fabro/nodes/implement/response.md

The diff looks exactly right. Let me summarize all the changes made:

Summary of changes

1. lib/crates/fabro-daytona/src/lib.rs — Core refactor

• Added GitCloneParams struct with url: String and branch: Option<String> fields, publicly exported
• Changed DaytonaSandbox field from clone_branch: Option<String> to clone_params: Option<GitCloneParams>
• Updated new() signature: last param changed from clone_branch: Option<String> to clone_params: Option<GitCloneParams>
• Updated reconnect(): clone_params: None (was clone_branch: None)
• Refactored initialize():
  • Removed let cwd = std::env::current_dir() — no longer reads the process-global cwd
  • Replaced match detect_repo_info(&cwd) with if let Some(ref clone_params) = self.clone_params
  • Some arm: uses clone_params.url / clone_params.branch directly (URL is already HTTPS from caller)
  • else arm: creates empty working directory (the old Err arm logic)
• Removed unused import use fabro_github::ssh_url_to_https

2. lib/crates/fabro-cli/src/commands/run.rs — Production callers

• Main run path: Constructs GitCloneParams from origin_url (with ssh_url_to_https) and detected_base_branch, passes it to DaytonaSandbox::new()
• Doctor path: Already passes None — type changed from Option<String> to Option<GitCloneParams> but None is valid for both

3. lib/crates/fabro-workflows/tests/daytona_integration.rs — Test fixes

• create_env_with_github_app: Detects repo info with detect_repo_info(&cwd) and builds GitCloneParams before calling new(), preserving clone behavior for all tests that use this helper
• daytona_computer_use_browser_screenshot: Removed tempfile::tempdir() and set_current_dir() — passes None as last arg which now cleanly means "skip clone"
• daytona_playwright_mcp_sandbox_transport: Same — removed tempfile::tempdir() and set_current_dir()