Level5 - Your Agent Can Pay For Itself

Budget Management for AI Agents — USDC billing gateway. Deposit USDC on Solana; pay for LLM compute per token.

Security Scan

The skill's instructions, network calls, and lack of installs/credentials are consistent with its stated purpose as a Solana/USDC billing proxy, but it relies on a single sensitive dashboard URL (containing an API token) so users should be careful how they store/share that link and verify on-chain details before depositing funds.

Purpose & Capability

The name/description (USDC on Solana, per-token billing proxy) matches the SKILL.md: registration, dashboard, pricing, and service discovery all relate to billing. The skill does not request unrelated credentials or binaries.

Instruction Scope

All runtime instructions are limited to registering with api.level5.cloud, showing the returned dashboard URL, optionally checking a remote SKILL.md on user request, and pointing users to deposit funds. The skill explicitly instructs not to act without user confirmation. The notable data-handling detail: the dashboard URL contains the API token and the skill instructs agents to present and bookmark it — this is a sensitive bearer token and a potential leakage risk if handled carelessly.

Install Mechanism

Instruction-only skill with no install spec, no downloads, and no code files — minimal disk/system footprint and low install risk.

Credentials

The skill requires no pre-existing environment variables or credentials. It receives an API token from the service at registration; that token appears to be the sole credential and is communicated via the dashboard URL. This is proportionate to the billing-purpose but creates a sensitive single-token-in-URL workflow that users must protect.

Persistence & Privilege

always is false and the skill explicitly states it is passive until user activation. It does not request persistent system privileges or attempt to modify other skills/configs.

Assessment

This skill appears to do what it says (registers a billing proxy, returns a dashboard URL containing an API token, and instructs how to fund and use Level5). Before using it: 1) Only allow the agent to register after you explicitly confirm — the skill itself instructs to ask first. 2) Treat the dashboard URL as a secret bearer token: do not paste it in public chat, shared docs, or logs. 3) Verify the Solana contract_address on-chain (and the service's docs) before depositing USDC — confirm contract/source code where possible. 4) Confirm the HTTPS endpoint and the Level5 service reputation; check privacy/policy and whether funds can be recovered. 5) If you want tighter security, prefer using a wallet or custody arrangement you control rather than sharing a single-URL token, and avoid letting autonomous agents manage deposits. If you want more assurance, provide the skill's maintainers, repo, or smart contract links so you can independently audit them; absence of a canonical code repo is a remaining risk.

Level5: Budget Management for AI Agents

Computational sovereignty is not a feature. It is a right.

Level5 is a transparent billing proxy between AI agents and LLM providers (OpenAI, Anthropic, OpenRouter). Agents deposit USDC to a Solana smart contract; Level5 handles per-token billing with real-time balance tracking.

Streaming support: Both streaming (SSE) and non-streaming requests are fully supported. Claude Code and other clients that default to stream: true work out of the box.

---

When to Activate

This skill is passive until the user explicitly requests it. Do not register, write files, or make network requests on skill load.

Activate when:

• The user asks to set up Level5, connect a billing account, or use Level5 as a proxy
• The user asks you to register for Level5
• A request fails with 402 Payment Required and you ask the user if they'd like to fund their account

Always ask before acting. Before registering, confirm with the user:

"I'll create a Level5 billing account at api.level5.cloud. This will route your LLM traffic through Level5 for per-token billing. I'll give you a dashboard URL to save — that URL is your account. Shall I proceed?"

Only proceed after the user confirms.

Registration

import requests


def register_level5() -> dict:
    """Register with Level5 and return credentials.

    Only call this after the user has explicitly confirmed they want to proceed.
    No files are written. The dashboard URL is displayed for the user to save.
    """
    reg = requests.post("https://api.level5.cloud/v1/register", timeout=10).json()
    api_token = reg["api_token"]
    dashboard_url = f"https://level5.cloud/dashboard/{api_token}"

    print(
        f"\n{'─' * 60}\n"
        f"  Level5 account created.\n\n"
        f"  Your dashboard URL — save this somewhere safe:\n\n"
        f"    {dashboard_url}\n\n"
        f"  This URL is your account portal. It contains your API token\n"
        f"  and lets you deposit USDC, view your balance, and get SDK\n"
        f"  configuration instructions.\n\n"
        f"  ⚠ There is no account recovery. If you lose this URL,\n"
        f"  you will need to register again.\n"
        f"{'─' * 60}\n"
    )

    return {"api_token": api_token, "dashboard_url": dashboard_url}

---

Version Check

This skill is versioned (semver, frontmatter version field). The canonical source is https://level5.cloud/SKILL.md.

Never auto-update. Overwriting skill files silently is a security risk — a compromised remote could change your agent's instructions. Version checks are opt-in: only run when the user explicitly asks (e.g. "check for Level5 skill updates").

import re
import requests
from pathlib import Path

SKILL_URL = "https://level5.cloud/SKILL.md"


def check_skill_version(local_path: Path) -> None:
    """Fetch remote SKILL.md version and alert if a newer version exists.

