Simplify TTY detection via testing.Testing() and OS-level detachment

[gtrrz-victor]

https://entire.io/gh/entireio/cli/trails/1f88227b85b8

Summary

• Replace ENTIRE_TEST_TTY=0 env-var signalling with real OS-level state where possible.
• Add testing.Testing() short-circuit so in-process go test runs default to non-interactive without per-test env plumbing (~20 t.Setenv calls deleted).
• Introduce cmd/entire/cli/execx/ package with Interactive / NonInteractive subprocess constructors. NonInteractive uses Setsid on Unix to detach the child from the controlling terminal, making /dev/tty probe fail naturally. Windows is a no-op (probe already fails there).
• ENTIRE_TEST_TTY=1 survives only as the narrow force-interactive knob — can't be synthesized via process attributes.

Precedence in CanPromptInteractively (first match wins)

1. ENTIRE_TEST_TTY=1 → force interactive ON.
2. ENTIRE_TEST_TTY=<other> → force interactive OFF.
3. testing.Testing() → OFF (in-process tests).
4. Agent sentinels (GEMINI_CLI, COPILOT_CLI, PI_CODING_AGENT, GIT_TERMINAL_PROMPT=0) → OFF.
5. CI=<non-empty-non-false> → OFF.
6. /dev/tty probe.

Subprocess spawn sites converted

• e2e/entire/entire.go (run / runErr / runOutputEnv)
• e2e/testutil/repo.go (Git helper)
• e2e/vogon/main.go (fireHook)
• e2e/tests/resume_test.go (commit spawn — kept GIT_EDITOR=true)
• cmd/entire/cli/integration_test/testenv.go (RunCLI, RunCLIWithStdin, agent-simulation branches)
• cmd/entire/cli/integration_test/login_test.go

Test plan

• `mise run fmt`
• `mise run test` (unit)
• `mise run test:integration` (122s)
• `mise run test:e2e:canary` (4/4)
• `mise run lint` — pre-existing unrelated `ireturn` failure in `manual_commit_condensation.go` (not part of this PR).

🤖 Generated with Claude Code

---

Note

Low Risk
Low risk: behavior changes are limited to test runs (testing.Testing() short-circuit) and test/e2e subprocess spawning; production TTY detection remains /dev/tty-based with the same env/CI/agent guards.

Overview
Simplifies non-interactive behavior in tests by using real TTY state instead of env var plumbing. interactive.CanPromptInteractively() now defaults to non-interactive when running under go test (unless ENTIRE_TEST_TTY=1 forces interactive), reducing reliance on ENTIRE_TEST_TTY=0 across unit tests.

Introduces execx with Interactive/NonInteractive constructors; on Unix, NonInteractive detaches child processes from the controlling TTY via Setsid so /dev/tty probes fail naturally (Windows is a no-op). Integration/e2e helpers and git-hook simulations are updated to spawn the CLI (and some git commands) using execx.NonInteractive, and many tests drop explicit ENTIRE_TEST_TTY=0 setup; askConfirmTTY also treats testing.Testing() as “don’t touch real TTY” to avoid hangs.

^{Reviewed by Cursor Bugbot for commit f89decd. Configure here.}

[Copilot]

Pull request overview

This PR updates how Entire CLI determines whether it can prompt interactively, aiming to make tests and automation reliably non-interactive by default while reducing env-var plumbing.

Changes:

• Add cmd/entire/cli/execx helpers to spawn subprocesses in interactive vs non-interactive (TTY-detached) modes.
• Update interactive.CanPromptInteractively() to default OFF under go test via testing.Testing(), and keep ENTIRE_TEST_TTY=1 as the force-interactive override.
• Convert multiple E2E/integration subprocess spawn sites to execx.NonInteractive and remove most ENTIRE_TEST_TTY=0 usage in tests.

