15 rules on top
The patterns that kept coming back across all 675 lessons, each one a run that failed.
Never invent data to fill a gap
If the source lacks a number, queue the gap and say so.
Markdown · Pythonlive
Every mistake your AI agent will make, already made
675 lessons from shipping 60+ projects with OpenCode, Claude Code, Codex and Copilot, sorted into 12 chapters with the 15 that matter most on top.
2platforms: Markdown, Python
MITopen source licence
Why it exists
Every line cost a failed run.
Hard-won lessons from building and shipping 60+ projects with AI coding agents (OpenCode, Claude Code, Codex, Copilot, Gemini, Vector) on a locked-down Windows work laptop, as one person who is an engineer, not a programmer.
Every line here cost at least one failed run.
In short
How to use the handbook.
Drop-in for agents
One line in AGENTS.md, CLAUDE.md or copilot-instructions.md points your agent at the whole handbook.
Build your own
tools/build.py turns your own lesson notes into the same format and refuses to publish if a private term survives.
What's inside
The rules that kept coming back.
Check every capture before judging it
Rounds were lost to bad evidence (a page that had not loaded, an error toast, an unstyled sheet), not bad work.
Parallel agents need their own checkout
One worktree or clone per task, fetch and rebase before every push, claim queue items before starting, and kill only your own processes.
Tool calls die; files survive
Shell calls time out and interpreters restart, so long jobs write their progress to files and resume from them.
"Make it like X" means X's interaction model
Check what the reference actually does before building (a 3D walkable space is not a 2D floor plan in the same colours).
The app is the source of truth
Order of work: app, then brand, then landing page.
When the human says "finish it", finish it
Work around hand-off tasks instead of asking, but treat real access walls (org policy, other people's accounts, codes only the human can type) as evidence, not obstacles to route around.
Questions
Good questions.
What does it run on?
Markdown, Python. You can try it in the browser with nothing to install.
Is it finished?
Not yet: it's marked “live” and still changing. The README always has the current state.
Is it free?
The source is public on GitHub under the MIT licence.
Who made it?
Juwen Lie, an R&D mechanical engineer in Batam, Indonesia, as an independent personal project.
Keep exploring
More from the same workbench.
Let your agent skip the mistakes.
Every mistake your AI agent will make, already made