    Never overwrites the local file. Call only when the user explicitly asks
    to check for updates — do not run automatically on startup or on a schedule.
    """
    try:
        remote_text = requests.get(SKILL_URL, timeout=10).text
    except Exception:
        return

    def parse_version(text: str) -> str:
        m = re.search(r"^version:\s*(\S+)", text, re.MULTILINE)
        return m.group(1) if m else "0.0.0"

    from packaging.version import Version
    local_version = parse_version(local_path.read_text())
    remote_version = parse_version(remote_text)

    if Version(remote_version) > Version(local_version):
        print(
            f"\n[Level5] A new skill version is available: "
            f"{local_version} → {remote_version}\n"
            f"  Ask me to upgrade the Level5 skill to get the latest "
            f"models, pricing, and features.\n"
        )
    else:
        print(f"\n[Level5] Skill is up to date (v{local_version}).\n")

---

Service Discovery

Machine-readable service descriptor:

https://level5.cloud/.well-known/agent-service.json

Agent frameworks and orchestrators can fetch this to auto-configure Level5 without manual setup.

---

Three-Step Onboarding

Step 1: Register

curl -X POST https://api.level5.cloud/v1/register

Response:

{
  "api_token": "abc-123-def-456",
  "deposit_code": "A1B2C3D4E5F6A7B8",
  "status": "pending_deposit",
  "instructions": {
    "contract_address": "<sovereign_contract_address>",
    "dashboard_url": "https://level5.cloud/dashboard/abc-123-def-456"
  }
}

Step 2: Fund and Configure

Visit your dashboard URL to deposit USDC and get SDK configuration instructions for Claude Code, Codex, OpenCode, and Cursor. All setup steps are on the dashboard.

https://level5.cloud/dashboard/{YOUR_API_TOKEN}

---

Dashboard

Once registered, your dashboard is available at:

https://level5.cloud/dashboard/{YOUR_API_TOKEN}

The dashboard shows your USDC balance, deposit and usage history, and agent connection instructions for Claude Code, Codex, OpenCode, and Cursor.

Bookmark it. This URL is the only way to access your dashboard without re-registering.

For agents: If a user asks how to find their dashboard, construct the URL from their api_token and provide it directly:

dashboard_url = f"https://level5.cloud/dashboard/{api_token}"
print(f"Your dashboard: {dashboard_url}")

---

API Reference

Method	Path	Auth	Description
GET	/health	None	Service health — returns component status
GET	/v1/pricing	None	Current model pricing
POST	/v1/register	None	Register new agent, get API token + deposit code
GET	/proxy/{api_token}/balance	Token in path	Check USDC balance
GET	/proxy/{api_token}/transactions	Token in path	Transaction history
POST	/proxy/{api_token}/v1/chat/completions	Token in path	OpenAI-format proxy
POST	/proxy/{api_token}/v1/messages	Token in path	Anthropic-format proxy

---

GET /health

curl https://api.level5.cloud/health

Response (healthy): HTTP 200

{
  "status": "ok",
  "version": "1.0.0",
  "components": {
    "redis": "ok",
    "postgres": "ok"
  }
}

Response (degraded): HTTP 503

{
  "status": "degraded",
  "version": "1.0.0",
  "components": {
    "redis": "ok",
    "postgres": "error"
  }
}

---

POST /v1/register

curl -X POST https://api.level5.cloud/v1/register

Response: HTTP 200

{
  "api_token": "abc-123-def-456",
  "deposit_code": "A1B2C3D4E5F6A7B8",
  "status": "pending_deposit",
  "instructions": {
    "contract_address": "BBAdcqUkg68JXNiPQ1HR1wujfZuayyK3eQTQSYAh6FSW",
    "dashboard_url": "https://level5.cloud/dashboard/abc-123-def-456"
  }
}

Rate limited to 10 requests per minute per IP.

---

GET /v1/pricing

curl https://api.level5.cloud/v1/pricing

Response: HTTP 200

{
  "pricing": {
    "anthropic/claude-sonnet-4-6": {
      "input_per_1m": 3300000,
      "output_per_1m": 16500000,
      "cache_write_per_1m": 4125000,
      "cache_read_per_1m": 330000
    },
    "anthropic/claude-opus-4-6": {
      "input_per_1m": 5500000,
      "output_per_1m": 27500000,
      "cache_write_per_1m": 6875000,
      "cache_read_per_1m": 550000
    }
  },
  "currency": "USDC",
  "denomination": "microunits (6 decimals, 1 USDC = 1_000_000)"
}

---

GET /proxy/{api_token}/balance

curl https://api.level5.cloud/proxy/{YOUR_API_TOKEN}/balance

Response: HTTP 200

{
  "api_token": "abc-123-def-456",
  "usdc_balance": 5000000,
  "is_active": true
}

usdc_balance is in USDC microunits (6 decimals). 5 000 000 = 5.00 USDC.

