Docs · Claude Code

Melira works with Claude Code

Melira protects your AI agent’s memory and instruction files. When you use Claude Code, it can show you that protection right where you work - no dashboard, no login.

Two things you get

A status line at the bottom of Claude Code: 🛡 PoisonZero 🟢 v1.7.2 · 12 blocked · last action: 3h ago - health light, version, how many changes were blocked, and how long ago Melira last acted. 🟡 means it is running but reporting a problem; when it is not running the line reads 🛡 PoisonZero 🔴 not running.
A /poisonzero command (or its short form /pz) you can type in Claude Code: it shows how much Melira blocked this month, broken down by attack type.

Turn it on

Run once:

poisonzero setup-claude

This adds the /poisonzero command and its short form /pz. The status line is a separate step: poisonzero setup-claude --statusline-only. It is only set up if Claude Code is installed (~/.claude exists) and you don’t already have a status line. The same one-shot command setup works for Codex, Gemini CLI and Cursor too. Unattended install: PZ_SETUP_AGENTS=claude sets up the commands, PZ_SETUP_STATUSLINE=1 the status line.

It never overrides your setup

If you already use a custom status line, Melira leaves it untouched. To show both, have your own status-line script also run poisonzero stats --statusline and append the output.

Fully local

The numbers come from a local stats file on your machine. Nothing leaves your device, and Melira never writes into Claude’s memory. Related: memory poisoning and fail-closed design.