frames spread through the sourcing clip, what the camera saw.
Screenshot: the board when sourcing began.
Screenshot: the board when sourcing ended.
fields
What this step does:
Clip: fields, the PCB editor from 2026-11-08T20:18:16Z to 2026-11-08T20:31:18Z. Motion only: 10.4 min of real time, 451 s of it had something moving; the clock in the corner is the real elapsed time.
Contact sheet: nine frames spread through the fields clip, what the camera saw.
fields, visit 2
What this step does:
Clip: fields, visit 2, the PCB editor from 2026-11-09T00:57:06Z to 2026-11-09T01:09:31Z. Motion only: 10.4 min of real time, 313 s of it had something moving; the clock in the corner is the real elapsed time.
Contact sheet: nine frames spread through the fields, visit 2 clip, what the camera saw.
3d
What this step does:
Clip: 3d, the PCB editor from 2026-11-09T01:37:59Z to 2026-11-09T01:38:34Z. Motion only: 0.4 min of real time, 6 s of it had something moving; the clock in the corner is the real elapsed time.
Contact sheet: nine frames spread through the 3d clip, what the camera saw.
export, visit 2
What this step does:
Clip: export, visit 2, the PCB editor from 2026-11-09T02:33:55Z to 2026-11-09T02:34:26Z. Motion only: 0.4 min of real time, 10 s of it had something moving; the clock in the corner is the real elapsed time.
Contact sheet: nine frames spread through the export, visit 2 clip, what the camera saw.
Delivered
What the AI said when it delivered (the video is at the top of this page):
Ghost Hand rev 2 board: 30 x 22 mm two-layer molecule, routed, poured, Fields-analysed, two-sided silkscreen with the light-row fix, exported (board-25); fab files and design zip in wiki release john/ghost-hand 0.2.0 Rev 2. Video: 144 s of real screen clips from this run (silkscreen landing, Fields, nets, 3D, AI Flow dashboard).
Plan usage, the second cost line
Percent of each plan limit used, as the provider reports it, at the start of the run and at its end.
| Provider | Limit | At start | At end | Used by this run |
|---|---|---|---|---|
| antigravity | 3p week | 0.00 % | 0.00 % | 0.00 % |
| antigravity | week | 0.00 % | 0.00 % | 0.00 % |
| claude | Fable | 0.00 % | 0.00 % | 0.00 % |
| claude | week | 8.00 % | 14.00 % | 6.00 % |
| codex | week | 1.00 % | 1.00 % | 0.00 % |
The limits are per account, not per run: every AI thread signed in as this user draws on the same numbers, so a run made while other threads were working shows their usage too. Read it as an upper bound unless this run was the only thread active.
Google Antigravity: tokens, quota and cost
0 model calls: 0 fresh input, 0 cached input and 0 output tokens. The run used 0.00 % of the weekly limit, about $0.00 of a $199.99/month plan.
| Step | Calls | Fresh input | Cached input | Output |
|---|
Token counts are the difference of Antigravity's own per-call records (its conversation files) between step snapshots; exact when nothing else used Antigravity during the run.
Given back
2 issue(s) or PR(s) filed by this run's AI to make the tool better for every Adom user:
- issue https://wiki.adom.inc/adom/adom-aiflow/issues/178
- issue https://wiki.adom.inc/adom/adom-aiflow/issues/179
Adom processes PRs within minutes during business hours: a non-stop PR intake runs all day.