Skip to main content
Run the Wizard in your application repository to carry a Straddle integration through planning, implementation, sandbox testing, and production readiness review. Complete the quickstart prerequisites first.

​
Start the program

Choose how the Wizard works with your coding agent.
Start the guided run and select your installed, signed-in command-line client:
The Wizard opens Claude Code, Codex, or cursor-agent in the terminal for the program. The agent continues through the listed steps and pauses when it needs your decision or approval.
Confirm the project context and integration choices. The program runs Setup, Plan, Integrate, Test, and Go Live readiness. When the repository uses another payment provider, Migrate runs after Plan and writes its own plan and report.

​
Review decisions as the agent works

Read straddle-integration-plan.md before approving the proposed application changes and test scenarios. The agent records your approval against that version of the plan. An edit to the plan requires another review and approval. Review sandbox request previews separately from code changes. A preview identifies the environment, acting account, operation, and payload. The skill asks for approval before it sends the write. On a resumed run, it presents sandbox writes for approval again. Claude Code Auto runs show a checklist in the status line. Codex Auto runs show it on a local page whose address the Wizard prints. In Cursor and Manual runs, the Wizard reads the files after the session or when you return.

​
Read the results

The following files record the work at the repository root. Reports after Plan identify the approved plan they used. The Wizard counts them as finished only when their completion status and plan hash match. Use the test evidence’s live-observed rows to identify behavior verified against Straddle sandbox. Inspect progress or open the recorded session log:
status shows the saved run and local configuration. log creates a local HTML page from recorded session events and available transcripts. Review the report and Verify before merging checklist with the code changes.

​
Resume from the first unfinished step

Continue from the same repository after resolving the finding or question that stopped the run:
The Wizard reads the plan and reports to choose the next step. In Auto, it reopens the agent session when a saved session ID is available, or starts a session at that step. In Manual, it prints the next message to paste. Change the mode with resume --mode auto or resume --mode manual. A partial or blocked report keeps its step unfinished. Changing an approved plan also makes reports for the earlier plan unfinished. Fix the reported issue and resume; the work already written stays in the repository.

​
Finish after sandbox testing

When Test is complete and Go Live is the only remaining step, the Wizard offers Finish here (skip Go Live) if a Go Live report has not already been written for that plan. This records the skipped review in the run. When you’re ready to assess production readiness, run:
Read the readiness report and resolve its gaps before moving the integration to production.