sunpeak AGENTS.md

A set of project instructions for sunpeak, a framework for building interactive MCP Apps that run inside AI chat hosts such as ChatGPT and Claude.

In plain words
What is it for?
Use it when developing sunpeak apps, testing them locally or in continuous integration, working with the embedded inspector, or supporting multiple chat hosts.
Why use it?
It keeps project terminology, navigation, testing, and design choices consistent while developers build and inspect these apps.

Instructions file for CodexOpenCode

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add instructions/alignbase/sunpeak/agents-md
Clone the repo
git clone --depth 1 https://github.com/Alignbase/sunpeak

Made for: Codex, OpenCode.

Per session 8,763 This file is loaded in full into every session.
When invoked 8,763 The same file — it is already loaded in full.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.08763 $0.08763
Opus 5 $0.04381 $0.04381
Sonnet 5 $0.01753 $0.01753
Haiku 4.5 $0.00876 $0.00876

Measured yesterday against content hash f6ebca83aba0, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

sunpeak AGENTS.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

AGENTS.md · 348 lines

How it starts

The opening of the file, as written. The whole thing — 348 lines — stays where its author put it; the contents beside it link to each section on GitHub.

sunpeak

Note that "sunpeak", except where required in URLs or code, is always lowercase.

sunpeak is a framework for building MCP Apps with interactive UIs that run inside AI chat hosts (ChatGPT, Claude, and future major hosts). Built on top of the MCP Apps SDK (@modelcontextprotocol/ext-apps).

ChatGPT apps are now submitted and published as plugins. The app portion is still based on MCP Apps or the OpenAI Apps SDK, so sunpeak's protocol and runtime architecture do not change. Adding a custom app requires ChatGPT Developer mode, enabled from the bottom-left user menu under Settings > Security and login > Developer mode. Local developer-mode apps are then created from the ChatGPT homepage under Plugins > +, while public apps are submitted as MCP-backed plugins through the OpenAI Platform.

The value proposition of the sunpeak framework is to help developers and their agents:

  1. Test MCP Apps locally and automatically (in CI/CD) using a replica of the ChatGPT and Claude runtimes.
    1. Save time manually testing all possible host, server, app, ui, and backend states.
    2. Protect developers from 4-click manual refreshes on every code change in each host.
    3. Cancel all the $20 per person per host per month testing accounts.
    4. Avoid burning host credits on every test and code change.
  2. Verify MCP tools work across multiple LLM models via the eval framework. Evals connect to the MCP server, discover tools, and send prompts to multiple models (GPT-4o, GPT-4o-mini, o4-mini, Claude Sonnet, Gemini 2.0 Flash) via the Vercel AI SDK. Cases can include App Context for follow-up turns that depend on model-visible UI state. Each case runs N times per model and reports pass/fail counts, so developers can confirm that tool descriptions, schemas, and App Context work reliably on smaller and cheaper models. Opt-in via sunpeak test --eval because evals cost money.
  3. Automate the real-host testing loop with live tests (sunpeak test --live): scripts that drive a real browser into ChatGPT (and other hosts as they're supported), send prompts that trigger MCP tool calls against the developer's server, and assert against the actually-rendered app via Playwright. Live tests catch what the inspector can't (real MCP connection behavior, real LLM tool invocation, host-specific iframe rendering, production resource loading) and replace the manual prompt-and-click loop. Opt-in because they hit real accounts.
  4. Build multi-platform MCP Apps in a structured way that's easy to understand and get started.
  5. Test their MCPs in ChatGPT with HMR and Claude with automatic rebuilds and refresh notifications.
  6. Embed the Inspector as a React component inside any third-party React app to render arbitrary MCP Apps. The <Inspector app={...} onCallTool={...} /> shape exposes the same double-iframe runtime, host shells, and conversation chrome as the CLI inspector, but without any sunpeak-owned servers or /__sunpeak/* runtime dependencies. Tool calls flow through a callback the embedder owns; resource HTML is passed as a string. Embedders host the static sandbox proxy (dist/sandbox-proxy.html) on a separate origin from their own app — or fall back to the same-origin srcdoc variant for zero deployment. This is a first-class use case; treat it as load-bearing when designing changes.

Read the full file on GitHub · 348 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. yesterday First seen · 348 lines · 8,763 tokens per session scan A f6ebca83aba0

Subscribe to this mod's changes

sunpeak AGENTS.md is an instructions file published in the GitHub repository Alignbase/sunpeak (214 stars, last pushed 3d ago), licensed MIT. It adds 8,763 tokens to every session, about $0.0438 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.