maintainability-agent 3.7.6__tar.gz → 3.7.8__tar.gz

{maintainability_agent-3.7.6/src/maintainability_agent.egg-info → maintainability_agent-3.7.8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: maintainability-agent
-Version: 3.7.6
+Version: 3.7.8
 Summary: Maintainability gate and AI remediation prompt generator for CI.
 Author: Marshall Guillory
 License: MIT
@@ -46,7 +46,7 @@ Dynamic: license-file
 **A deterministic, offline maintainability audit whose output is a _bounded
 work order_ for an AI coding agent** — a copy-paste prompt, per finding, that
 says *fix exactly these and refactor nothing else*. Chat-primary; CLI for CI.
-Version **3.7.6**.
+Version **3.7.8**.
 
 **Languages parsed:** Python, Java, C, C++, C#, Go, Rust, PHP, Ruby, Swift,
 COBOL, Fortran (free-form *and* fixed-form), and the JS/TS/HTML family — each
@@ -424,13 +424,13 @@ it was generated against — a provenance record, not a claim about HEAD.
 
 | Metric | Value |
 |---|---:|
-| Maintainability estimate | 4.5 / 5 |
+| Maintainability estimate | 4.4 / 5 |
 | Verified grade | B |
-| Files scanned | 511 |
+| Files scanned | 498 |
 | Hard gate failures | 0 |
 
 A **B**, and the report says why: the grade is verified against the evidence
-*floor* of 4.2 rather than the 4.5 point estimate, because an unmeasured aspect
+*floor* of 4.1 rather than the 4.4 point estimate, because an unmeasured aspect
 prices at 0 when a grade has to be defended. Every threshold gate is opted
 **on** for this repository's own CI, so drifting below the bar fails the build
 rather than the README.
@@ -473,7 +473,7 @@ instead of an invokable skill:
 This repo ships `action.yml`, usable as a composite action:
 
 ```yaml
-- uses: marshallguillory86/maintainability-agent@v1.0.0
+- uses: marshallguillory86/maintainability-agent@v3.7.8
   with:
     config: maintainability-agent.json
     changed-only: main...HEAD

{maintainability_agent-3.7.6 → maintainability_agent-3.7.8}/README.md

@@ -5,7 +5,7 @@
 **A deterministic, offline maintainability audit whose output is a _bounded
 work order_ for an AI coding agent** — a copy-paste prompt, per finding, that
 says *fix exactly these and refactor nothing else*. Chat-primary; CLI for CI.
-Version **3.7.6**.
+Version **3.7.8**.
 
 **Languages parsed:** Python, Java, C, C++, C#, Go, Rust, PHP, Ruby, Swift,
 COBOL, Fortran (free-form *and* fixed-form), and the JS/TS/HTML family — each
@@ -383,13 +383,13 @@ it was generated against — a provenance record, not a claim about HEAD.
 
 | Metric | Value |
 |---|---:|
-| Maintainability estimate | 4.5 / 5 |
+| Maintainability estimate | 4.4 / 5 |
 | Verified grade | B |
-| Files scanned | 511 |
+| Files scanned | 498 |
 | Hard gate failures | 0 |
 
 A **B**, and the report says why: the grade is verified against the evidence
-*floor* of 4.2 rather than the 4.5 point estimate, because an unmeasured aspect
+*floor* of 4.1 rather than the 4.4 point estimate, because an unmeasured aspect
 prices at 0 when a grade has to be defended. Every threshold gate is opted
 **on** for this repository's own CI, so drifting below the bar fails the build
 rather than the README.
@@ -432,7 +432,7 @@ instead of an invokable skill:
 This repo ships `action.yml`, usable as a composite action:
 
 ```yaml
-- uses: marshallguillory86/maintainability-agent@v1.0.0
+- uses: marshallguillory86/maintainability-agent@v3.7.8
   with:
     config: maintainability-agent.json
     changed-only: main...HEAD

{maintainability_agent-3.7.6 → maintainability_agent-3.7.8}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "maintainability-agent"
-version = "3.7.6"
+version = "3.7.8"
 description = "Maintainability gate and AI remediation prompt generator for CI."
 readme = "README.md"
 # 3.11, not 3.10. `_discovery`, `_pillars` and `_runner` all import

{maintainability_agent-3.7.6 → maintainability_agent-3.7.8/src/maintainability_agent.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: maintainability-agent
-Version: 3.7.6
+Version: 3.7.8
 Summary: Maintainability gate and AI remediation prompt generator for CI.
