From e971edc0dd93b45f5bd0825fd46b300e22385a4d Mon Sep 17 00:00:00 2001
From: JYunth <jyunth28@gmail.com>
Date: Mon, 9 Nov 2026 16:22:12 +0530
Subject: [PATCH 1/2] loopguard: a heredoc's lines are not test runs

status_of_tests split the raw command, so a heredoc line that starts with a
test runner counted as a test run of its own. A turn whose last real run
failed, followed by `gh pr create --body-file - <<'EOF'` with `pytest -q`
in the body, was recorded as ended green, and the judge was told the last
verification passed.

It now drops heredoc bodies before splitting, as exit_status_covers already
does. A runner that reads a heredoc (`pytest -q <<EOF`), or runs after one,
is now judged by its own status, where before it was read as "mixed".
---
 tests/test_live_core.py           | 26 ++++++++++++++++++++++++++
 tokeneconomizer/loopguard/turn.py |  2 +-
 2 files changed, 27 insertions(+), 1 deletion(-)

diff --git a/tests/test_live_core.py b/tests/test_live_core.py
index 4dae487..c54da93 100644
--- a/tests/test_live_core.py
+++ b/tests/test_live_core.py
@@ -1336,6 +1336,32 @@ def test_a_non_test_command_mentioning_a_runner_is_not_a_verification(cmd):
     assert g.verifications == 1 and not g.resolved_proxy()
 
 
+def test_a_heredoc_writing_a_script_runs_no_tests():
+    assert turn.status_of_tests("cat <<EOF > ci.sh\nnpm test\nEOF") is None
+
+
+@pytest.mark.parametrize("cmd", ["gh pr create --body-file - <<'EOF'\npytest -q\nEOF",
+                                 "gh issue comment 7 --body-file - <<EOF\nnpm test\nEOF"])
+def test_a_heredoc_line_naming_a_runner_is_not_a_verification(cmd):
+    """A heredoc's lines are the input of the command that reads them, as in
+    `exit_status_covers`: a body written with one runs no tests, so the failing run before it
+    is still the turn's last verification."""
+    assert turn.status_of_tests(cmd) is None
+    g = guard.TurnGuard(turn_key="t")
+    g.fold("Edit", {"file_path": "/r/a.py", "old_string": "x", "new_string": "y"}, "ok")
+    g.fold("Bash", {"command": "pytest -q"}, "FAILED (failures=1)", 1)
+    g.fold("Bash", {"command": cmd}, "ok", 0)
+    assert g.verifications == 1 and not g.resolved_proxy()
+
+
+@pytest.mark.parametrize("cmd", ["pytest -q <<EOF\ny\nEOF",
+                                 "cat <<EOF > ci.sh\nnpm test\nEOF\npytest -q"])
+def test_a_runner_beside_a_heredoc_is_still_judged_by_its_status(cmd):
+    """A runner that reads a heredoc, or runs after one, is the last command: its status is the
+    tests' own."""
+    assert turn.status_of_tests(cmd) == "own"
+
+
 TAP_FAILING = "# tests 9\n# suites 0\n# pass 8\n# fail 1\n# cancelled 0"
 
 
diff --git a/tokeneconomizer/loopguard/turn.py b/tokeneconomizer/loopguard/turn.py
index f973696..f942654 100644
--- a/tokeneconomizer/loopguard/turn.py
+++ b/tokeneconomizer/loopguard/turn.py
@@ -141,7 +141,7 @@ def status_of_tests(cmd):
     (Codex) is judged by its script."""
     if not isinstance(cmd, str):
         return None
-    pieces, pipefail = _pieces(cmd)
+    pieces, pipefail = _pieces(_split_heredocs(cmd)[0])
     found = None
     for i in range(0, len(pieces), 2):
         stage = pieces[i]

From 66942914c350789a5bf2bddcc6a2d19e41e2b1f3 Mon Sep 17 00:00:00 2001
From: Aaron <aaronjli@berkeley.edu>
Date: Tue, 10 Nov 2026 09:15:01 -0800
Subject: [PATCH 2/2] loopguard: drop a Codex script's heredocs after the
 unwrap

