Sealevel Guard

Agent-native Solana security infrastructure for evaluating protocol and code risk before agents deploy, integrate, trade, or move capital.

Sealevel Guard is an Anchor-first Solana audit system designed for machine consumers. Instead of producing only a human-readable report, it aims to return a structured risk brief that upstream agents can use to allow, warn, deny, or escalate an action.

[!NOTE] Current status: stage 1, local skill-native workflow. The implemented path is program address or local path -> source resolution -> specialist review -> risk brief.

Why It Exists

Solana agents can already execute.

They can ship code, integrate protocols, route capital, and call payment or data services. What they still lack is a native security judgment layer before they act.

Sealevel Guard is being designed to fill that gap for:

• deployment agents deciding whether code is safe enough to ship,
• integration agents deciding whether a program is safe enough to depend on,
• treasury agents deciding whether capital should flow into a protocol,
• trading agents deciding whether a venue or execution path is too risky.

The goal is to turn Solana security review into a machine-readable decision primitive.

What The Product Is Today

Sealevel Guard currently exists as a local trust gate for Solana agents. The intended operator flow is:

1. install or load the skill into a host runtime such as Claude Code
2. point an agent at a Solana program address or local Anchor path
3. let the local runtime execute the review
4. receive a structured risk brief and report

Today, the product is best understood as:

• an Anchor-first security judgment layer for Solana agents,
• a local review capability that can be invoked from agent runtimes,
• and a machine-readable risk brief generator for agent decisions.

What The Product May Become Later

If the local agent workflow proves out, Sealevel Guard can later expand into a hosted, agent-facing service layer built around:

1. Quick Scan
   • lightweight intake and coarse triage
2. Quoted Audit
   • quoted review jobs priced by codebase complexity and scope
3. Risk Brief
   • structured output for downstream agents and automated decisions

That is future product surface built on top of the current agent workflow.

What The First Release Focuses On

The initial scope is intentionally narrow: Anchor-first Solana audit workflows.

Planned audit skill areas:

• access control,
• PDA integrity,
• account constraints,
• CPI risk,
• token and vault invariants,
• governance and upgradeability risk.

How To Use

Sealevel Guard currently supports one implemented execution path:

program address or local Anchor repo -> source bundle -> specialist review -> risk brief

Prerequisites

• Node.js 20 or newer
• network access to a Solana RPC endpoint and verified-build metadata
• Claude Code if you want to run the product as intended from a local host runtime
• optional: Codex CLI if you want to experiment with a Codex-backed specialist runtime

Quick Start

Install:

npx skills add NewmanXBT/sealevel-guard

Run review:

/sealevel-guard:sealevel-guard-review <program_address_or_local_path>

Outputs

By default Sealevel Guard writes review artifacts under:

• artifacts/reviews/<target>/resolution.json
• artifacts/reviews/<target>/bundle-manifest.json
• artifacts/reviews/<target>/specialist-findings.json
• artifacts/reviews/<target>/judged-risk-brief.json
• artifacts/reviews/<target>/report.md

See How To Use for the full operator flow and runtime notes.

Example Risk Brief

The core artifact is intended to be agent-readable:

{
  "target": "repo_or_program",
  "risk_score": 71,
  "recommendation": "deny",
  "ship_blocker": true,
  "findings": [
    {
      "skill": "cpi-risk",
      "severity": "high",
      "confidence": 0.84,
      "title": "Unvalidated CPI target",
      "evidence": ["programs/vault/src/lib.rs:118"]
    }
  ]
}

What Exists In This Repo Today

This repository currently contains the stage 1 product framing and current agent-facing implementation for the project:

• Narrative
• Positioning
• Product Spec
• How To Use
• Architecture
• Roadmap
• X Article Draft

Positioning

Sealevel Guard is not a generic crypto chatbot, a chain-agnostic Rust scanner, or a thin pay-per-request x402 demo.

Right now it is a local trust gate for Solana agents and operators. Over time, it can become security clearing infrastructure for the Solana agent economy: a system agents call before they deploy code, integrate programs, or route funds.