bab-ilu: Skill for Claude Code

/setup

Bab-ilu's configuration surface is intentionally tiny: Claude Code handles its own auth via claude login, and one optional MCP second-opinion LLM can be wired up if you want cross-model review. That's it. /setup walks you through the Review LLM.

Safe to re-run at any time — only updates keys you choose to configure.

Inputs

• No arguments required.
• Reads: .env (current state), config/setup-guide.md (reference).

Outputs

• Updated .env with any newly configured LLM_* keys.
• A summary of current configuration status.

Wiki Interaction

• None. /setup runs before any vault exists and never writes to wiki/.

Workflow

Step 1: Read configuration reference

Read config/setup-guide.md for the complete description of the Review LLM setup, provider table, and verification script.

Step 2: Detect current state

python3 -c "
import sys, os
sys.path.insert(0, 'tools')
try:
    import _env  # loads .env into os.environ
except Exception:
    pass
keys = {
    'LLM_API_KEY':  'Review LLM (API key)',
    'LLM_BASE_URL': 'Review LLM (base URL)',
    'LLM_MODEL':    'Review LLM (model)',
}
for k, label in keys.items():
    v = os.environ.get(k, '').strip()
    print(f'SET:{k}' if v else f'UNSET:{k}')
"

Also detect the Python environment:

ls .venv/ 2>/dev/null && echo "venv:present" || echo "venv:absent"
python3 --version

Step 3: Show configuration status

Present a concise status summary:

Bab-ilu Configuration Status
============================
✓  Claude Code auth          — managed by `claude login`

Optional:
✗  Review LLM (LLM_* ×3)     — not set  (cross-model MCP review unavailable)

Ask: "Configure the Review LLM now, or skip?"

Step 4: Configure Review LLM (if user opts in)

Explain: "The Review LLM powers mcp-servers/llm-review — a second model that critiques Bab-ilu artifacts independently of Claude. Works with any OpenAI-compatible API. Without it the MCP server simply does not start; no Bab-ilu skill hard-depends on it."