 Author: Marshall Guillory
 License: MIT
@@ -46,7 +46,7 @@ Dynamic: license-file
 **A deterministic, offline maintainability audit whose output is a _bounded
 work order_ for an AI coding agent** — a copy-paste prompt, per finding, that
 says *fix exactly these and refactor nothing else*. Chat-primary; CLI for CI.
-Version **3.7.6**.
+Version **3.7.8**.
 
 **Languages parsed:** Python, Java, C, C++, C#, Go, Rust, PHP, Ruby, Swift,
 COBOL, Fortran (free-form *and* fixed-form), and the JS/TS/HTML family — each
@@ -424,13 +424,13 @@ it was generated against — a provenance record, not a claim about HEAD.
 
 | Metric | Value |
 |---|---:|
-| Maintainability estimate | 4.5 / 5 |
+| Maintainability estimate | 4.4 / 5 |
 | Verified grade | B |
-| Files scanned | 511 |
+| Files scanned | 498 |
 | Hard gate failures | 0 |
 
 A **B**, and the report says why: the grade is verified against the evidence
-*floor* of 4.2 rather than the 4.5 point estimate, because an unmeasured aspect
+*floor* of 4.1 rather than the 4.4 point estimate, because an unmeasured aspect
 prices at 0 when a grade has to be defended. Every threshold gate is opted
 **on** for this repository's own CI, so drifting below the bar fails the build
 rather than the README.
@@ -473,7 +473,7 @@ instead of an invokable skill:
 This repo ships `action.yml`, usable as a composite action:
 
 ```yaml
-- uses: marshallguillory86/maintainability-agent@v1.0.0
+- uses: marshallguillory86/maintainability-agent@v3.7.8
   with:
     config: maintainability-agent.json
     changed-only: main...HEAD

{maintainability_agent-3.7.6 → maintainability_agent-3.7.8}/src/maintainability_agent.egg-info/SOURCES.txt

@@ -203,11 +203,13 @@ tests/test_claimed_languages.py
 tests/test_cli.py
 tests/test_clone_group_class.py
 tests/test_cobol_declarations.py
+tests/test_code_scanning_gets_alerts_not_the_record.py
 tests/test_cognitive_complexity.py
 tests/test_cohort_statistics.py
 tests/test_concealment_properties.py
 tests/test_config_shape.py
 tests/test_consent_and_grant.py
+tests/test_consent_without_command_has_a_remedy.py
 tests/test_consumer_migration.py
 tests/test_contract_claims.py
 tests/test_corroborate.py
@@ -224,6 +226,7 @@ tests/test_declaration_grading.py
 tests/test_declaration_ranges.py
 tests/test_declaration_source.py
 tests/test_declarative_parsers.py
+tests/test_declared_capabilities_match_the_delegate.py
 tests/test_declared_imports.py
 tests/test_delegated_security_pillar.py
 tests/test_demo_tree.py
@@ -344,6 +347,7 @@ tests/test_scope_conformance.py
 tests/test_scoring_calibration.py
 tests/test_security_delegate_versions_agree.py
 tests/test_security_findings_keep_their_handoff.py
+tests/test_security_pillar_asserts_a_scanner_set.py
 tests/test_security_pillar_runs_and_shows.py
 tests/test_semantic_findings_stay_inside_the_audited_files.py
 tests/test_semantic_policy.py

{maintainability_agent-3.7.6 → maintainability_agent-3.7.8}/src/maintainability_audit/__init__.py

@@ -1,3 +1,3 @@
 """Maintainability Audit CI."""
 
-__version__ = "3.7.6"
+__version__ = "3.7.8"

{maintainability_agent-3.7.6 → maintainability_agent-3.7.8}/src/maintainability_audit/_delegated_pillar.py

@@ -23,10 +23,12 @@ number: a delegated pillar's numbers belong to the tool that measured
 them, and a consumer that recomputed anything would be a second opinion
 nobody asked for.
 
-**Absence is the normal case and stays the documented one.** No file
-means the pillar reports exactly what it reported before — delegated,
-`NotApplicable`, naming the other tool. That path is not a failure and
-does not warn: most repositories run one tool.
+**Absence is stated, never silent.** Since D177 every audit runs
+secure-code-agent itself; this reader serves `--security-pillar PATH`,
+where a pipeline hands over a document the delegate already wrote. A
+document that is missing or cannot be trusted yields `None`, and the
+caller records why in the pillar entry rather than reporting an empty
+pillar as a clean one.
