Security Hardening

This guide focuses on secure defaults and common mistakes in real deployments.

Baseline Security Profile

const agent = await ReactiveAgents.create()
  .withProvider("anthropic")
  .withGuardrails()
  .withBehavioralContracts({
    deniedTools: ["code-execute"],
    maxIterations: 10,
  })
  .withIdentity()
  .withAudit()
  .withKillSwitch()
  .build();

Guardrails and Contracts

• Keep .withGuardrails() enabled for all user-facing entry points.
• Use behavioral contracts to constrain tool access by policy, not by prompt text.
• Prefer allowlists/denylists for tools in high-trust environments.

MCP Hardening

• Prefer streamable-http with explicit auth headers for remote servers.
• For stdio/Docker, keep containers minimal and ephemeral (--rm).
• Separate host CLI env from container env; only pass required secrets.
• Always use deterministic cleanup (await using or runOnce()).

Secret Management

• Never embed secrets in docs examples committed to source control.
• Keep per-server credentials in environment variables.
• Pass only minimal auth headers per MCP server.

Tool Risk Reduction

• Disable code-execute unless strictly required.
• Require approval for state-changing tools where possible.
• Isolate file-write scope to approved directories.

Identity and Audit

• Enable .withIdentity() for RBAC and delegation controls.
• Enable .withAudit() to preserve action history for investigations.
• Subscribe to security-relevant events and alert in near real-time.

Incident Readiness

• Wire kill switch activation into on-call procedures.
• Add alerts for repeated guardrail violations and budget exhaustion.
• Keep a tested rollback path for model/provider configuration changes.

Deployment Checklist

• Guardrails enabled
• Behavioral contracts defined
• Kill switch enabled
• MCP transports authenticated and scoped
• Agent disposal guaranteed
• Audit logging enabled
• Budget limits configured
• On-call alerts wired