DryforgeJust explain it.
We'll take the rest.
Your agent works like a senior developer.
From the first messy idea through the last line of code — one plugin, the whole thread.
/plugin marketplace add fn-opt/dryforge /plugin install dryforge
Don't stack code on a guess
Ask properly up front and you spend less time building the wrong thing and tearing it down. A few questions now beat a big rewrite later.
The behavior you want, the edge cases, the rules that have to hold — none of that is in the code. Those get asked until they're clear.
The agent researches the implementation. It only stops to show trade-offs when the choice actually changes the result.
Answers it can read from the codebase, or choices that don't change the outcome, it handles on its own.
If a bad assumption only shows up at the last check, you asked too late.
More process does not mean better work
More checks look safer. Then the agent works to clear the checklist, and the thing you wanted built comes second. Dryforge draws the lines that must hold and leaves judgment inside them. When models get better, an old procedure does not hold them back.
If it has to step outside that scope, it does not go ahead on its own. It asks again.
“Done” is not evidence. What counts is a command that actually ran.
What you decided and why is written into the project. The next conversation starts by reading it.
Just explain it. /ready will take it from there.
Install the plugin, then write one line about what you want.
Two commands. One loop.
The next task starts here
If the project already has code, run /migration once at the start.
/ready takes unfinished thoughts
A one-line idea, a spec another tool dumped, scattered notes — put any of it in. Whatever the code can answer, it fills in. It only asks you what the code cannot.
/go does not treat every job the same
A small edit does not get the heavy process. Every result is checked. Work that's hard to undo gets split and checked more carefully.
/migration reads the project you already have
This is not a greenfield-only tool. It reads the code you have, lays out the structure and the rules, and confirms only what the code cannot settle.
your-project/ ├── CLAUDE.md What Claude Code reads first ├── AGENTS.md What Codex reads first └── docs/ ├── architecture.md Structure and flow ├── business-rules.md How the product is supposed to behave ├── security.md Security policy ├── standards.md Rules that have to hold ├── operations.md Install, build, deploy ├── contracts.md What we send and receive └── tracking/ Decisions and progress
Questions
When does it run?
Only when you type /ready, /go, or /migration. It does not jump into ordinary chat or other work.
Does it change the setup I already have?
No. It attaches as a plugin on Claude Code or Codex. No extra server, no extra account. Work stays in the folder and git you already use. Remove the plugin and the documents stay in the project.
Do I have to use it for everything?
No. If the change is small and already clear, just do it. Dryforge fits the work where a wrong call is expensive to undo.