 
 **A file that cannot be trusted is refused rather than partly believed.**
 Wrong schema, unreadable JSON, a shape that is not an object — each
@@ -108,7 +110,8 @@ def read_delegated(root: Path, relative: str = DEFAULT_PILLAR_PATH) -> dict[str,
     the file is absent, it is not a regular file, it does not parse, it
     is not an object, or it announces a schema this tool does not know.
     All five collapse to one answer because the caller does the same
-    thing with each — keep the placeholder that names the other tool.
+    thing with each — it states that the handed-over document was absent
+    or could not be trusted.
 
     Read through the regular-file primitive like every other path the
     audited tree chooses (D145): a FIFO here would hang the audit rather

{maintainability_agent-3.7.6 → maintainability_agent-3.7.8}/src/maintainability_audit/_mcp_gate.py

@@ -18,6 +18,7 @@ from pathlib import Path
 from typing import Any
 
 from ._mcp_setup import run_tests_pending, setup_pending, setup_questions
+from ._test_execution import consented_without_command
 from ._user_config import mark_repo_seen
 from .config import VERSION, load_config
 
@@ -100,6 +101,16 @@ def _choose_next(root: Path) -> dict[str, Any]:
             "own test suite for a coverage reading (test effectiveness). "
             "Choose reconfigure to set it up."
         )
+    elif consented_without_command():
+        # The opt-in was answered yes and the command never reached the
+        # person's tier, so `run_tests_pending` reads it as asked-and-done
+        # and said nothing. The suite then does not run on any audit, and
+        # the report explained the refusal without naming a remedy.
+        prompt += (
+            " Note: this host is opted in to running the test suite, but no "
+            "test command is recorded for it, so no suite runs. Choose "
+            "reconfigure to record the command."
+        )
     result["choice_needed"] = {
         "name": "next_action",
         "prompt": prompt,

{maintainability_agent-3.7.6 → maintainability_agent-3.7.8}/src/maintainability_audit/_security_delegate.py

@@ -61,9 +61,17 @@ from ._runner import Invocation, Outcome, run
 #: `paths:` globs against absolute finding paths. This audit always hands it an
 #: absolute root, so under 0.12.1 every reviewed `paths:` suppression is ignored
 #: and its finding counts as live in the pillar every skin shows (D179).
-SUPPORTED_FLOOR = (0, 12, 2)
+#:
+#: **Raised to 0.12.7 by D186.** The delegate config now declares what this
+#: tool is — `capabilities` — and secure-code-agent sweeps unknown top-level
+#: keys strictly, so anything below 0.12.6 refuses the configuration outright
+#: rather than ignoring the key: the audit would not run at all. 0.12.6 runs
+#: it and prints no reason under the declared axes (its D25), which is a grade
+#: moved in silence. The floor is the first release where the declaration is
+#: both accepted and disclosed.
+SUPPORTED_FLOOR = (0, 12, 7)
 SUPPORTED_CEILING = (1,)
-REQUIREMENT = "secure-code-agent>=0.12.2,<1"
+REQUIREMENT = "secure-code-agent>=0.12.7,<1"
 
 #: How long the child may run before the pillar is reported as unmeasured.
 DEFAULT_TIMEOUT_SECONDS = 300

{maintainability_agent-3.7.6 → maintainability_agent-3.7.8}/src/maintainability_audit/_test_execution.py

@@ -113,6 +113,25 @@ def opted_in_command() -> list[str]:
     return list(command) if isinstance(command, list) else []
 
 
+def consented_without_command() -> bool:
+    """Opted in to running the suite, with no command in the person's tier.
+
+    `run_tests_pending` covers the config written before the opt-in existed:
+    no `test_execution` key at all, so the question was never asked. This
+    covers the state after it — the answer is recorded, `requested` is true,
+    and `expected_commands.test` never reached the user tier because setup
+    captured the consent without capturing the program.
+
+    Neither surface said anything about it. The report explained why nothing
+    ran and named no remedy, and the MCP gate treats a present `test_execution`
+    key as asked-and-answered, so it offered no discovery line either. A person
+    who said yes got silence on every door.
+    """
+    answers = user_config_answers() or {}
+    requested = bool((answers.get("test_execution") or {}).get("requested"))
+    return requested and not opted_in_command()
+
+
 def _bounded_suite_timeout(configured: Any) -> int:
     """A suite timeout the audited tree cannot weaponise.
