Execution agent
Builds an accepted spec or approved fix and opens a pull request.
The Execution agent is the one that actually writes code. It takes an accepted spec — or an approved issue fix — implements it on its own branch, runs your tests, and opens a pull request. There's no interview; it works autonomously.
How it starts
You don't start Execution yourself. A sweep runs every couple of minutes and picks up any spec you've accepted or any issue you've approved for a fix, then begins the build. The moment it starts, the spec or issue locks so it can't be edited mid-build.
How it works
It first breaks the work into a task checklist you can watch, then implements task by task, running your project's test commands as it goes. When everything passes it commits, pushes its branch, and opens a pull request.
If it needs a scope decision it pauses and asks — you reply and it carries on. You can do the same with a finished build: under Keep working on it, tell Fragua what to tweak, attach any new files, and it confirms the plan with you before changing code.
Accepting the result
When it's done the spec or issue moves to implemented, with either a ready or a draft pull request depending on whether the tests passed and the work fit the budget. If something blocked it — a failed push, say — it stops at implementation failed with the details you need to recover by hand.
You review and merge the pull request the same as any other change, then mark the spec or issue shipped. Need a clean slate instead? Start over clears the branch and reshapes from scratch.
How a build ends
Tests passed and the work fit — a pull request ready to review and merge.
CI wasn't green or the budget ran out — a draft PR you can pick up and finish.
Something blocked it, like a failed push — it stops with the details you need to recover.
Whatever the outcome, you review and merge the pull request yourself, then mark the spec or issue shipped.
Execution runs off the two surfaces you drive — specs and issues.