CLAWT

Verifiable AI Agent Platform on EigenCompute

CLAWT deploys personal AI agents into Trusted Execution Environments (TEEs) with cryptographically signed responses. Users sign in with Ethereum wallets (MetaMask + SIWE), deploy agents, and submit natural-language tasks.

Architecture

frontend/     Next.js web UI (Vercel)
backend/      Express orchestration API (VPS)
agent/        TEE agent with HD wallet (EigenCompute)
skills/       39 executable skill definitions
graph/        12-node capability navigation graph
shared/       Shared modules (paytoll-client.js)
scripts/      Registry maintenance scripts

Skills Registry

The skills/ directory contains 39 skills organized by domain:

• DeFi (17): Aave lending, DEX swaps, token prices, pool discovery
• Social (5): Twitter search, post, user/tweet lookup
• AI (4): LLM access (Anthropic, Google, OpenAI), image generation
• Identity (6): ENS names, wallet validation, token balances
• Text Tools (4): Summarize, translate, humanize
• Search (1): Tavily AI-optimized web search
• Tools (2): GitHub CLI, knowledge graphs

Each skill has a SKILL.md with YAML frontmatter defining:

• name, description, version, author
• requires_env — environment variables needed
• execution — shell commands to run

PayToll Skills

Skills prefixed with paytoll- use x402 micropayments for API access. They share a common paytoll-client.js module.

Skill Graph

The graph/ directory contains navigation nodes that help the agent discover relevant skills:

• index.md — Entry point listing all domains
• Domain MOCs: defi.md, social.md, ai.md, identity.md, text-tools.md, search.md, tools.md
• Sub-MOCs: aave-lending.md, dex-trading.md, ens-management.md
• Concept: x402-payments.md

Nodes use [[wikilinks]] to reference other nodes and skills.

Registry JSON

registry.json is auto-generated from skill metadata:

python scripts/generate-registry.py

This produces a machine-readable index with skill IDs, descriptions, content hashes, and environment requirements.

Validation

Validate graph integrity (wikilinks, reachability, orphans):

python scripts/validate-graph.py

CI Workflow

.github/workflows/update-registry.yml automatically regenerates registry.json when skills change on the main branch.

Adding a New Skill

1. Create skills/<skill-id>/SKILL.md with frontmatter
2. Add execution script (run.js, scripts/*.py, etc.)
3. Link from appropriate graph node using [[skill-id]]
4. Run python scripts/validate-graph.py to verify
5. Commit — CI regenerates registry.json

Environment Variables

Skills declare required env vars in their requires_env field. The agent filters available skills based on which vars are set.

License

MIT