 
@@ -153,7 +172,9 @@ def run_test_suite(root: Path, config: dict[str, Any]) -> dict[str, Any] | None:
             "ran": False, "exit_code": None, "passed": False,
             "detail": (
                 "opted in, but no test command is recorded in the user tier; "
-                "the repository's documented command is not run on its own say-so"
+                "the repository's documented command is not run on its own say-so. "
+                "Reconfigure to record the command you consent to, or add it to "
+                "expected_commands.test in the user configuration"
             ),
             "coverage_percent": None,
         }

{maintainability_agent-3.7.6 → maintainability_agent-3.7.8}/src/maintainability_audit/config.py

@@ -20,7 +20,7 @@ from ._operator_reads import (
 # still the door every caller comes through.
 __all__ = ["DEFAULT_CONFIG", "DEFAULT_IDIOM_GROUPS", "PROJECT_URL", "VERSION"]
 
-VERSION = "3.7.6"
+VERSION = "3.7.8"
 
 PROJECT_URL = "https://github.com/marshallguillory86/maintainability-agent"
 

{maintainability_agent-3.7.6 → maintainability_agent-3.7.8}/tests/test_authorship_gates.py

@@ -77,19 +77,32 @@ def _step_scripts() -> dict[str, str]:
     assert scripts.keys() == {
         "Each commit carries the repository's commit identity",
         "Each commit is signed",
+        "Commit messages name only the commit identity",
+        "No forbidden term in the change",
     }
     return scripts
 
 
 SCRIPTS = _step_scripts()
 
+#: Stands in for the FORBIDDEN_TERMS secret. A made-up term, deliberately:
+#: the real terms live only in the secret and must never be written here.
+EXAMPLE_TERM = "example-forbidden-term"
 
-def _run_gate(repo: Path, name: str) -> subprocess.CompletedProcess[str]:
-    """Substitute the only Actions expression used by the checked script."""
+
+def _run_gate(
+    repo: Path, name: str, *, terms: str = EXAMPLE_TERM,
+) -> subprocess.CompletedProcess[str]:
+    """Substitute the only Actions expression used by the checked script.
+
+    ``terms`` is what the secret would hold; an empty string is the secret
+    being unset.
+    """
     script = SCRIPTS[name].replace("${{ github.base_ref }}", "main")
     return subprocess.run(
         ["bash", "-c", script],
         cwd=repo,
+        env={**os.environ, "FORBIDDEN_TERMS": terms},
         text=True,
         capture_output=True,
         check=False,
@@ -131,10 +144,13 @@ def signed_repo(tmp_path: Path) -> tuple[Path, Callable[..., str]]:
         signed: bool = True,
         author: str = COMMIT_IDENTITY,
         committer: str = COMMIT_IDENTITY,
+        message: str | None = None,
+        content: str | None = None,
     ) -> str:
         nonlocal counter
         counter += 1
-        (repo / f"change-{counter}").write_text(f"{counter}\n", encoding="utf-8")
+        (repo / f"change-{counter}").write_text(
+            content if content is not None else f"{counter}\n", encoding="utf-8")
         _require_git(repo, "add", ".")
         result = _git(
             repo,
@@ -147,7 +163,7 @@ def signed_repo(tmp_path: Path) -> tuple[Path, Callable[..., str]]:
             "--author",
             f"Marshall Guillory <{author}>",
             "-m",
-            f"change {counter}",
+            message if message is not None else f"change {counter}",
             env={**os.environ, "GIT_IDENTITY_OVERRIDE": "1"}
             if committer != COMMIT_IDENTITY
             else None,
@@ -325,3 +341,86 @@ def test_the_signature_gate_still_checks_merge_commits() -> None:
         "the conflict resolution its author wrote; rebase a stale branch "
         "rather than weakening this to make 'Update branch' pass (D105)."
     )
+
+
+def test_a_trailer_naming_another_address_fails(
+    signed_repo: tuple[Path, Callable[..., str]],
+) -> None:
+    """A co-author trailer is where an outside address reached history before.
+
+    The identity step reads author and committer only, so a trailer in the
+    message passed every check and landed on the protected branch.
+    """
+    repo, commit = signed_repo
+    commit(message="change\n\nCo-authored-by: Someone Else <someone@example.com>")
+
+    result = _run_gate(repo, "Commit messages name only the commit identity")
+
+    assert result.returncode == 1, result.stdout
+    assert "names an address other than the commit identity" in result.stdout
+
+
+def test_a_message_naming_only_the_commit_identity_passes(
+    signed_repo: tuple[Path, Callable[..., str]],
+) -> None:
+    """The allowlist is the identity itself, so naming it is not a failure."""
