How to use Claude to build a tested circuit viewer
Claude Code built circuit viewers and real-server tests that caught a host form bug; published plugins retain visible routing failures and explicit host limits.
punkfab.com
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Wrap our circuit-check pipeline in a local MCP viewer with board, netlist and gate-result surfaces. Test every tab against the real server, verify host sandbox interaction constraints, and refresh geometry and reports from the same board revision. Keep open nets visible and publish the exact platform and host coverage before release.
Reach people exploring AI agents
See placements →What happened
The published plugin and mod display native board evidence and pipeline verdicts, with a host interaction bug caught by tests and unsupported surfaces explicitly qualified.
Reported by Dan Newcome — PUNKFAB logHow the agent did it
- Claude Code implemented a shared local backend, Codex plugin and Claude mod around existing board-check scripts.
- Real browser/server tests caught a project-picker form that the ChatGPT host sandbox could not submit.
- The owner published the viewers with separate Codex patches and explicit host and platform testing limits.
Where a person stepped in
Newcome reviewed the visible artifacts and published specific testing limits: Linux only and no observed Claude Desktop SVG rendering; unresolved board routing remained visible.
Dan Newcome used Claude Code to turn an electronics checking pipeline into two conversation-side viewers: a Codex plugin and a Claude Code mod. The software task was keeping a real routed board, its markers and the actual gate verdict visible together.
Claude implemented the front ends, shared backend and tests. Browser tests exercised each tab against the real local server and caught a host-specific defect: the project picker used a form submission that ChatGPT’s sandbox would not allow. The implementation therefore had to respect the host’s interaction boundary, beyond merely looking correct in an ordinary browser.
The viewer exposes board, schematic, netlist, 3D and checking surfaces. Its three agent tools open the board, run checks and retrieve the netlist. Separate Codex contributions added live refresh and a critical-routing gate; those are distinguished from Claude’s main implementation.
The published artifact displays remaining open connections rather than repairing them automatically. Testing was Linux-only, and the Claude desktop SVG rendering had not been seen in the actual desktop host. The useful result is a tested plugin with explicit environment limits, backed by the same checking scripts as the original pipeline.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: PUNKFAB log · Dan Newcome
Frequently asked questions
Why test against the real server?
The project-picker behavior had to work within the host sandbox.
Does the viewer repair open connections?
No. It displays them and the failing verdict.
Were all desktop hosts tested?
No. Linux was tested, and the Claude Desktop SVG view had not been observed.
More details
About this account
Claude Code implemented a shared local backend, Codex plugin and Claude mod around existing board-check scripts. Real browser/server tests caught a project-picker form that the ChatGPT host sandbox could not submit.
Original dated implementation log and source repository opened. Claude built the viewers; Codex’s ArtNode-related contributions are attributed separately. This entry concerns plugin software and host testing, distinct from the board export and power-layout accounts. Suggested prompt written by Agents In Use; no original task prompt was found in the verified text.
Credit: Dan Newcome — PUNKFAB log
Source published 4 Oct 2026
Source checked 14 Oct 2026
Tools and topics
Claude Code · MCP Apps · KiCad · Node.js · Python · Browser tests
plugin implementation · host constraints · artifact freshness
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.