The agent that grows with you
Python aiai-agentai-agentsanthropicchatgptclaudeclaude-codecodexhermeshermes-agentllmnous-researchopenai
fix(agent): keep shell exit codes when recording pipelines
BrokeSkill2b7ab9371136
diff --git a/agent/verification_evidence.py b/agent/verification_evidence.pyindex 5f2b8b5d995..32c12dca6ce 100644--- a/agent/verification_evidence.py+++ b/agent/verification_evidence.py@@ -244,7 +244,7 @@ def _segment_status(),default=sequence_start,)- if pipeline_status and match_index == pipeline_start:+ if pipeline_status and len(pipeline_status) >= 2 and match_index == pipeline_start:return pipeline_status[0]return Noneif not operators:diff --git a/agent/verification_stop.py b/agent/verification_stop.pyindex 6a3ecf7eaca..f5f835fc6a4 100644--- a/agent/verification_stop.py+++ b/agent/verification_stop.py@@ -16,9 +16,9 @@ _MAX_CHANGED_PATHS_IN_NUDGE = 8_MAX_STATUS_SUMMARY_CHARS = 1200_UNPIPED_RUN_WARNING = (- " Run the check bare: no `| tail`, `| head`, or `| grep` filter. A piped run's exit "- "status belongs to the filter, not the check, so it cannot count as evidence; the "- "terminal already truncates long output and spills the rest to a file."+ " Run the check directly. If you pipe it, keep the check as the first command in "+ "the pipeline so its exit status can be attributed; the terminal already truncates "+ "long output and spills the rest to a file.")# Prose/data extensions and extension-less prose filenames (case-insensitive) withdiff --git a/tests/agent/test_verification_evidence.py b/tests/agent/test_verification_evidence.pyindex 1fa91d2e035..1fb86b6740f 100644--- a/tests/agent/test_verification_evidence.py+++ b/tests/agent/test_verification_evidence.py@@ -117,6 +117,20 @@ def test_piped_verifier_without_pipeline_status_is_not_evidence(tmp_path, monkey) is None[email protected]("pipeline_status", [[0], [1]])+def test_piped_verifier_with_single_pipeline_status_is_not_evidence(+ tmp_path, monkeypatch, pipeline_status,+):+ """A one-element PIPESTATUS belongs to a compound command (``!``, ``if``), not a pipe."""+ monkeypatch.setenv("HERMES_HOME", str(tmp_path / ".hermes"))+ _python_project(tmp_path)++ assert classify_verification_command(+ "pytest -q 2>&1 | tail -20", cwd=tmp_path, session_id="s1",+ exit_code=0, pipeline_status=pipeline_status,+ ) is None++@pytest.mark.parametrize("command",[diff --git a/tests/agent/test_verification_stop.py b/tests/agent/test_verification_stop.pyindex 10a4c3c1dd3..865390e2723 100644--- a/tests/agent/test_verification_stop.py+++ b/tests/agent/test_verification_stop.py@@ -182,8 +182,8 @@ def test_path_under_a_deleted_directory_does_not_nudge(tmp_path, monkeypatch):assert build_verify_on_stop_nudge(session_id="s1", changed_paths=[gone]) is None-def test_nudge_demands_an_unpiped_verification_run(tmp_path, monkeypatch):- """A piped run reports the filter's exit status, so the ledger can never record it."""+def test_nudge_asks_for_an_attributable_pipeline(tmp_path, monkeypatch):+ """A piped run counts only when the check leads the pipeline."""monkeypatch.setenv("HERMES_HOME", str(tmp_path / ".hermes"))_node_project(tmp_path)changed = _edited(tmp_path, "src/app.ts")@@ -192,7 +192,7 @@ def test_nudge_demands_an_unpiped_verification_run(tmp_path, monkeypatch):nudge = build_verify_on_stop_nudge(session_id="s1", changed_paths=[changed])assert nudge is not None- assert "no `| tail`" in nudge+ assert "keep the check as the first command in the pipeline" in nudge# ---------------------------------------------------------------------------# Fix C: documentation/prose edits carry no verifiable behavior and must neverdiff --git a/tests/hermes_cli/test_agent_env_advertisement.py b/tests/hermes_cli/test_agent_env_advertisement.pyindex aaf0bd18856..7251be96f8c 