+    repo, commit = signed_repo
+    commit(message=f"change\n\nCo-authored-by: Marshall Guillory <{COMMIT_IDENTITY}>")
+
+    result = _run_gate(repo, "Commit messages name only the commit identity")
+
+    assert result.returncode == 0, result.stdout
+
+
+def test_a_forbidden_term_in_a_message_fails_in_any_case(
+    signed_repo: tuple[Path, Callable[..., str]],
+) -> None:
+    repo, commit = signed_repo
+    commit(message=f"change mentioning {EXAMPLE_TERM.upper()}")
+
+    result = _run_gate(repo, "No forbidden term in the change")
+
+    assert result.returncode == 1, result.stdout
+    assert "identity or message contains a forbidden term" in result.stdout
+
+
+def test_a_forbidden_term_added_to_a_file_fails_without_printing_it(
+    signed_repo: tuple[Path, Callable[..., str]],
+) -> None:
+    """The failure names the file, never the matched text.
+
+    A check that echoed the term into a public CI log would publish the
+    thing it exists to keep out.
+    """
+    repo, commit = signed_repo
+    commit(content=f"a line carrying {EXAMPLE_TERM}\n")
+
+    result = _run_gate(repo, "No forbidden term in the change")
+
+    assert result.returncode == 1, result.stdout
+    assert "adds a forbidden term" in result.stdout
+    assert EXAMPLE_TERM not in (result.stdout + result.stderr).lower()
+
+
+def test_a_clean_change_passes_the_forbidden_term_check(
+    signed_repo: tuple[Path, Callable[..., str]],
+) -> None:
+    repo, commit = signed_repo
+    commit()
+
+    result = _run_gate(repo, "No forbidden term in the change")
+
+    assert result.returncode == 0, result.stdout
+
+
+def test_the_forbidden_term_check_fails_closed_without_the_secret(
+    signed_repo: tuple[Path, Callable[..., str]],
+) -> None:
+    """An unset secret must not read as a clean change."""
+    repo, commit = signed_repo
+    commit()
+
+    result = _run_gate(repo, "No forbidden term in the change", terms="")
+
+    assert result.returncode == 1
+    assert "this check proved nothing" in result.stdout

maintainability_agent-3.7.8/tests/test_code_scanning_gets_alerts_not_the_record.py

@@ -0,0 +1,83 @@
+"""What this repository uploads to code scanning is what should become an alert (D187).
+
+secure-code-agent writes two SARIF documents (its D22). `--sarif-output` is the
+record: every result, a suppressed one carrying its `suppressions` entry and
+reason. `--code-scanning-sarif-output` is the same document without the
+suppressed results — the test tree, documentation, and every reviewed
+`.scignore.yaml` entry.
+
+The difference matters because code scanning opens an alert for every result it
+is handed, `suppressions` array or not, and this repository's branch protection
+requires every resulting review thread resolved before a merge. Uploading the
+record put seventeen threads on one pull request, all of them `assert` in a
+pytest file, and blocked it. That is not a one-off: it is the outcome for every
+pull request that adds a test, forever, and the shape secure-code-agent already
+fixed upstream and this repository had not adopted.
+
+Nothing is hidden by the change. The record is still written, still kept as a
+build artifact, and still carries every finding; so do the JSON report, the
+Markdown report and the work order. What changes is which document becomes
+somebody's blocking alert.
+"""
+
+from __future__ import annotations
+
+import re
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[1]
+WORKFLOWS = ROOT / ".github" / "workflows"
+
+#: The record. Never the thing uploaded to code scanning.
+_RECORD = "secure-code.sarif"
+#: The filtered document, which is.
+_FILTERED = "secure-code.code-scanning.sarif"
+
+
+def _uploads() -> list[tuple[str, str]]:
+    """Every `sarif_file:` value handed to the code-scanning upload action."""
+    found = []
+    for path in sorted(WORKFLOWS.glob("*.yml")):
+        for match in re.finditer(r"^\s*sarif_file:\s*(\S+)\s*$", path.read_text(encoding="utf-8"), re.MULTILINE):
+            found.append((path.name, match.group(1)))
+    return found
+
+
+def test_the_security_record_is_never_uploaded_to_code_scanning() -> None:
+    """The record carries the test tree; code scanning turns each one into a thread."""
