Greenlight
checking mode
A quality gate for AI agent output

Agents write fast.
Greenlight makes sure they write true.

Support replies, listings and outreach drafted by agents ship with invented refunds, made-up specs and statistics nobody measured. Greenlight checks every draft against its brief, scores it with an independent evaluator panel, rewrites what fails, and only lets passing work through.

3 roundsmaximum rewrite loop
0–100panel score per round
Every claimtraced back to the brief
The work to gate
Instant check runs locally as you type—
Sentences with figures or promises get traced against the brief.
Ship threshold80
1
Pre-check
SERV Reasoning
2
Evaluation
Reppo panel
3
Rewrite
SERV Reasoning
4
Verdict
ship or block
0/100
waiting to run
Pick a case on the left and run the gate. The score journey appears here, round by round — click any point or round tab to inspect it.

No rounds yet. When the gate runs, each round's verdict, rubric, flagged claims and the rewrite diff land here for inspection.

Audit trail
  • Gate idle. Every check, verdict and rewrite is recorded here, in order.