This is the judge-facing runbook: problem first, evidence second, architecture close.
- Problem: AI agents can reason, but they cannot safely pay each other per request.
- Approach: x402 paywalled endpoints on Stellar + Claude orchestration + strict budget guardrails.
- Proof: clickable on-chain transaction hashes during the demo.
- Orchestrator intro (10s): show task box, budget slider, wallet balances.
- Agent Registry (15s): show all 4 agents, model and price tiers.
- API Status (20s): show premium endpoint list and payment flow surface.
- Main run (60-80s): run task with 0.15 budget and let SSE stream events.
- Proof click 1 (12s): click TX from result chips and show Stellar Expert details.
- Budget guardrail run (35-45s): run second task with 0.02 budget.
- Proof click 2 (12s): open Transactions page, click first hash, show it resolves.
- Close (10s): return to API Status then Transactions to reinforce architecture + proof.
Note:
- The recorder now keeps proof clicks in the same browser tab (no pop-out tab switching).
- Narration text file:
DEMO_VOICEOVER.txt - The automation uses this file to generate narration audio.
- Edit wording there if you want your own tone before rendering.
npm run record:narratedThis performs:
- local server start
- strict preflight (
x402+ on-chain verify) - website-only scripted recording
- narration generation
- final MP4 mux
Final output path is printed and also saved in recordings/latest-narrated.txt.
npm run preflightreachesREADY.- Main run reports real
x402settlement events. - At least one transaction hash is clicked and shown on Stellar Expert.
- Second run demonstrates budget-limited behavior.
- No debug detours in the final submitted video.