Tools for working with coding agents without losing track of them, the Windows apps the platform should have shipped, and write-ups where the numbers come from records rather than recollection.
Claude Code will tell you your rate-limit percentage if you stop and ask it. This puts the number where you already glance: a native tray icon that redraws itself crisply at any DPI and updates on its own. It reads the session files already on your disk, so there is no account and no service.
Two product lines and a lab. A tool can sit in two of them.
Your Claude Code rate-limit percentage, always in the tray, as a crisp DPI-aware icon.
Docker on Windows without Docker Desktop: upstream Moby in a WSL2 distro it owns, on the pipe your tools already look for.
PlayStation Remote Play as a real Windows app — ten-bit colour, hardware decode on the card you already have, and no setup on first run.
SSH on Windows without the suite around it: the terminal grid drawn on the GPU, and a live prompt in under 400 ms.
Owns the writes to your ROADMAP, CHANGELOG and IMPROVEMENTS, so the format is a schema at insertion instead of a convention you remember.
Reads your staged diff, asks an LLM for a Conventional Commits message, shows it to you, and commits.
Drives a Windows desktop app from a test — four verdicts rather than two, so a check that never ran never reads as one that passed.
Makes 3D assets by declaring what they must satisfy and searching for the parameters that do, on Blender, Godot and a mesh service.
A piano Claude Code can write scores for, play and teach from: the notes fall onto the keys and you learn to play them.
A character-level GPT written from scratch in plain Java with no ML library at all — bigrams, then backprop by hand, then a small transformer.
What a Claude Code task really costs, and how far each plan carries you — every coefficient measured from real session transcripts.
Where the hours actually went on an AI-assisted project, attributed from session records and git history rather than recalled.
How to run a project where Claude Code writes the code and a person decides what gets built, with every rule resting on graded evidence.
Both of publish-release's cleanliness checks set benchmarks/results/gate-*.md aside, and the end of a release names the row left to commit.
The watch test holds the watcher to the window's own 300 ms quiet, settles and zeroes its count before writing, so a late inotify event cannot read as a burst.
A session fixture loads the operation registry before any test reads it, so a test run alone or on any xdist worker finds every operation.
A family keeps its name as written and its sheet takes a safe, unique file name, so a question mark no longer crashes a sitting (design recorded in docs/specs/adoption.md).
An answer counts only from when its sitting was laid out, so a re-laid sitting waits and review.state names its unanswered families (design recorded in docs/specs/acceptance-spec.md).
tools/operations_ts.py puts the tree's src first on the path, refuses another polyweave and names the one it described, so a stale install cannot write the SDK.
roadkeep governs the backlog and its desktop app reads every one, commitclerk writes the commit and winwright reports what a test observed.
Read the method →/plugin marketplace add alegauss/alegauss.github.ioroadkeep, winwright, polyweave, piano.
See the plugins →Every product here is made by Alexandre Oliveira (alegauss), a software developer in Brazil. He also builds Viglet, open-source enterprise search and content tools (Viglet Turing, Viglet Shio and Viglet Dumont), Cursarei, an independent directory of Brazilian courses, and the Japode apps.
No dashboards, no accounts, no background services you didn't ask for. A tool should do the one thing its name promises.
The tools collect nothing and phone nothing home. What runs locally stays local — the only network call is the one you configure yourself.
Where there are numbers, the inputs are yours to change and the method is written down. A model you can argue with beats a figure you have to trust.