Using with Claude Code

Nitpick works with or without Claude Code. This guide covers the Claude Code integration.

Two ways Claude Code fits in

1. As a provider

llm:
 provider: claude-code

Nitpick’s CLI runs normally, but LLM calls route through the claude subprocess (instead of direct API). Use when you want billing to route through your Claude Code account.

Requires @anthropic-ai/claude-code installed globally.

2. As the runtime

Install the bundled skills into ~/.claude/skills/:

./scripts/install-skill.sh

Symlinks skills/nitpick/, skills/prd-e2e-orchestrator/, and skills/multi-role-e2e/ into your Claude Code skills directory.

Then:

claude

And ask conversationally:

“I need to test my app at https://app.example.com. I changed the task form. Credentials: admin@example.com / pass123.”

Claude will:

1. Activate the nitpick skill
2. Run the crawler (via Bash tool)
3. Propose affected pages (in chat)
4. Wait for your confirmation
5. Delegate per-page testing via the Agent tool to subagents running prd-e2e-orchestrator
6. Present Phase 3.5 summaries in chat for approval
7. Show you the unified report inline

No CLI. All interaction is conversational.

When to use which

	CLI / HTTP API	Claude Code (runtime)
Interface	Terminal / REST	Conversational
Config	nitpick.yaml + .env	Inline in chat if you want
LLM	Your API key	Your Claude Code session
Parallelism	Node subprocess model	Agent tool (subagents)
Best for	CI, scripts, scheduled	Ad-hoc investigation

Uninstalling the skill

rm ~/.claude/skills/{nitpick,prd-e2e-orchestrator,multi-role-e2e}