Claude Next — Security skill for Claude Code

claude-next

**One-key handoff for deep Claude Code sessions.** Pass your work to a fresh window with an **independent audit** — no manual summary pasting, no copy-pasted "here's what I was doing".

[](./LICENSE) [](https://www.npmjs.com/package/claude-next) [](https://github.com/llmapi-pro/claude-next)

---

The problem

Claude Code sessions degrade past ~400K tokens. You've seen it:

• Fabricates file paths that never existed
• Skips TODOs it explicitly acknowledged earlier
• Contradicts its own setup decisions from 200 turns ago

The natural fix is **opening a new window**. But the manual workflow is painful:

1. Ask old Claude: "summarize what we were doing, write an opening message for a new window"
2. Copy-paste into new window
3. Ask new Claude: "does this match what's actually in the codebase?"
4. Ask old Claude to verify, loop

Steps 3-4 exist because **old Claude writing its own handoff produces self-congratulatory summaries** — it remembers claiming to push the image, forgets whether it actually did.

`claude-next` automates the loop, and adds an independent fresh-context auditor that checks every claim against the real filesystem before handoff.

---

What it does

 old window                    new window
 ──────────                    ──────────
 /next           →  handoff + audit stored
                                    │
                                    ▼
                        first message: "continue A"
                                    │
           drift-check + inject ◄───┘
                                    │
                                    ▼
                    Claude picks up exactly where you left off

• Old window: /next — produces a structured handoff, spawns a fresh-context subagent to audit