I just released Patchmill v1.0
It is a complete rebuild. The old workflow engine is gone, and a small kernel takes its place: it starts Pi, claims issues, reads git and the issue host, and checks what the agent actually did. Everything that needs judgment, such as how to plan, code, review, and write, now lives in skills you can read and change. The production code went from 53,542 lines to 4,300.
On top of that kernel, this release adds two things we've wanted for a long time. You can now plan an issue with Patchmill in a chat, approve the spec from a dialog, and let a background worker build it. And you can run Patchmill without touching your tracker's labels at all. The skill pack also grew: frontend design with real HTML mockups, an independent frontend review, plain-English writing rules, and an implementation skill that always gets a fresh reviewer.
This is a breaking release. There is no upgrade path from 0.20 configuration or saved state. Read Upgrading below before you run it on an existing repository.
https://t.co/bP2F0FaoAI
the most import skill in 2026 is still being able to tell your agents "wtf is this crap, no, not like this"
if you stop doing that, you're basically fucked, despite what some devfluencers would like you to think.
curious how this is effected by N humans. hypothesis:
for
N=0-1 Agent adversarial review mostly effective
N=2 Spy vs Spy, but still pretty effective
N>2 Spy^N creates toxic context and time fragmentation
Quality plummets.
I had my agents analyze thousands of recent adversarial reviews in my system. Something like a quarter overall were blocked by the reviewer, and a meta-review showed the issues raised by agentic reviewers were almost all legit. I disallow filing P3s, which helps a lot. But in a nutshell, they work for me and I'm keeping them.
Some mistakes you might be making:
Your effort level is too high
Your skills are too long
You're not aligning before you start work
You're aligning too often, and too hard
Your automated checks are too lax
Your custom tooling is too token-inefficient
You aren't doing enough automated review
You aren't reviewing previous coding sessions (with an agent)
You're doing too much human review
You aren't vibe coding enough prototypes
Your agent can't verify its own work
You're babysitting the agent too much
You've overloaded your global steering files
Your codebase is hard to navigate
You aren't building a shared language
Your codebase architecture allows bad tests to be written
You aren't thinking strategically enough
You aren't thinking enough