fix(agents): address sync review debt

.github/scripts/issue_format.py

@@ -199,6 +199,33 @@ def _section_text(body: str, start: int) -> str:
     return "\n".join(lines[start + 1 : end])
 
 
+def _without_fenced_code(text: str) -> str:
+    """Remove Markdown fences so examples cannot satisfy issue requirements."""
+    kept: list[str] = []
+    fence: tuple[str, int] | None = None
+    for line in text.splitlines():
+        match = re.match(r"\s{0,3}(`{3,}|~{3,})", line)
+        if match:
+            marker = match.group(1)
+            if fence is None:
+                fence = (marker[0], len(marker))
+            elif (
+                marker[0] == fence[0]
+                and len(marker) >= fence[1]
+                and re.fullmatch(
+                    rf"\s{{0,3}}(?:`{{{fence[1]},}}|~{{{fence[1]},}})\s*",
+                    line,
+                )
+            ):
+                # Closing fences are marker-only (optional whitespace); trailing
+                # content such as a language tag must not end the fence.
+                fence = None
+            continue
+        if fence is None:
+            kept.append(line)
+    return "\n".join(kept)
+
+
 @dataclass
 class Report:
     ok: bool = True
@@ -247,7 +274,9 @@ def validate(body: str) -> Report:
     tasks_at = _find(body, REQUIRED["Tasks"])
     if tasks_at is not None:
         task_items = re.findall(
-            r"^\s*[-*]\s*\[[ xX]\]\s*(.+)$", _section_text(body, tasks_at), re.M
+            r"^\s*[-*]\s*\[[ xX]\]\s*(.+)$",
+            _without_fenced_code(_section_text(body, tasks_at)),
+            re.M,
         )
         if not task_items:
             report.problems.append(
@@ -260,7 +289,7 @@ def validate(body: str) -> Report:
 
     acceptance_at = _find(body, REQUIRED["Acceptance Criteria"])
     if acceptance_at is not None:
-        acceptance = _section_text(body, acceptance_at)
+        acceptance = _without_fenced_code(_section_text(body, acceptance_at))
         if not GATE.search(acceptance):
             report.problems.append(
                 "`Acceptance Criteria` names no test, runnable command or observable "

.github/workflows/agents-63-issue-intake.yml

@@ -146,7 +146,8 @@ jobs:
           )
         )
       ) &&
-      !contains(github.event.issue.labels.*.name, 'agents:auto-pilot')
+      (github.event_name != 'issues' ||
+      !contains(github.event.issue.labels.*.name, 'agents:auto-pilot'))
     name: Normalize workflow inputs
     runs-on: ubuntu-latest
     outputs:

config/template-drift-allowlist.txt

@@ -43,9 +43,9 @@
 [pair.1]
 main = .github/workflows/agents-63-issue-intake.yml
 template = templates/consumer-repo/.github/workflows/agents-issue-intake.yml
-main_sha256 = e0ded7e7b98ff8abe3cb60b69c40bd8472d4c4d857d037540d52aefcc332dd7a
-template_sha256 = 1381165cbf31a6fbe632bd9667337b7ac042d777af8965fc155f8351b66a233a
-reason = Intentional divergence re-baselined 2026-08-08: both intake surfaces now reject closed issue label events before bridge/PR creation. Root remains the numeric-prefixed Workflows-internal superset while the consumer template retains its pinned action and minimal bridge contract.
+main_sha256 = f63f28d047db415dac9036ead0dc111a55ae999179fdb44cb458c8a39d0e6c67
+template_sha256 = 617cb594296c26ef8510b03dc43e835396a50ce84ed0d8a10d9e41b3e93f6772
+reason = Intentional divergence re-baselined 2026-08-09: both intake surfaces keep the auto-pilot skip only for issue events so workflow_dispatch remains independent of issue payload labels. Root remains the numeric-prefixed Workflows-internal superset while the consumer template retains its pinned action and minimal bridge contract.
 
