Ways to work
Patchrooms is one feedback layer: reviewers comment in a live app, the context reaches your coding agent. How the agent acts on it is the workflow. Pick by how much autonomy is safe; move up the ladder as trust grows.
-
Review
You point at elements in a running app and leave feedback. The agent pulls it, proposes a fix, and shows you the diff. You approve. The default when you are actively building.
When: Active development, anything where you want eyes on every change.
-
Testing / QA
Use the open reports as a QA checklist. The agent reproduces each on its page, confirms whether it is still broken, and marks it triaged or closed. It reports. It does not patch.
When: Pre-release passes, regression sweeps, signing off a build.
-
Auto-fix loop
A timed loop (e.g. every 5 minutes) pulls new feedback, fixes it, and closes it unattended. Best for a steady stream of small, low-risk reports where a bad change is cheap to revert.
When: High-volume small fixes on a repo with git + CI.
-
Propose loop
The same loop, but fixes land on a branch and reports stay in-progress with the diff surfaced. A human approves batches instead of every single change.
When: Clearing a backlog without losing review.
-
Spec-writing
When a report is too big or vague to fix on a guess, the agent turns it into a short spec (what is asked, affected screens, acceptance check) and leaves the report triaged. You review the spec before any code is written.
When: Ambiguous or large feedback, product decisions.
-
Accumulate & ask
When the agent keeps hitting unknowns, it collects the open questions across reports and surfaces them in one batch, instead of interrupting you once per report.
When: Early on a new codebase, lots of missing context.