● 12 articles on AI, development and infrastructure
Gemini CLI 0.63.0 stops even in YOLO mode with "Untrusted Command Flags Detected" on harmless commands. Here is why it happens, how to confirm it and the two ways to unblock.
Claude Code's auto memory index only loads 200 lines or 25KB, and the cut leaves out exactly the newest memories. The exact limit, the command to measure it, and how to rewrite MEMORY.md.
How to read the /context grid in Verboo Code, what each category means, the automatic warnings and how to free context with /memory, /mcp and /compact.
Articles on AI, coding, and developer tools. No spam.
Write your project rules once and Verboo Code loads them in every session. The 4 file levels, AGENTS.md vs CLAUDE.md, @ imports and per-folder rules with paths.
Across 24 AI-generated covers, 11 came out with text even with "no text" in the prompt. What fixed it was swapping objects that carry writing for abstract shapes.
/plan makes the agent read the project and write a plan before editing any file. See the commands, where the plan is saved and how to approve it.
Every WSL agent command fails with "os error 2" before the shell starts. The cause is in the Codex source, and a mount --bind fixes it while issue #49731 stays open.
Bash and Edit blocked by "auto mode classifier gave no verdict"? The server-side check failed, not your action. The 4 ways back to work, with commands.
/compact swaps the history for a summary. Learn how to say what to keep, when automatic compaction fires, and what to do when it fails.
A gender error went live in all 3 languages before anyone caught it. We wrote, tested, and fixed a bug in the checker script itself before publishing this article.
Four posts published this week had a diagram note cut off past the card border, fixed by hand every time. The real bug: the generator never measured text width before drawing. See the fix, with code, applied to render.py.
Today, when a task fails in Verboo Code with nobody watching, it just stays failed. Hook chains change that: an event-driven recovery layer that evaluates declarative rules and dispatches an action, like recovering with a fresh agent or notifying the team. A real feature, off by default. See the exact command to turn it on, the config JSON and the safety guards, all verified in the verbeux-ai/code source code.