Manifestack

Manifestack is an open-source skill for coding agents such as Claude Code, Cursor and Codex. Ask it a question like "what breaks first at 50k users?" and it reports the bills, outages and compliance gaps ahead of you, each finding with a fix. Prices and limits are read from vendors' own public pages.

What It Is

Manifestack is an agent skill that helps pick, audit and guard the third-party services in a software stack. It reads your repo (JavaScript, Python and Go dependencies, config files and env var names) and checks each service against requirements such as budget, user growth and data residency. It is read-only: it writes only inside .manifestack/, and needs no API keys or account.

How the workflow runs

• New project: it asks about users, budget, data rules and team, then proposes and prices a stack from launch to 100k users.
• Existing project: it scans the product and reports what breaks or costs too much, ranked by severity, with timing, cost and a fix per finding.
• Vendor comparison: /manifestack compare neon supabase weighs two or three vendors on your workload, including what a switch would take.
• Every change: in Claude Code and Cursor, a hook flags each newly added vendor SDK and checks it against your requirements.

Memory in STACK.md

Requirements, plans and the reasons behind decisions are stored in .manifestack/STACK.md in your repo. Entries can carry a revisit trigger, for example a database size threshold, so later audits know when to recommend a change. Decisions are written only after you confirm, and secrets are never stored.

Setup path

It requires Node.js 22 or newer. Run npx manifestack to detect installed agents, or pass --agent claude-code, cursor or codex. It also detects Copilot, Windsurf, OpenCode, Cline and Gemini CLI, and can be installed as a Claude Code plugin or via the skills CLI. The project is MIT licensed.