A heredoc that ends a bash -lc '<script>' closes on EOF' in the recorded command, so dropping heredoc bodies before the unwrap also dropped the rest of the script, and the test run before the heredoc was lost. _pieces now drops them after the unwrap, for both status_of_tests and exit_status_covers.
---
 tests/test_live_core.py           | 12 ++++++++++++
 tokeneconomizer/loopguard/turn.py | 10 ++++++----
 2 files changed, 18 insertions(+), 4 deletions(-)

diff --git a/tests/test_live_core.py b/tests/test_live_core.py
index c54da93..0c6b593 100644
--- a/tests/test_live_core.py
+++ b/tests/test_live_core.py
@@ -1362,6 +1362,18 @@ def test_a_runner_beside_a_heredoc_is_still_judged_by_its_status(cmd):
     assert turn.status_of_tests(cmd) == "own"
 
 
+@pytest.mark.parametrize("cmd, want", [
+    ("bash -lc 'pytest -q <<EOF\ny\nEOF'", "own"),
+    ("bash -lc \"pytest -q && cat <<'EOF'\nsummary\nEOF\"", "own"),
+    ("bash -lc \"gh pr create --body-file - <<'EOF'\npytest -q\nEOF\"", None),
+])
+def test_a_heredoc_that_ends_a_codex_script_is_read_inside_the_script(cmd, want):
+    """Codex records `bash -lc '<script>'`. A heredoc that ends the script closes on `EOF'` there,
+    not on a line of its own, so its body is left out after the unwrap: the run before it still
+    counts, and a runner named in its body still does not."""
+    assert turn.status_of_tests(cmd) == want
+
+
 TAP_FAILING = "# tests 9\n# suites 0\n# pass 8\n# fail 1\n# cancelled 0"
 
 
diff --git a/tokeneconomizer/loopguard/turn.py b/tokeneconomizer/loopguard/turn.py
index f942654..cf2f2e9 100644
--- a/tokeneconomizer/loopguard/turn.py
+++ b/tokeneconomizer/loopguard/turn.py
@@ -71,8 +71,9 @@
 
 def _pieces(cmd):
     """`cmd` as its commands and the separators between them (`[cmd, sep, cmd, ...]`), with
-    quoted text masked, and whether it sets `pipefail`. A command recorded as `bash -lc
-    '<script>'` (Codex) is split by its script."""
+    quoted text masked and heredoc bodies left out, and whether it sets `pipefail`. A command
+    recorded as `bash -lc '<script>'` (Codex) is split by its script: its heredocs are read
+    there, since one that ends the script closes on `EOF'`, not on a line of its own."""
     try:
         parts = shlex.split(cmd)
     except ValueError:
@@ -80,6 +81,7 @@ def _pieces(cmd):
     if (len(parts) >= 3 and os.path.basename(parts[0]) in _SHELLS
             and parts[1].startswith("-") and "c" in parts[1]):
         cmd = parts[2]
+    cmd = _split_heredocs(cmd)[0]       # a heredoc's lines are input, not commands
     pieces = _SEP_RE.split(re.sub(r"[(){}]", " ", _QUOTED_RE.sub("Q", cmd)))
     while len(pieces) > 1 and not pieces[-1].strip():      # a trailing `;` or newline
         del pieces[-2:]
@@ -123,7 +125,7 @@ def exit_status_covers(cmd):
     lines are the input of the command that reads them, not commands of their own."""
     if not isinstance(cmd, str) or not cmd.strip():
         return False
-    pieces, pipefail = _pieces(_split_heredocs(cmd)[0])
+    pieces, pipefail = _pieces(cmd)
     for i in range(0, len(pieces) - 1, 2):              # every command but the last
         sep = pieces[i + 1]
         if sep == "&&" or (pipefail and sep in ("|", "|&")):
@@ -141,7 +143,7 @@ def status_of_tests(cmd):
     (Codex) is judged by its script."""
     if not isinstance(cmd, str):
         return None
-    pieces, pipefail = _pieces(_split_heredocs(cmd)[0])
+    pieces, pipefail = _pieces(cmd)
     found = None
     for i in range(0, len(pieces), 2):
         stage = pieces[i]