Feature request: license the Claude Code harness (source-available is sufficient)

Summary

Open -- or make source-available -- the Claude Code harness (CLI, GUI, mobile
client, tool definitions, prompt scaffolding) while keeping the model and the
business closed. The harness is a commodity client; the moat is the model behind
api.anthropic.com, and every fork still calls it.

Why now -- the secrecy is already spent

@anthropic-ai/claude-code v2.1.88 (2026-03-31) shipped a source map exposing the
full harness: roughly 512K lines of unobfuscated TypeScript across ~1900 files. The
harness is already public in practice. The only open question is licensed exposure
vs. the current unlicensed leak -- this is a cleanup, not a concession.

Why it's low-risk

• The moat is the model, not the client.
• The harness is already replicated many times over (OpenAI Codex CLI, OpenCode, Aider).
• Opening it changes nothing about competitive position: forks still depend on the

closed model, weights, inference stack, and account/billing.

• Lock-in deepens through ecosystem, not obscurity.

Why it helps

• A licensed source tree lets the paying Max/Pro base contribute directly -- point

their own Claude at the real source, reproduce harness bugs, and send PRs against
actual code instead of guessing from a leaked source map.

• It gives external contributors a real target for the harness-side issues that

currently can't be diagnosed without the source.

If competitor reuse is the concern

Source-available (readable, buildable, no competitor redistribution) is enough. Full
OSI-open licensing is not required for any of the above.

Scope

| Open / source-available | Stays closed |
| ----------------------- | --------------------- |
| CLI | Model + weights |
| GUI | Training data + methods |
| Mobile client | Inference stack |
| Tool definitions | Account / billing |
| Prompt scaffolding | |