Production project
Verifiable Autonomous AI Agent Project
Design and verify an autonomous treasury management AI agent that monitors oracle metrics, runs policy engine checks, simulates transactions, requests operator approval above spending thresholds, and executes non-custodial rebalances via an ERC-4337/EIP-7702 smart account with ZK execution verification.
Project evidence
0/3 milestones · rubric 0% · pass 70%
Milestone 1: Policy Engine & Safety Gates
A policy engine specification defining contract allowlists, spending limits, and human gate triggers.
Acceptance criteria
- Policy engine intercepts tool calls before transaction generation.
- Spending limits and contract allowlists are explicitly checked.
Use: context · drivers · decision · alternatives · consequences · risks · validation · rollback · evidence
Milestone 2: ERC-4337 Smart Account Integration
An ERC-4337 UserOperation builder and Paymaster gas sponsorship configuration.
Acceptance criteria
- Transactions execute via UserOperations through EntryPoint contract.
- Session keys constrain sub-agent transaction permissions.
Use: context · drivers · decision · alternatives · consequences · risks · validation · rollback · evidence
Milestone 3: ZK Proof Verification & Finality
An on-chain ZK verifier integration validating model execution receipts before settlement.
Acceptance criteria
- zkVM proof receipts are verified by smart contract before payout.
- Model weight hashes match the on-chain provenance registry.
Use: context · drivers · decision · alternatives · consequences · risks · validation · rollback · evidence
Criterion-level self-assessment
Project rubric
Architecture coherence
Architecture coherence for AI Architecture Decision Workbook is explicit, testable, and connected to submitted evidence. · weight 20%
Current level: Missing or contradicted by the submitted evidence.
Implementation evidence
Implementation evidence for AI Architecture Decision Workbook is explicit, testable, and connected to submitted evidence. · weight 15%
Current level: Missing or contradicted by the submitted evidence.
Evaluation quality
Evaluation quality for AI Architecture Decision Workbook is explicit, testable, and connected to submitted evidence. · weight 20%
Current level: Missing or contradicted by the submitted evidence.
Operational readiness
Operational readiness for AI Architecture Decision Workbook is explicit, testable, and connected to submitted evidence. · weight 15%
Current level: Missing or contradicted by the submitted evidence.
Security and privacy
Security and privacy for AI Architecture Decision Workbook is explicit, testable, and connected to submitted evidence. · weight 20%
Current level: Missing or contradicted by the submitted evidence.
Decision communication
Decision communication for AI Architecture Decision Workbook is explicit, testable, and connected to submitted evidence. · weight 10%
Current level: Missing or contradicted by the submitted evidence.
Portable evidence
Attach project records
Text, HTTPS links, repository paths, and imported Markdown/JSON/text are stored locally and included in exports.
Project review checklist
Unresolved risks
One risk per line. A passing score does not erase open risk.