Uncle Bob's ub-mode Puts Quality Gates on AI Coding Agents

AI coding agents are rarely short on speed. They're short on a definition of "done that isn't embarrassing." Uncle Bob, author of Clean Code, has open-sourced ub-mode, an agent skill that stuffs old-school engineering discipline into the automated coding loop.

ub-mode at a glance

• Who built it: Uncle Bob, author of Clean Code

• What it is: an open-source skill for AI coding agents

• Order of work: make it run, then make it right; tests first

• Quality gates: complexity, duplication, and mutation testing

"Run first, then fix" blocks fake completion

A lot of agents jump straight into big structural rewrites, leave the build broken and the suite red, and walk away from a half-finished mess. ub-mode insists the code runs before it gets cleaned up. That sequence is a constraint, not a slogan.

Tests-first also redefines done as verifiable behavior, not "the model said it finished." For automated coding, that beats another paragraph of prompting.

Metrics as gates, not vibes

Complexity, duplication and mutation testing are veteran tools for checking whether code is healthy. Wiring them as gates tells the agent: fail these checks and you haven't delivered.

AI coding tools keep getting stronger; what's often missing is a standard. Uncle Bob shipping a skill himself guarantees attention. The useful part is whether teams can drop it into the agent workflows they already run, so automated output has to clear a real quality bar.