Security Model

This SDK executes real financial operations. Treat security as a first-class runtime feature, not an add-on.

Stendar is infrastructure software only and does not provide investment, legal, accounting, tax, brokerage, or custodial services.

Core Guardrails

• Fail-closed policies: policy evaluation errors deny execution.
• Simulation-first flow: runtime simulates before signing or broadcasting.
• Wallet boundary: signing is delegated to wallet adapters; runtime does not custody funds.
• Auditability: AuditLogger writes hash-chained JSONL entries.
• Bridge safety: bridge paths avoid implicit replay and surface explicit status.
• No committed secrets: .env, private keys, wallet JSON, bearer tokens, and provider keys must stay outside git.
• Controlled agent surface: external agents should use MCP control or full mode when they need proposal, approval, and audit semantics.

SDK Audit Logging

The core SDK AuditLogger can be used by strategy/runtime integrations.

import { AuditLogger } from '@stendar/core';

const audit = new AuditLogger({
  enabled: true,
});

audit.log({
  actor: { type: 'agent', id: 'strategy-engine' },
  action: 'execute_swap',
  resource: 'solana-mainnet',
  outcome: 'success',
  details: { strategy: 'dca-sol' },
});

const verification = await audit.verify();
console.log(verification.valid);

Runtime Policy Behavior

• Attach PolicyRule[] to DeFiRuntime to enforce spending, rate, protocol, token, slippage, drawdown, exposure, and time-window constraints.
• Unknown or malformed Policy v2 evaluations must not auto-allow money-affecting execution.
• Use dry-run and preview flows before live execution.
• Prefer explicit approvals for high-notional or high-risk actions.
• Public LangChain and Vercel AI adapters expose direct in-process runtime tools. Use them only from applications that also own policy attachment, approval UX, and audit logging.
• MCP control and full modes require local-runtime bearer authentication and disable direct runtime execution tools unless explicitly overridden.

Provider And Wallet Boundary

• Treat RPC/provider responses as untrusted until validated.
• Keep wallet private keys and signing credentials out of code, logs, docs, and issue reports.
• Use environment variables or secret managers for local development credentials.
• Rotate provider credentials if logs, screenshots, or artifacts accidentally include token-like values.

Threat Model Areas

Key categories covered in the repository threat model:

• malicious provider/plugin behavior
• compromised RPC/provider responses
• agent prompt-injection misuse
• policy bypass attempts
• replay or duplicate transaction submission
• wallet secret leakage
• package supply-chain compromise

See the repository-level Threat Model for the public SDK-focused model.