fix(issue-23-24): merge review - the environment form of the auto-trust switch reaches the handler; a rate-limited dialog does not hide the next one

Two probes from the merge-side codex review (its report was lost to an API
outage; the probes and their results were recovered from its transcript):

- create_session.sh and resume_session.sh set AUTO_TRUST=1 and then
  exported MAM_AUTO_TRUST from it, so the documented MAM_AUTO_TRUST=0
  without a flag reached handle_startup_dialogs as 1: the operator had
  turned automatic trust off and it stayed on. Now AUTO_TRUST defaults to
  the environment value and an explicit --auto-trust/--no-auto-trust wins.
- _mam_dialog_pick exited on the first dialog inside its resend cooldown
  instead of looking further down the screen, so a skipped (--no-auto-trust)
  or just-answered trust prompt hid the codex update notice below it, which
  the docs say is always answered. It continues to the next dialog now; S5
  and S5b (one send per cooldown, one retry after it) still hold.

Regressions: S6c, S5f (startup path), R3b and C3 (resume/create, env only,
flag over env). Each fails without the fix. The create skill doc names the
environment form and its precedence.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VCgu9Fa9iT7TbJUSzdq4Ff

tests/test_issue23_24_startup_path.py

@@ -578,6 +578,32 @@ def test_s6_no_auto_trust_leaves_trust_dialogs_alone_but_answers_notices(
     assert sent == ["send-keys -t test-sess 2"], (res.stderr, sent)
 
 
+def test_s6c_a_skipped_trust_dialog_does_not_hide_the_notice_below_it(mam_sandbox):
+    """S6c (merge review, codex probe): with --no-auto-trust and a codex screen
+    that shows the trust prompt and the update notice together, the trust
+    prompt is skipped and recorded, and the skip's cooldown must not stop the
+    picker from answering the notice on the next poll."""
+    res, sent = _run_dialogs(
+        mam_sandbox, "codex",
+        "Do you trust the contents of this directory?\n > 1. Yes, continue\n"
+        "Update available\n 1. Update now\n 2. Skip\n 3. Skip until next version",
+        timeout=3, env={"MAM_AUTO_TRUST": "0", "MAM_DIALOG_RESEND_S": "30"})
+    assert sent == ["send-keys -t test-sess 2"], (res.stderr, sent)
+    assert "no-auto-trust" in res.stderr, res.stderr
+
+
+def test_s5f_a_dialog_inside_its_cooldown_does_not_block_the_next_one(mam_sandbox):
+    """S5f (merge review, codex probe): two dialogs on one screen, auto-trust on.
+    The trust prompt is answered first; while it is rate-limited the notice
+    must still be answered, once each."""
+    res, sent = _run_dialogs(
+        mam_sandbox, "codex",
+        "Do you trust the contents of this directory?\n > 1. Yes, continue\n"
+        "Update available\n 1. Update now\n 2. Skip\n 3. Skip until next version",
+        timeout=3, env={"MAM_DIALOG_RESEND_S": "60"})
+    assert sent == ["send-keys -t test-sess Enter", "send-keys -t test-sess 2"], (res.stderr, sent)
+
+
 def test_s6b_create_session_no_longer_gates_dialogs_on_claude(mam_sandbox):
     """S6b (source read): the #24 root cause was the `AGENT = claude` gate at
     the create_session call site, so the gate's absence is the regression."""