[Audit] LOW: No rate limiting on API endpoints

[NickCalabs]

Severity: LOW

Description

The HTTP server in src/server.ts has no rate limiting on any endpoints. While it's bound to localhost by default, this could still be problematic in some scenarios.

Vulnerable Endpoints

POST /agents/:name/run - No rate limit

app.post("/agents/:name/run", async (c) => {
  // Fire-and-forget agent run
  // No check for duplicate runs or rate limiting
});

POST /tools/call - No rate limit

app.post("/tools/call", async (c) => {
  // Direct tool invocation
  // Could be expensive (shell commands, file operations)
});

POST /agents - No rate limit

app.post("/agents", async (c) => {
  // Agent creation/registration
});

Risk Scenarios

1. Localhost Compromise: If localhost is compromised (malicious browser extension, local malware), attacker can spam requests
2. Accidental DoS: Buggy client code in a loop could spawn thousands of agent runs
3. Resource Exhaustion: Each agent run spawns MCP servers, runs LLM calls, etc.
4. Cost Explosion: Anthropic API calls cost money, unlimited runs = unlimited cost

Current Mitigation

• Server binds to 127.0.0.1 only (not 0.0.0.0)
• No network exposure by default

Impact

• Runaway costs from accidental infinite loops
• System resource exhaustion (memory, CPU, file descriptors)
• Database growth (run logs)
• Difficult to kill runaway processes

Recommendation

Add basic rate limiting with a simple in-memory store:

import { Hono } from "hono";

interface RateLimitEntry {
  count: number;
  resetAt: number;
}

const rateLimits = new Map<string, RateLimitEntry>();

function checkRateLimit(
  key: string, 
  maxRequests: number, 
  windowMs: number
): { allowed: boolean; retryAfter?: number } {
  const now = Date.now();
  const entry = rateLimits.get(key);
  
  if (!entry || now > entry.resetAt) {
    rateLimits.set(key, { count: 1, resetAt: now + windowMs });
    return { allowed: true };
  }
  
  if (entry.count >= maxRequests) {
    return { allowed: false, retryAfter: Math.ceil((entry.resetAt - now) / 1000) };
  }
  
  entry.count++;
  return { allowed: true };
}

// Apply to endpoints:
app.post("/agents/:name/run", async (c) => {
  const limit = checkRateLimit(`run:${c.req.param("name")}`, 10, 60_000); // 10 per minute
  if (!limit.allowed) {
    c.header("Retry-After", String(limit.retryAfter));
    return c.json({ error: "Rate limit exceeded" }, 429);
  }
  // ... rest of handler
});

Or use existing middleware like hono-rate-limiter.

Configuration

Make rate limits configurable in ~/.agentd/config.yaml:

rate_limits:
  agent_runs_per_minute: 10
  tool_calls_per_minute: 100
  agent_creation_per_hour: 5

---

Created by security audit