opcode-lucho

opcode-lucho is a desktop application (Tauri 2 + Rust on the backend, React 18 + TypeScript on the frontend) that gives Claude Code a graphical interface: project and session management, custom agents, MCP servers, checkpoints, and a token-usage dashboard. It is not an original product of mine or a web service, and it is not affiliated with Anthropic.

It is a fork of winfunc/opcode, licensed under AGPL-3.0. GitHub does not mark it as a technical fork because the repository is independent and imported the upstream history, so the only way to make that known is for me to say it: of the repository’s 522 commits, 321 are mine (250 excluding merges, across ten days of work from July 2 to July 11, 2026). The rest— the app concept, Tauri scaffolding, Claude Code agents, MCP, checkpoints, and many components—came from upstream. Cargo.toml still credits the original authors; I did not touch that line.

The fork starts at commit 70c16d8 (October 13, 2025, the last upstream commit before my first one). The README says it directly in an [!IMPORTANT] notice:

“This is a fork. opcode-lucho is a personal fork of winfunc/opcode (originally by Asterisk), licensed under AGPL-3.0… do not report issues about this fork upstream.”

I inherited the app’s entire concept and scaffolding: Tauri 2 + React, Claude Code agents, MCP, checkpoints, the web server (web_server.rs), and most of src/components/. I also inherited the AGPL-3.0 license and credits for the highest-volume upstream authors (Vivek R, Mufeed VH, Kiran Johns), which I left untouched in Cargo.toml.

What I did not inherit was any of the how it is maintained: no green CI, no internal documentation, and no navigation model I knew from memory. That is what follows.

None of this existed upstream. Every piece entered through its own PR—72 in total, with all work reviewed before it merged into master.

Before rewriting a view, I had to understand how the existing one worked, and upstream did not include that documentation. I ended up writing docs/views/: 16 English documents, one per view, written after building each corresponding feature—not as plans, but as the maps that would have saved me time had they existed earlier.

navigation-model.md is the one I use most. It describes how the sidebar, tab manager, and title bar relate, exactly the area rebuilt above. Documenting an unfamiliar codebase became as much a part of the work as writing new code.

The token-usage dashboard came from upstream, but I ended up rewriting it end to end, and it is the piece where I invested the most time. src-tauri/src/commands/usage.rs (+1,383 / −329 lines) is the most modified file in the entire repository—more than any navigation file. On the frontend, there is a whole new src/components/usage-detail/ directory (17 files), plus dashboard/DashboardBento.tsx and aggregation logic extracted to lib/usageAggregation.ts (+349).

The screenshot shows the result, not the starting point. Upstream had a usage view with disconnected metrics. I rewrote both how they are calculated (the Rust backend, where most of the diff landed) and how they are read: an eleven-metric bento at a glance, a model filter panel, and a temporal chart that crosses cost, tokens, and sessions with three representations—bars, candles, and line—instead of one fixed chart.

Upstream describes the improvement as “much faster loading on large histories.” I do not have a measured before-and-after for that claim, so I will not invent a number. It does match why I touched the backend in the first place: the original usage view did not handle a large session history well, and rewriting usage.rs answered that problem rather than a design whim.

Separating aggregation into a pure module (usageAggregation.ts) instead of mixing it into components was deliberate. It is the same criterion applied to the other twelve new modules under src/lib/: derived frontend logic lives apart from the view consuming it.

I also protected master and dev with no-bypass rulesets, including for me as admin. They require a PR and a green bun run check, and block force pushes. One fact I learned along the way: protection rulesets do not exist on GitHub Free for private repositories. The order was forced into public first, protection second, and that is what I did.

The project is closed as a version, not distributed. Anything from here will be targeted fixes, not a return to this pace of work.