100644--- a/tests/hermes_cli/test_agent_env_advertisement.py+++ b/tests/hermes_cli/test_agent_env_advertisement.py@@ -53,6 +53,7 @@ class TestWrapCommandAdvertisesHarness:env._snapshot_ready = Falseenv._session_id = "testsession0"env._cwd_marker = "__HERMES_CWD_testsession0__"+ env._pipe_marker = "__HERMES_PIPE_testsession0__"env._snapshot_path = "/tmp/hermes-snap-testsession0.sh"env._snapshot_passthrough_names = set()return env._wrap_command(command, "/tmp")diff --git a/tests/tools/test_docker_environment.py b/tests/tools/test_docker_environment.pyindex 1cc01ca8ac4..0d1418f951b 100644--- a/tests/tools/test_docker_environment.py+++ b/tests/tools/test_docker_environment.py@@ -115,6 +115,7 @@ def _make_execute_only_env(forward_env=None):env._snapshot_path = "/tmp/hermes-snap-test123.sh"env._cwd_file = "/tmp/hermes-cwd-test123.txt"env._cwd_marker = "__HERMES_CWD_test123__"+ env._pipe_marker = "__HERMES_PIPE_test123__"env._snapshot_ready = Trueenv._last_sync_time = Noneenv._init_env_args = []diff --git a/tests/tools/test_terminal_pipeline_status.py b/tests/tools/test_terminal_pipeline_status.pyindex e0b5d43b37d..c236869cb6e 100644--- a/tests/tools/test_terminal_pipeline_status.py+++ b/tests/tools/test_terminal_pipeline_status.py@@ -2,6 +2,8 @@import subprocess+import pytest+from tools.environments.base_session_env import _wrap_command_scriptCWD_MARKER = "__HERMES_CWD_test__"@@ -16,8 +18,12 @@ def _script(command: str) -> str:)+def _run_result(command: str) -> subprocess.CompletedProcess:+ return subprocess.run(["bash", "-c", _script(command)], capture_output=True, text=True, check=False)++def _run(command: str) -> str:- return subprocess.run(["bash", "-c", _script(command)], capture_output=True, text=True, check=False).stdout+ return _run_result(command).stdoutdef test_pipeline_status_captures_the_leading_command():@@ -32,8 +38,33 @@ def test_plain_command_reports_one_status():assert f"{PIPE_MARKER}1{PIPE_MARKER}" in output-def test_wrapper_captures_pipestatus_before_anything_else():- script = _script("echo hi")+ ("command", "expected"),+ [+ ("if false; then echo yes; fi", 0),+ ("while read line; do echo \"$line\"; done < /dev/null", 0),+ ("set -o pipefail; false | true", 1),+ ("! false", 0),+ ("! true", 1),+ ("echo 'unbalanced", 2),+ ("echo hi \\", 0),+ ("exit 3", 3),+ ],+)+def test_wrapper_preserves_shell_exit_code(command, expected):+ """``PIPESTATUS`` is side data: the wrapper must return the shell's own ``$?``."""+ assert _run_result(command).returncode == expected+++def test_wrapper_does_not_leak_internals_on_syntax_error():+ result = _run_result("echo 'unbalanced")++ assert "__hermes_ps" not in result.stderr+++def test_wrapper_does_not_leak_internals_after_a_trailing_backslash():+ result = _run_result("echo hi \\")- assert script.index('__hermes_ps=("${PIPESTATUS[@]}")') < script.index("umask 077")- assert script.index("__hermes_ec=${__hermes_ps[${#__hermes_ps[@]}-1]}") < script.index(PIPE_MARKER)+ assert result.returncode == 0+ assert "__hermes_ps" not in result.stdout+ assert "__hermes_ps" not in result.stderrdiff --git a/tools/environments/base_session_env.py b/tools/environments/base_session_env.pyindex ac94acc3ad3..57d2785931d 100644--- a/tools/environments/base_session_env.py+++ b/tools/environments/base_session_env.py@@ -169,8 +169,8 @@ def _wrap_command_script('export GIT_PAGER="${GIT_PAGER:-cat}" PAGER="${PAGER:-cat}"',# ``--`` keeps hyphen-prefixed directory names from being parsed as options.f"builtin cd -- {quoted_cwd} || exit 126",- f"eval '{escaped}\n__hermes_ps=(\"${{PIPESTATUS[@]}}\")'",- "__hermes_ec=${__hermes_ps[${#__hermes_ps[@]}-1]}",+ f"eval '{escaped}\n\n__hermes_rc=$? __hermes_ps=(\"${{PIPESTATUS[@]}}\"); (exit $__hermes_rc)'",+ "__hermes_ec=$?","umask 077"]if snapshot_ready and persist_env:parts.append(