Verify, approve, run, re-check

Outcome

You know every step between a claim and a trusted answer, and who takes each one.

Does not do

No step can be skipped: nothing runs before a person approves the plan.

Every check follows the same loop, whether it starts in the console, the CLI, a CI job or an AI assistant. The names differ by channel; the steps and the rules do not.

1. Verify. Send the deployed URL and one sentence. Vraelis writes a plan and returns an approval link. Nothing has run and nothing is charged.
2. Approve. A person opens the link and approves the plan with one click. An API key cannot approve a plan, so a script or an assistant never signs off on its own check.
3. Run. The approved plan runs in a real browser on the live app, exactly as approved. The answer is Verified, Failed or Blocked, with the evidence.
4. Re-check. After a Failed, deploy the fix and run the same approved plan again. No new approval is needed within 24 hours of the approval, for up to 10 re-checks, on the same site.

The same loop in each channel

• Console: write the claim, approve under Review, read the result, and run it again from the result page.
• CLI: (retired command) prints the approval link, waits for the approval, runs, and exits 0, 1 or 2. (retired command) runs the same plan after a fix.
• API: POST /v1/verifications returns review_required and an approve_url. After the approval, resubmit with reviewed_plan_id to start the run, read GET /v1/verifications/{id}, and re-check with POST /v1/verifications/{id}/recheck.
• AI assistants over MCP: vraelis_verify, then vraelis_status while the person approves and the run finishes, then vraelis_recheck after a fix.