+    uploads = _uploads()
+    assert uploads, "no workflow uploads SARIF; this sweep would pass vacuously"
+    record = [(name, value) for name, value in uploads if value == _RECORD]
+    assert not record, (
+        f"the full secure-code-agent record is uploaded to code scanning: {record}. "
+        "Every suppressed finding becomes an alert, and under this repository's "
+        "required-conversation-resolution policy, a blocking review thread."
+    )
+
+
+def test_the_filtered_document_is_what_is_uploaded() -> None:
+    """Proving the record is absent is not proving the right file is present."""
+    uploaded = {value for _, value in _uploads()}
+    assert _FILTERED in uploaded, (
+        f"no workflow uploads {_FILTERED}; the security findings that should "
+        "become alerts would reach code scanning from nowhere"
+    )
+
+
+def test_the_audit_asks_the_delegate_to_write_the_filtered_document() -> None:
+    """An upload step naming a file nothing writes fails the job, loudly but late."""
+    writing = [
+        path.name
+        for path in sorted(WORKFLOWS.glob("*.yml"))
+        if f"--code-scanning-sarif-output {_FILTERED}" in path.read_text(encoding="utf-8")
+    ]
+    assert writing, (
+        f"no workflow passes --code-scanning-sarif-output {_FILTERED}; the upload "
+        "step would name a file the security audit never produced"
+    )
+
+
+def test_the_filtered_document_is_not_committed() -> None:
+    """Generated per run, like every other audit output this repository writes."""
+    ignored = (ROOT / ".gitignore").read_text(encoding="utf-8").splitlines()
+    assert _FILTERED in ignored, f"{_FILTERED} is not in .gitignore"

maintainability_agent-3.7.8/tests/test_consent_without_command_has_a_remedy.py

@@ -0,0 +1,66 @@
+"""Consent to run the suite, with no command, must name a way out (D185).
+
+`test_execution.requested` true with no `expected_commands.test` in the user
+tier is a reachable state: setup recorded the answer without recording the
+program. In that state no suite runs on any audit.
+
+Neither surface said what to do about it. The report explained the refusal and
+stopped there, and the MCP gate treats a present `test_execution` key as
+asked-and-answered, so `run_tests_pending` is false and no discovery line is
+offered. A person who opted in got silence on every door, which is how this
+repository's own audit reported "it did not run" for weeks.
+"""
+from __future__ import annotations
+
+from pathlib import Path
+
+import pytest
+
+from maintainability_audit import _mcp_gate, _test_execution
+
+
+@pytest.fixture
+def consented(monkeypatch):
+    """User tier that opted in and carries no command."""
+    monkeypatch.setattr(
+        _test_execution, "user_config_answers",
+        lambda: {"test_execution": {"requested": True}})
+    return None
+
+
+def test_the_report_detail_names_a_remedy(consented, tmp_path):
+    """The operator is told how to fix it, not only why it happened."""
+    result = _test_execution.run_test_suite(
+        tmp_path,
+        {"test_execution": {"requested": True},
+         "expected_commands": {"test": ["pytest", "-q"]}})
+
+    assert result is not None and result["ran"] is False
+    detail = result["detail"].lower()
+    assert "reconfigure" in detail or "expected_commands" in detail, (
+        f"the refusal names no remedy: {result['detail']!r}")
+
+
+def test_the_predicate_sees_consent_without_a_command(consented):
+    """`run_tests_pending` cannot see this state; this is what does."""
+    assert _test_execution.consented_without_command() is True
+
+
+def test_a_recorded_command_is_not_reported_as_missing(monkeypatch):
+    """The predicate is about the gap, not about opting in."""
+    monkeypatch.setattr(
+        _test_execution, "user_config_answers",
+        lambda: {"test_execution": {"requested": True},
+                 "expected_commands": {"test": ["pytest", "-q"]}})
+    assert _test_execution.consented_without_command() is False
+
+
+def test_the_mcp_gate_offers_reconfigure_for_the_gap(consented, monkeypatch, tmp_path):
+    """The chat door surfaces it too, where `run_tests_pending` is false."""
+    monkeypatch.setattr(_mcp_gate, "run_tests_pending", lambda root: False)
+
+    result = _mcp_gate._choose_next(Path(tmp_path))
+
+    prompt = (result.get("choice_needed") or {}).get("prompt", "").lower()
+    assert "no test command is recorded" in prompt, (
+        f"the gate says nothing about the gap: {prompt!r}")