v0.6.0: Claude Code provider (no API key)
Highlights
Claude Code provider: run copperhead without an OpenAI or Anthropic API key.
If you already have Claude access through your Claude Code login, copperhead can now use it as a reasoning backend. No API key, no separate billing setup:
npm install -g copperhead
copperhead do "add a second RGB LED on an RTC-capable pin" --model claude-codeClaude Code runs reasoning-only inside copperhead: it only proposes actions, and copperhead stays the sole dispatcher for every file edit, KiCad ERC/DRC check, and commit gate. All the existing safety rails (spec-gated edits, verification-gated commits, git-snapshot rollback) are unchanged. If the Agent SDK ever emits a tool call, the run fails loudly rather than letting an edit bypass those gates.
Details
- Select it with
--model claude-code, or pin a model id with--model claude-code:<model-id>. - Auth uses your saved Claude Code login. If you are not logged in, run
claude setup-tokenand setCLAUDE_CODE_OAUTH_TOKEN. copperhead never reads the credential itself. - A claude-code run strips any billed
ANTHROPIC_API_KEY/OPENAI_API_KEYfrom the SDK subprocess, so it never quietly falls back to paid usage even when one is set. - The
@anthropic-ai/claude-agent-sdkis an optional dependency, so a normal install stays lean: only Claude Code users pull it in, and its absence surfaces an actionable install message. - The direct OpenAI and Anthropic providers, and the deterministic
check/verifycommand, are unchanged.
Also in this release
- The OpenAI provider now supports reasoning models and Gemini.
- Supplier BOM export for JLCPCB, DigiKey, and Mouser.
- A fab documentation-presence check added to the release gate.
npm: https://www.npmjs.com/package/copperhead/v/0.6.0
Full changelog: v0.5.0...v0.6.0