---

GET /proxy/{api_token}/transactions

curl "https://api.level5.cloud/proxy/{YOUR_API_TOKEN}/transactions?page=1&limit=50"

Response: HTTP 200

{
  "api_token": "abc-123-def-456",
  "page": 1,
  "limit": 50,
  "transactions": [
    {
      "id": 1,
      "type": "DEPOSIT",
      "usdc_amount": 10000000,
      "provider": null,
      "model": null,
      "input_tokens": null,
      "output_tokens": null,
      "cache_write_tokens": null,
      "cache_read_tokens": null,
      "created_at": "2026-02-22T10:00:00+00:00"
    },
    {
      "id": 2,
      "type": "DEBIT",
      "usdc_amount": -330,
      "provider": "anthropic",
      "model": "claude-sonnet-4-6",
      "input_tokens": 15,
      "output_tokens": 25,
      "cache_write_tokens": null,
      "cache_read_tokens": 120,
      "created_at": "2026-02-22T10:01:00+00:00"
    }
  ]
}

---

POST /proxy/{api_token}/v1/messages

Anthropic-compatible. Supports streaming (stream: true) and non-streaming.

curl https://api.level5.cloud/proxy/{YOUR_API_TOKEN}/v1/messages \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4-6",
    "max_tokens": 1024,
    "stream": true,
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Response headers include:

X-Balance-Remaining: 4999670

---

POST /proxy/{api_token}/v1/chat/completions

OpenAI-compatible. Supports streaming and non-streaming.

curl https://api.level5.cloud/proxy/{YOUR_API_TOKEN}/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o",
    "stream": true,
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Response headers include:

X-Balance-Remaining: 4999670

---

X-Balance-Remaining Header

Every proxy response includes X-Balance-Remaining — the agent's current USDC balance in microunits after the request is processed. Agents can use this to track spend without polling the balance endpoint.

response = client.messages.create(...)
remaining = int(response.headers.get("x-balance-remaining", 0))
if remaining < 1_000_000:  # < 1 USDC
    alert("Balance low — please top up")

---

Cache Billing

Anthropic models support prompt caching. Level5 tracks cache write and read tokens separately and bills them at the rates shown in /v1/pricing.

• cache_write_tokens: tokens written to a new cache block (1.25x base input rate)
• cache_read_tokens: tokens served from cache (0.10x base input rate — 10x cheaper)

Both fields are visible in transaction records and factored into X-Balance-Remaining.

---

Integration Examples

Python — Anthropic SDK

import anthropic

client = anthropic.Anthropic(
    base_url="https://api.level5.cloud/proxy/{YOUR_API_TOKEN}",
    api_key="level5",
)

response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Analyze market conditions"}],
)
print(response.content[0].text)

Python — OpenAI SDK

from openai import OpenAI

client = OpenAI(
    base_url="https://api.level5.cloud/proxy/{YOUR_API_TOKEN}/v1",
    api_key="level5",
)

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Analyze market conditions"}],
)
print(response.choices[0].message.content)

After Registration

# User has confirmed setup and saved their dashboard URL.
# Use the api_token returned by register_level5() to configure the SDK.
creds = register_level5()  # call only after user confirms
api_token = creds["api_token"]

import anthropic
client = anthropic.Anthropic(
    base_url=f"https://api.level5.cloud/proxy/{api_token}",
    api_key="level5",
)

---

Error Codes

Code	Meaning	Action
200	Success	Request completed, balance debited
400	Bad Request	Invalid JSON body
401	Unauthorized	Invalid or inactive API token
402	Payment Required	Insufficient USDC balance — deposit more
429	Rate Limited	Too many /v1/register calls — wait 60s
502	Upstream Error	LLM provider error — retry with backoff
503	Service Unavailable	Level5 infrastructure issue — check /health

---

Supported Models

Prices in USDC microunits per 1M tokens (6 decimals, 1 USDC = 1 000 000).

Provider	Model	Input/1M	Output/1M	Cache Read/1M
anthropic	claude-sonnet-4-6	3 300 000	16 500 000	330 000
anthropic	claude-opus-4-6	5 500 000	27 500 000	550 000
anthropic	claude-haiku-4-5-20251001	1 100 000	5 500 000	110 000
openai	gpt-5.2	1 750 000	14 000 000	175 000
openai	gpt-5.3-codex	1 750 000	14 000 000	175 000
openai	gpt-4o	2 500 000	10 000 000	1 250 000
openai	o3	2 000 000	8 000 000	500 000
openrouter	qwen3-max	1 320 000	6 600 000	264 000
openrouter	qwen3-coder-plus	1 100 000	5 500 000	220 000
openrouter	glm-5	1 100 000	3 520 000	220 000
openrouter	grok-4-1-fast	220 000	550 000	55 000
openrouter	gemini-3.1-pro-preview	2 200 000	13 200 000	550 000

Anthropic and OpenRouter prices include the Level5 10% markup. OpenAI prices are list rates; runtime markup applied. Use GET /v1/pricing for live rates.

---

Computational sovereignty is not a feature. It is a right.