Safe On-Chain Agent Skill

Production-grade safety middleware and execution environment for autonomous AI agents on Solana.

Install Now View GitHub Repo Full README

Key Features

Simulation-First Execution: Every transaction is dry-run against mainnet state before requesting signatures to prevent failures and save compute fees.
Intelligent Error Recovery: Intercepts raw Solana errors (like 0x1771) and feeds actionable, semantic feedback back to the LLM for autonomous self-healing.
Scoped Permissions & Wallet Hygiene: Enforces zero-trust architecture using session keys, embedded wallets, and granular amount/action/time-bound allowances.
Composability: Designed to drop directly into the Solana AI Kit as safety middleware without rewriting existing logic.

Demo Concept: Self-Healing Jupiter Swap

Witness the power of semantic error recovery. When the agent attempts a swap but the market moves, it doesn't panic.

The Flow

1. Agent: Proposes swapping 10 USDC for SOL.

2. Middleware: Simulates the transaction.

3. Result: Simulation fails with `SlippageToleranceExceeded`.

4. Middleware: Injects semantic context back to the LLM.

5. Agent: Autonomously recalculates route with 0.5% higher slippage.

6. Middleware: Re-simulates successfully, then executes.

How to Install

Install the skill locally or globally using our simple one-liner bash script. This handles cloning the repository and setting up the mandatory `.claude/rules`.

curl -sSL https://raw.githubusercontent.com/Cryptojigi/safe-onchain-agent-skill/main/install.sh | bash

For custom installation options (like project-local setups), download and run the interactive installer:

curl -O https://raw.githubusercontent.com/Cryptojigi/safe-onchain-agent-skill/main/install-custom.sh
chmod +x install-custom.sh
./install-custom.sh