Evermind — Development skill for Claude Code

---

Why

Every request in Claude Code re-reads the whole conversation, so usage grows with context size. A long chat gets expensive and slow, and `/clear` throws away what you learned.

evermind fixes both:

• It restarts from a small context before the chat gets heavy: writes a short handoff, then you type /clear.
• It remembers everything from the moment it is installed: prompts, replies and tool calls are captured, durable facts are extracted, and relevant older memory is injected back before each prompt.
• It shows you what is happening, live, in a minimalist dashboard that opens by itself.

In our own measurements on real sessions, this cut estimated usage by roughly 30 to 40 percent on long chats.

How it works

flowchart LR
    A[You chat] --> B[Hook: capture
prompts, replies, tools]
    B --> C[(Local memory
facts + history)]
    A --> D{Context over 150K
or cache cold?}
    D -- yes --> E[Handoff
Goal / State / Next]
    E --> F[You type /clear]
    F --> G[Fresh session
digest + handoff injected]
    C --> G
    C -- relevant facts --> H[Recall hook
before each prompt]
    H --> A

Piece	What it does
refresh_check.py	Hook. Triggers a refresh when context passes 150K tokens, or when the cache went cold over 60K+.
handoff_check.py	Validates