Move shell environment setup into install-hooks

[svarlamov]

Summary

• move Unix shell profile detection and PATH mutation into the Rust install-hooks command
• move Windows user PATH and Git Bash profile configuration into the Rust install-hooks command
• keep dry-run, idempotency, shell preference, fallback, skip-path, output, and MDM ownership behavior aligned with the existing installers
• remove the migrated logic from install.sh and install.ps1 and replace the obsolete BATS coverage with Rust unit and TestRepo integration tests

Impact

The install scripts are shorter, while direct install-hooks invocations now perform the same shell environment setup as package installation.

Validation

• task fmt
• task lint
• task build
• task test
• Windows target test compilation with cargo check --tests --target x86_64-pc-windows-msvc
• bash -n install.sh
• git diff --check

[svarlamov]

• Move shell environment setup into install-hooks #1971 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[devin-ai-integration]

Devin Review found 1 potential issue.

[devin-ai-integration]

🟡 A shell profile that cannot be written makes hook installation report failure

Setting up shell PATH entries can now abort the whole hook-installation command (shell_env::configure(...)? at src/commands/install_hooks.rs:350) whenever a shell profile cannot be written, so hooks that actually installed fine are reported as failed.
Impact: If any shell profile (e.g. an unwritable ~/.bashrc/.zshrc) can't be updated, the user sees a misleading "Failed to set up IDE/agent hooks" message and a non-zero exit even though hooks were installed successfully.

How the coupling changes failure behavior

Previously the shell PATH/profile setup lived in install.sh / install.ps1 and was independent of hook installation: the installer ran install-hooks first (warning on failure) and then did shell configuration separately. Now run() calls shell_env::configure(...)? right after async_run_install (src/commands/install_hooks.rs:349-350). configure_unix propagates any IO error from OpenOptions::open/writeln!/create_dir_all (src/commands/install_hooks/shell_env.rs:107-118) via ?. When that error bubbles up, run() returns Err, which in src/commands/git_ai_handlers.rs:166-169 causes eprintln!("Install hooks failed") and std::process::exit(1). In install.sh the branch if ! ... install-hooks; then warn "Warning: Failed to set up IDE/agent hooks..." then fires even though every hook installed correctly. It also skips cleanup_legacy_envelope_logs() (src/commands/install_hooks.rs:354-356).

Prompt for agents

In src/commands/install_hooks.rs:350, shell_env::configure(...) is invoked with the `?` operator, which makes any failure to write a user's shell profile (e.g. an unwritable ~/.bashrc) fatal to the entire install-hooks command. This diverges from the previous behavior where shell environment setup lived in install.sh/install.ps1 and was best-effort/independent of hook installation success. The consequence is that a shell-profile write error now causes install-hooks to exit non-zero (see git_ai_handlers.rs:166-169) and makes install.sh print the misleading 'Failed to set up IDE/agent hooks' warning even though all hooks installed fine. Consider making shell_env::configure best-effort: log a warning on error (matching the previous non-fatal semantics) instead of propagating the error out of run(). Note that configure_unix itself already accumulates results and only fails on IO errors; you likely want to catch/log those rather than return Err from run().

---

Was this helpful? React with 👍 or 👎 to provide feedback.