OpenCode Refactorer
Staged local refactoring
Works through a refactor in your own checkout in stages that each leave the tests green. You can stop, review or ship at any point rather than facing one enormous diff at the end. Behaviour is preserved exactly — if it finds a bug along the way it notes it separately instead of fixing it inside the refactor, where it would be invisible in review.
Add to your OasisConnects with
No app connections are included in this template. Check its foundation and setup requirements before your first run.
- +Custom toolConnect any other tool from thousands of available integrations.
How it runs
- Foundation
OpenCode
- Reasoning
- Medium effort
- Per run
- Up to 50 turns
- Memory
- Learns across runs
- Browsing
- Off
- Category
- Engineering
- Opencode
- Local
- Refactoring
Frequently asked questions
How does OpenCode Refactorer keep a local refactor reviewable?
It works in stages, with tests green after each stage, so you can stop, review or ship without facing one enormous diff.
Will OpenCode Refactorer change behavior while cleaning up code?
It preserves behavior exactly. If it spots a bug, it records that separately rather than hiding a behavior fix inside the refactor.
Can I pause OpenCode Refactorer after one refactor stage?
Yes. Each stage leaves the tests green and can be stopped or reviewed before later stages continue.
Where does OpenCode Refactorer perform its code changes?
The catalog describes staged work in your own checkout, with no connected toolkits listed.
Agents like this one
Codex Bug Fixer
Give it a failing test or a stack trace and it works the problem in your local checkout. It reproduces the failure first and reads outward from where it actually breaks — never starting from a theory, because a fix for a bug you never saw fail is a guess. It fixes the cause rather than the symptom, and hands back a minimal diff with the suite green.
Try this Agent
Oasis PlatformEngineeringRefactor Planner
Maps out a refactor before anyone touches the code. It traces every caller, every test and every implicit contract the type system doesn't capture, then produces a staged plan where each stage leaves the system working. It names what could break at each stage — and says plainly when stable, rarely-touched code isn't worth refactoring at all.
Try this AgentOasis PlatformEngineeringKilo Feature Scaffolder
Lays down the boring skeleton of a new feature in your repo, in the shape this codebase already uses. It studies two or three comparable features first and copies their structure, naming and test layout, because consistency with the surrounding code matters more than any preferred structure. You start from a running stub with the genuinely undecided logic clearly marked.
Try this AgentOasis PlatformEngineeringRelease Scribe
Reads the commits since your last tag and writes the changelog. Each user-visible change is translated into what it means for someone using the software, and internal refactors, dependency bumps and test-only changes are dropped. When a commit message is unclear it reads the diff rather than paraphrasing the subject line.
Try this AgentOasis PlatformEngineeringTest Backfiller
Finds the code paths your suite never touches and writes tests that would genuinely catch a regression. Coverage gaps are ranked by blast radius, so money, auth and data writes outrank a well-covered utility missing one line. It matches your existing test style exactly and runs everything it writes — it never hands over a test it hasn't seen pass.
Try this AgentOasis PlatformEngineeringHermes Repo Agent
Runs long, self-directed jobs across your local repo — a sweep over many files, a dependency upgrade, a cleanup. It scopes the work and gets that agreed first, since a long job with the wrong definition is the most expensive failure here. It reports progress in batches as it goes, because silence for two hundred file edits isn't acceptable output.
Try this AgentOasis PlatformEngineeringBug Report Filer
Turns a customer complaint into a ticket an engineer can act on. It extracts what the user did, what happened, the environment and how many customers are affected, and searches existing tickets and error tracking for duplicates before filing anything. Where a repro step is missing it says so rather than inventing a plausible one, since a fabricated step wastes an afternoon.
+2
Try this AgentOasis PlatformEngineeringDevin Backlog Burner
Points Devin's cloud sessions at the small, well-specified tickets nobody gets to, and returns pull requests you can review. It refuses tickets that aren't ready — open design questions, unclear owners, no way to verify success — and says why, because choosing the right tickets is most of the value. Changes stay scoped, since drive-by refactors make an autonomous PR unreviewable.
Try this AgentOasis PlatformEngineeringDevin CLI Task Runner
Hands a scoped, well-specified ticket to Devin in your local environment and keeps the loop tight. It restates the task as acceptance criteria before starting, which is the highest-value step given autonomous agents fail most often on under-specified work. It stays inside the stated scope and notes unrelated problems rather than expanding the change.
Try this AgentOasis PlatformEngineeringPi Dev Assistant
A general hand in your local checkout. It defaults to reading and explaining, makes small changes when asked, and shows the diff rather than describing it. When a request is ambiguous it asks instead of picking, and it never runs destructive git operations or touches files outside the repo it was pointed at.
Try this AgentOasis PlatformEngineering
Put OpenCode Refactorer to work today
Add it to your Oasis, review its setup, and hand it the first task.