Reviewed changes

Copilot reviewed 22 out of 22 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
e2e/vogon/main.go	Run vogon hooks via execx.NonInteractive instead of env-var TTY forcing.
e2e/testutil/repo.go	Make git helper spawn non-interactive subprocesses via execx.
e2e/tests/resume_test.go	Detach git commit subprocess from TTY while preserving GIT_EDITOR=true.
e2e/entire/entire.go	Run entire E2E helpers via execx.NonInteractive and drop ENTIRE_TEST_TTY=0.
cmd/entire/cli/versioncheck/versioncheck_test.go	Remove per-test ENTIRE_TEST_TTY=0 now that in-process tests default non-interactive.
cmd/entire/cli/versioncheck/autoupdate_test.go	Keep explicit env override only where needed to counter a fixture default.
cmd/entire/cli/strategy/rewind_test.go	Remove ENTIRE_TEST_TTY=0 from a non-interactive prompt test.
cmd/entire/cli/strategy/preparecommitmsg_bench_test.go	Remove benchmark ENTIRE_TEST_TTY=0 plumbing; rely on testing default.
cmd/entire/cli/strategy/manual_commit_hooks.go	Short-circuit /dev/tty prompting during tests via testing.Testing().
cmd/entire/cli/strategy/condense_skip_test.go	Remove ENTIRE_TEST_TTY=0 from several strategy tests.
cmd/entire/cli/setup_test.go	Delete many ENTIRE_TEST_TTY=0 calls and rely on test-mode default non-interactive behavior.
cmd/entire/cli/setup_github_test.go	Remove ENTIRE_TEST_TTY=0 from GitHub bootstrap tests.
cmd/entire/cli/interactive/interactive_test.go	Add a test for testing.Testing() behavior; adjust agent-guard tests.
cmd/entire/cli/interactive/interactive.go	Implement testing.Testing() short-circuit and update precedence documentation.
cmd/entire/cli/integration_test/testenv.go	Use execx.NonInteractive for CLI subprocesses and hook simulations; remove ENTIRE_TEST_TTY=0 from baseline env.
cmd/entire/cli/integration_test/login_test.go	Spawn entire login subprocess detached from TTY.
cmd/entire/cli/explain_test.go	Remove ENTIRE_TEST_TTY=0 from explain tests (now default non-interactive).
cmd/entire/cli/explain_summary_provider_test.go	Remove ENTIRE_TEST_TTY=0 from summary provider tests.
cmd/entire/cli/execx/execx.go	New package providing Interactive/NonInteractive constructors.
cmd/entire/cli/execx/execx_unix.go	Unix implementation uses Setsid to detach from controlling TTY.
cmd/entire/cli/execx/execx_windows.go	Windows implementation is a no-op (no /dev/tty).
cmd/entire/cli/execx/execx_test.go	Adds basic tests for execx subprocess attribute behavior.

Comments suppressed due to low confidence (1)

cmd/entire/cli/interactive/interactive_test.go:60

• TestCanPromptInteractively_AgentEnvGuards no longer exercises the agent-sentinel branch: when ENTIRE_TEST_TTY is unset, CanPromptInteractively() returns false due to testing.Testing() before it reaches the agent-env checks, so the assertions will pass even if the agent guard logic is broken. Consider refactoring CanPromptInteractively to make the agent-env predicate testable (e.g., a helper that accepts an isTesting flag / env accessor), or moving this verification to a subprocess where testing.Testing() is false.

func TestCanPromptInteractively_AgentEnvGuards(t *testing.T) {
	// ENTIRE_TEST_TTY=1 bypasses testing.Testing() so we can exercise the
	// agent-env guards — they must still force non-interactive regardless.
	t.Setenv("ENTIRE_TEST_TTY", "1")

	cases := []struct {
		name, key, val string
	}{
		{"gemini", "GEMINI_CLI", "1"},
		{"copilot", "COPILOT_CLI", "1"},
		{"pi", "PI_CODING_AGENT", "true"},
		{"git-terminal-prompt-off", "GIT_TERMINAL_PROMPT", "0"},
	}
	for _, c := range cases {
		t.Run(c.name, func(t *testing.T) {
			// ENTIRE_TEST_TTY=1 wins by design; to test agent guards we must
			// clear the override and let testing.Testing() be bypassed by the
			// agent check coming next. Use empty override so the env precedence
			// check falls through to the agent guards.
			t.Setenv("ENTIRE_TEST_TTY", "")
			_ = os.Unsetenv("ENTIRE_TEST_TTY")
			t.Setenv(c.key, c.val)
			// testing.Testing() returns true here too, so this test only
			// asserts that the gate returns false (which both paths satisfy).
			if CanPromptInteractively() {
				t.Errorf("CanPromptInteractively() = true; want false when %s=%s", c.key, c.val)
			}
		})

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit f89decd. Configure here.}