bloodf/klaude.md

klaude Setup — MiniMax-powered Claude Code wrapper

What this does

klaude is a drop-in wrapper for claude that routes all model calls through MiniMax (MiniMax-M2.7) instead of Anthropic, using the Anthropic-compatible API. Regular claude is untouched — both commands coexist independently.

Prerequisites

claude CLI installed and accessible in PATH (~/.local/bin/claude or similar)
A MiniMax API key from https://platform.minimax.io/user-center/basic-information/interface-key
The ~/.klaude folder copied from the previous machine (or set up fresh below)

Setup Instructions

Paste the following prompt to Claude Code in a fresh session on the new machine:

Prompt to paste

I need you to set up (or restore) the `klaude` CLI wrapper on this machine.
The ~/.klaude folder may already exist (copied from another machine).

Do all of the following steps in order:

---

**STEP 1 — Find where `claude` is installed**

Run: `which claude`

Note the directory (e.g. ~/.local/bin). The klaude script must go there.

---

**STEP 2 — Ensure ~/.klaude structure exists**

Run:
```bash
mkdir -p ~/.klaude ~/.klaude/config-dir

STEP 3 — Create or verify ~/.klaude/config

If ~/.klaude/config does not exist, create it:

# klaude configuration — MiniMax API credentials
# Get your API key from: https://platform.minimax.io/user-center/basic-information/interface-key
MINIMAX_API_KEY=your_api_key_here

# API base URL — change to https://api.minimaxi.com/anthropic if you're in China
MINIMAX_BASE_URL=https://api.minimax.io/anthropic

If it exists, read it and confirm MINIMAX_API_KEY is set to a real key (not the placeholder).

STEP 4 — Sync symlinks in ~/.klaude/config-dir/

This step is critical. It must:

Check each expected symlink — repair it if broken or missing
Scan everything in ~/.claude/ and symlink anything not already linked (skip: settings.json, settings.local.json, history.jsonl, sessions, transcripts, projects, cache, todos, backups, debug, paste-cache, shell-snapshots, stats-cache.json, policy-limits.json, telemetry)

Run this script exactly:

CLAUDE_DIR="$HOME/.claude"
KLAUDE_DIR="$HOME/.klaude/config-dir"

# Items that must never be symlinked (auth, history, cache, session state)
SKIP=(settings.json settings.local.json history.jsonl sessions transcripts projects
      cache todos backups debug paste-cache shell-snapshots stats-cache.json
      policy-limits.json telemetry)

skip_item() {
  local name="$1"
  for s in "${SKIP[@]}"; do [ "$name" = "$s" ] && return 0; done
  return 1
}

echo "=== Syncing ~/.klaude/config-dir symlinks ==="
for item in "$CLAUDE_DIR"/*; do
  name=$(basename "$item")
  skip_item "$name" && continue
  target="$KLAUDE_DIR/$name"
  if [ -L "$target" ] && [ "$(readlink "$target")" = "$item" ]; then
    echo "  OK        $name"
  elif [ -L "$target" ]; then
    ln -sf "$item" "$target" && echo "  REPAIRED  $name"
  elif [ ! -e "$target" ]; then
    ln -sf "$item" "$target" && echo "  ADDED     $name"
  else
    echo "  SKIPPED   $name (real file exists, not a symlink)"
  fi
done
echo "=== Done ==="

Report all ADDED, REPAIRED, and SKIPPED items.

STEP 5 — Create or verify ~/.klaude/config-dir/settings.json

If it does not exist, create it by reading ~/.claude/settings.json and copying all non-auth fields (plugins, statusLine, env except Anthropic keys, etc.).

If it exists, read both files and confirm the plugin list and statusLine match ~/.claude/settings.json. Update any differences.

STEP 6 — Create or verify the klaude wrapper script

Check if the klaude script exists at the same directory as claude. If missing or outdated, write it with this exact content:

#!/usr/bin/env bash
# klaude — Claude Code wrapper using MiniMax models
# Config: ~/.klaude/config

CONFIG="$HOME/.klaude/config"
CONFIG_DIR="$HOME/.klaude/config-dir"

if [ ! -f "$CONFIG" ]; then
  echo "klaude: config file not found at $CONFIG" >&2
  exit 1
fi

source "$CONFIG"

if [ -z "$MINIMAX_API_KEY" ] || [ "$MINIMAX_API_KEY" = "your_api_key_here" ]; then
  echo "klaude: set MINIMAX_API_KEY in $CONFIG" >&2
  exit 1
fi

exec env \
  CLAUDE_CONFIG_DIR="$CONFIG_DIR" \
  ANTHROPIC_BASE_URL="${MINIMAX_BASE_URL:-https://api.minimax.io/anthropic}" \
  ANTHROPIC_AUTH_TOKEN="$MINIMAX_API_KEY" \
  API_TIMEOUT_MS="3000000" \
  CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC="1" \
  ANTHROPIC_MODEL="MiniMax-M2.7" \
  ANTHROPIC_SMALL_FAST_MODEL="MiniMax-M2.7" \
  ANTHROPIC_DEFAULT_SONNET_MODEL="MiniMax-M2.7" \
  ANTHROPIC_DEFAULT_OPUS_MODEL="MiniMax-M2.7" \
  ANTHROPIC_DEFAULT_HAIKU_MODEL="MiniMax-M2.7" \
  claude "$@"

Then run: chmod +x <path-to-klaude>

STEP 7 — Verify

Run:

klaude -p "reply with only: which model are you?"

Expected: MiniMax-M2.7 with no auth conflict warnings.

If there is an auth conflict warning, the CLAUDE_CONFIG_DIR symlink step may have accidentally linked an auth file. Re-check the skip list in Step 4.

Why CLAUDE_CONFIG_DIR? Without it, klaude inherits the stored Anthropic OAuth session from ~/.claude/, causing an "auth conflict" warning. A separate config dir has no stored session, so no conflict. All other config (plugins, MCP, skills, hooks) comes through symlinks.