Skip to content

hackathon-demo-guide.md

Latest commit

 

History

History
187 lines (133 loc) · 4.77 KB

hackathon-demo-guide.md

File metadata and controls

187 lines (133 loc) · 4.77 KB

Hackathon Demo Guide

This guide shows how to demo LeviathanMatrix AEP Open Core quickly.

1. Install

python3 -m venv .venv
. .venv/bin/activate
pip install -r requirements.txt

2. Run Tests

pytest -q -p no:cacheprovider

Expected result:

all tests passed

3. Run A Governed Action

python scripts/aep_cli.py run-text \
  --text "buy 1 USDC of SOL" \
  --agent-id demo-agent

What to point out:

  • the agent did not execute directly
  • AEP compiled the request first
  • AEP created an authorization case
  • AEP issued an Execution Pass
  • AEP wrapped the notional in a Capital Capsule
  • execution produced a receipt
  • review produced a final status

4. Show The Agent Runtime Demo Videos

The README includes two YouTube demos:

Use them to show that AEP can sit between an external agent runtime and the execution path, not only behind a standalone CLI.

5. Inspect The Case

Cases are written under:

artifacts/cases/

Open the generated case and point to:

  • producer
  • spec_id
  • request
  • intent
  • policy_output
  • authorization.issuance
  • authorization.decision.capital_capsule
  • execution
  • receipt
  • review

6. Export The Execution Claim

python scripts/aep_cli.py export-claim \
  --case-id <case_id>

The exported claim is the shareable summary of the governed action.

7. Demo A Denial

python scripts/aep_cli.py run-text \
  --text "buy 1000000 USDC of SOL" \
  --agent-id demo-agent

What to point out:

  • the request compiles
  • policy denies it
  • no capsule is issued
  • execution is blocked
  • review fails the case

This demonstrates fail-closed behavior.

8. Show The Solana Devnet Proof Anchor

AEP Open Core includes a minimal Anchor program for recording the execution lifecycle as a Solana Devnet proof anchor.

Program ID:

5LY2YsVpAhES2nq9TT7iQn4gGAy8vdb4nkE3XyQzMw4q

Confirmed Devnet transaction:

https://explorer.solana.com/tx/57c51zE8QZ3vrZ4My5Jgp1ssFAhEP8vRa72X8apg8QKxE3VSCY1J1gQCU8SZFU83gu2XTryWopGZvTRd4P6SV5Qx?cluster=devnet

What to point out:

  • the AEP decision happens off-chain;
  • the proof of the AEP lifecycle is anchored on Solana Devnet;
  • the chain account stores hashes, not raw strategy or private data;
  • judges can inspect the transaction in Solana Explorer.

9. Demo The Core Thesis

The one-line demo:

We are not showing that an agent can trade. Everybody can do that.
We are showing that an agent cannot trade unless its action becomes policy-bound, time-limited, notional-limited, and reviewable.

10. Judge Talking Points

Use these:

  • AEP is a deterministic execution kernel, not a prompt trick.
  • Execution Pass turns approval into a cryptographic capability object.
  • Capital Capsule turns broad capital access into a finite envelope.
  • Capability hash blocks scope mutation between authorization and execution.
  • Constitution rules make policy portable across agent runtimes.
  • Solana is the natural first target because high-speed agent finance needs high-speed execution policy.
  • The Devnet proof anchor gives the local AEP lifecycle a public Solana verification surface.

11. Thirty-Second Pitch

Most agent demos prove that an AI can call a tool.
LeviathanMatrix AEP proves something more important:
an AI cannot call a value-moving tool unless its action becomes a scoped, expiring, policy-bound execution object.

We turn agent execution into a deterministic lifecycle:
intent, constitution, risk score, execution pass, capital capsule, receipt, review.

This is the control plane agents need before they can safely operate in Solana-speed finance.

12. What To Show On Screen

Open these files during the demo:

  • aep/kernel.py: the lifecycle orchestration
  • programs/aep-proof-anchor/src/lib.rs: the minimal Solana proof anchor
  • policy_engine/engine.py: deterministic policy scoring and reason codes
  • aep/issuance.py: Execution Pass and capability hash
  • aep/capsule.py: Capital Capsule lifecycle and notional guard
  • aep/capsule_pricing.py: capsule pressure model
  • docs/algorithm-notes.md: algorithm explanation
  • docs/design-rationale.md: why the architecture exists

13. Solana Framing

Say this clearly:

Solana made machine-speed capital actions practical.
AEP makes machine-speed capital actions governable.

Then explain:

  • low fees make repeated agent actions viable
  • fast confirmation means weak policy fails faster
  • high throughput needs automated execution boundaries
  • agentic payment flows need policy objects, not just wallet signatures
  • long-tail assets make scoped authority safer than broad approvals