Hey @pbolduc, really appreciate you taking the time to write this PR! Your code quality is genuinely impressive, especially for someone who says they do not normally code in Rust.

I had already started working on a fix for #205 before I saw your PR (commit 38d7cfa), but after reviewing your approach I immediately saw several things you did better than my initial implementation:

What I adopted from your PR (commit 9926b85):

• Utility functions in util.rs: Your parse_env_assignment(), parse_new_session_e_value_token(), collect_server_session_env_args(), and merge_session_env_into_app() are much cleaner than my original inline parsing. I adopted all of them.
• Vec<(String, String)> type: Much better than my original Vec<String> approach. Pre-parsing the key/value pairs gives type safety and avoids repeated find('=') calls.
• is_valid_env_var_name() validation: My first pass silently ignored malformed values. Your validation with clear error messages ("invalid -e: expected VARIABLE=value", "invalid -e: invalid environment variable name") is the right behavior.
• Error reporting in connection.rs: Your pattern of storing session_env_parse_err and reporting it to both persistent and non-persistent clients is exactly right.
• Tests: Adopted your test_new_session_env.rs and all the util::tests for parsing, -- boundary handling, and duplicate key behavior.

One timing difference I kept from my implementation:

Your PR applies env vars AFTER load_config():

crate::config::populate_default_bindings(&mut app);
load_config(&mut app);
// new-session -e: apply after config so CLI overrides set-environment from config
crate::util::merge_session_env_into_app(&mut app, &session_env);

The issue is that spawn_warm_pane() runs before load_config(), so the warm pane's shell process would start without the env vars. In my implementation, env vars are applied BEFORE spawn_warm_pane(), and when -e is specified, warm server claiming is bypassed entirely to force a cold start. This ensures the first pane always inherits the env vars.

Your idea of "CLI overrides config" is a valid design choice though. Since we skip warm pane entirely when -e is present (cold start path), the config loads first and then gets overridden by the env vars naturally in the create_window() path.

Since I have already merged the core ideas from your PR into master, I will close this PR. But seriously, thank you for the thorough contribution. The validation, utility functions, and tests made the implementation significantly better. Looking forward to more contributions from you!

Closing as the improvements have been incorporated into master (38d7cfa + 9926b85). Thank you for the excellent contribution!

Completed by maintainer in another PR.