Reflow docs
Reflow proves your AI’s PR works in a real browser before you merge it. These guides cover the flow contract, deterministic browser language, run surfaces, and the evidence Reflow returns.
Start here
Section titled “Start here”- Getting started — activate a team, connect an agent, and run the first flow.
Writing flows
Section titled “Writing flows”- The flow format — structure a markdown flow with frontmatter, intent, and executable steps.
- RFL — the Reflow Language — use the deterministic statements that drive the browser.
- Guiding the agent — write intent and invariants for the parts of a flow that need a model.
- Playground: record and debug — record real interactions into RFL and replay a document line by line.
Running
Section titled “Running”- GitHub Action — run flows against pull-request previews and report the result in GitHub.
- Mailbox testing and test accounts — test signup, magic-link, OTP, and seeded-account journeys.
- Browser self-targets — start and test trusted code inside the browser task image.
- Custom images — push and run team container images for the enterprise harness.
Autoheal
Section titled “Autoheal”- Autoheal — understand which selector failures can be repaired and how proposed changes are persisted for review.
Visual testing
Section titled “Visual testing”- Visual testing — add explicit screenshot checkpoints, compare baselines, and review diffs.
Agents & MCP
Section titled “Agents & MCP”- MCP reference — connect a coding agent over OAuth and use Reflow’s flow, run, session, mailbox, and container tools.
API & tokens
Section titled “API & tokens”- API and personal access tokens — authenticate scripts and CI, start runs, inspect evidence, and manage flows over HTTP.