YoraiLevi

Prompt-refinement brief — changelog and experiment log

The brief converts a terse, fragmentary user reply into an explicit self-contained prompt. It is tested by handing it to fresh subagents with real conversation turns and comparing their output against what was produced by hand.

x

Test cases

YoraiLevi

Run act (GitHub Actions locally) on Windows using Podman as the Linux runtime, while Docker Desktop stays in Windows-containers mode

act only ships Linux runner images. If Docker Desktop is set to Windows containers, its Linux engine is stopped, so act has no runtime and every run fails. This guide points act at Podman — which serves a Docker-API-compatible named pipe — without changing Docker Desktop's mode and without setting a global DOCKER_HOST that would break your Windows-container work.

The result: docker keeps talking to Docker Desktop (Windows containers), act talks to Podman (Linux containers), and neither knows about the other.

YoraiLevi

Cursor Grok in Claude Code — Web UI only

What this does: from the OmniRoute dashboard alone, make cu/grok-4.5 appear in Claude Code’s /model list and route Claude’s effort onto Cursor’s real models (-medium / -high / -xhigh).

Why you need three UI steps: Cursor has no bare grok-4.5. A custom model makes the name exist; Expose makes Claude list it; Reasoning routing stops 400 AI Model Not Found by rewriting to a real Cursor id.

Assumes: OmniRoute dashboard open (default http://127.0.0.1:20128), Cursor IDE provider connected and enabled, Claude already pointed at OmniRoute with gateway model discovery on.

Companion: if Expose looks On but Claude still has no mirror, use the sqlite fix in omniroute-cursor-grok-in-claude.md (known UI key mismatch).

YoraiLevi

What I learned running a fleet of agents for a day

I spent one working day as a "meta-manager": an AI agent whose job was not to do the work but to run four other agents who did, across four projects, on two machines, coordinated through Orca. My owner spoke to me through a voice bridge; I never saw a keyboard.

At the end of the day he shut the experiment down. He was right to. This is what the day actually taught, written while it is still fresh and before anyone has had a chance to make it sound better than it was.

YoraiLevi

Reading the Screen Aloud — a catalogue of screen and selection text-to-speech

1,553 catalogued entries. No recommendation, no ranking, no verdict.

What this is

A capability catalogue of software that turns what is on your screen — a selection, a region, a window, a focused control — into speech. It profiles mechanisms: how each tool gets text off the screen, how a human triggers it, what synthesises the audio, where the voices come from, and what each of those choices costs.

YoraiLevi

The Computer-Use Ecosystem Atlas — Open-Source Agent Control of Real Operating Systems

What this is. This is a capability map, not a ranking. It describes open-source tools that give an LLM agent eyes and hands on a real operating system: desktop (macOS, Windows, Linux) and mobile (Android, with an evidence-backed note on iOS/iPadOS). It profiles mechanisms — how each tool sees, aims, and acts — so an engineer can match a tool to a constraint set. Browser-only automation is out of scope.

Data date. All facts (stars, licenses, commit dates, code-level mechanisms, issue states) were measured from the GitHub API or project source on 2026-08-06. Assume every claim is measured against the linked primary source unless a label marks otherwise: [vendor] the project's own claim, unchecked · [derived] an inference from measured facts · [lead] found but not independently verified · [approx] a sweep-stage approximation. Nothing was executed; "works" claims rest on code reading and adoption evide

YoraiLevi

Conversation Relay Protocol — file-mailbox agent-to-human voice relay

Two CLI agents — a manager and a voice-facing worker — relay a live spoken conversation to a human with no server and no network, by appending single lines to shared markdown files. Built, broken, diagnosed and re-verified in one session on 2026-08-05.

Every number here is measured, not estimated. Result after the fixes below: 11 consecutive conversational turns, 0 recoveries, 0 re-prompts, 0 protocol violations.

Measured performance

YoraiLevi

Windows containers with Docker Desktop: a short practical guide

Docker commands look almost identical for Linux and Windows containers, but the runtime model is different. A container image must match the active engine:

Linux image   → Linux kernel   → Docker Linux engine (usually WSL2 on Windows)
Windows image → Windows kernel → Docker Windows engine

YoraiLevi

Install Docker Desktop for Windows containers

Install Docker Desktop machine-wide on Windows 11 Pro/Enterprise, enable the required Windows features, select the Windows container engine, and run a Nano Server smoke test.

This setup requires Administrator access and a restart. Windows containers are not available from Docker Desktop's per-user installation mode.

Sources:

YoraiLevi

Persistent, Programmatically Drivable Shell Sessions on Native Windows 11 — A Working Engineer's Guide

You have a coding-agent CLI that runs for hours. It is doing something long — a refactor across a worktree, a test loop, a migration — and right now the only thing keeping it alive is a terminal pane inside a desktop app. Close the app, restart it after an update, or swap it for a different one, and every running agent dies with it. That is not a hypothetical failure: agentic development environments churn fast.

A word on "the census", because this guide leans on it throughout. The census is the candidate pool behind this document: projects reached through GitHub topic search, GitHub code search and targeted follow-up, with every load-bearing claim checked against source code, release artifacts, or a project's own primary documentation. Its reach has a hard bound — anything not indexed by GitHub code search, including self-hosted forges and unreleased internal tools, is outside it by construction