Question

Claude Code is a TUI application that we wrap in a PTY. The Claude Code development team must also wrap their TUI in some kind of terminal harness for testing. Understanding their approach could:

1. Reveal a better testing strategy for unleashed
2. Identify a supported integration point we're missing
3. Help understand how they test on Windows / git-bash (if they do)
4. Inform whether our PTY byte-matching approach aligns with or fights their architecture

Research Tasks

• Search for Anthropic blog posts about Claude Code testing infrastructure
• Check the anthropics/claude-code repo for test harness code
• Look for any public talks / conference presentations about Claude Code internals
• Check if they use Playwright, expect/pexpect, or a custom terminal emulator for testing
• Investigate whether they test on Windows at all (git-bash, PowerShell, cmd.exe)
• Look for any GitHub discussions or issues about Windows-specific permission behavior

Context

Unleashed exists partly because Claude Code's permission system doesn't handle git-bash on Windows well — commands with pipes and chains get bastardized. The /sync-permissions skill was an attempt to work within the system but hit limits. Understanding how the team intended permissions to work on Windows would clarify whether we're fighting a bug or a design decision.

Bigger picture

If the Claude Code team has a terminal wrapper for testing, we might be able to reuse or learn from it rather than maintaining our own PTY wrapper.

Source

Deep audit discussion (2026-02-12).