 [pair.2]
 main = .github/workflows/agents-71-codex-belt-dispatcher.yml

langsmith-fleet-worker-attempt.json

@@ -1,13 +1,13 @@
 {
   "agent": "codex",
   "cli_version": "0.144.1",
-  "emitted_at": "2026-08-09T07:45:12.662386Z",
+  "emitted_at": "2026-08-09T11:43:24.687070Z",
   "execution_profile": "codex-default",
   "fallback_models": [
     "gpt-5.5"
   ],
   "operation_role": "worker",
-  "pr_number": "3004",
+  "pr_number": "3005",
   "requested_model": "gpt-5.6-terra",
   "runner": "reusable-codex-run",
   "schema": "langsmith-fleet/v1",

scripts/langchain/issue_formatter.py

@@ -170,11 +170,11 @@ def _strip_reuse_marker(text: str) -> str:
 VERIFY_HINT_REGEX = re.compile(r"\(verify:\s*([^\n)]+)\)", re.IGNORECASE)
 SAFE_VERIFY_COMMAND_RE = re.compile(
     r"^(?:"
-    r"(?:python(?:3)?\s+-m\s+)?pytest\b"
+    r"(?:python(?:3)?\s+-m\s+)?(?:pytest|unittest)\b"
     r"|node\s+--test\b"
     r"|(?:npm|pnpm|yarn)\s+(?:run\s+)?(?:test|vitest|jest|playwright)\b"
     r"|(?:make|just|cargo|go|dotnet)\s+(?:test|check)\b"
-    r"|gh\s+(?:workflow\s+run|run)\b"
+    r"|gh\s+(?:workflow\s+run|run)\s+\S+"
     r"|curl\s+\S+"
     r")",
     re.IGNORECASE,
@@ -477,10 +477,34 @@ def _strip_original_issue_blocks(text: str) -> str:
         kept.append(text[cursor : match.start()])
         depth = 1
         end = match.end()
+        fence: tuple[str, int] | None = None
         for tag in _DETAILS_TAG_RE.finditer(text, match.end()):
+            # Literal HTML in the verbatim Original-Issue fence is content, not
+            # structural markup.  Only count tags outside Markdown fences.
+            before = text[end : tag.start()]
+            for line in before.splitlines():
+                fence_match = re.match(r"\s{0,3}(`{3,}|~{3,})", line)
+                if not fence_match:
+                    continue
+                marker = fence_match.group(1)
+                if fence is None:
+                    fence = (marker[0], len(marker))
+                elif (
+                    marker[0] == fence[0]
+                    and len(marker) >= fence[1]
+                    and re.fullmatch(
+                        rf"\s{{0,3}}(?:`{{{fence[1]},}}|~{{{fence[1]},}})\s*",
+                        line,
+                    )
+                ):
+                    # Closing fences are marker-only; language tags / trailing
+                    # text must not toggle the fence state.
+                    fence = None
+            end = tag.end()
+            if fence is not None:
+                continue
             depth += -1 if tag.group(0).startswith("</") else 1
             if depth == 0:
-                end = tag.end()
                 break
         else:
             # Leave malformed markup intact rather than silently discarding it.

templates/consumer-repo/.github/scripts/issue_format.py

@@ -150,7 +150,6 @@ def _task_has_concrete_target(item: str) -> bool:
         return True
     # Command names in prose ("make the UI better", "go improve it") are not
     # concrete targets. Require a command-shaped invocation instead.
-    # Allow a leading backtick so `make test` / `npm test` count.
     return re.search(rf"(?:^|[\s`]){_TASK_COMMAND}", item, re.I) is not None
 
 
@@ -200,6 +199,33 @@ def _section_text(body: str, start: int) -> str:
     return "\n".join(lines[start + 1 : end])
 
 
+def _without_fenced_code(text: str) -> str:
+    """Remove Markdown fences so examples cannot satisfy issue requirements."""
+    kept: list[str] = []
+    fence: tuple[str, int] | None = None
+    for line in text.splitlines():
+        match = re.match(r"\s{0,3}(`{3,}|~{3,})", line)
+        if match:
+            marker = match.group(1)
+            if fence is None:
+                fence = (marker[0], len(marker))
+            elif (
+                marker[0] == fence[0]
+                and len(marker) >= fence[1]
+                and re.fullmatch(
+                    rf"\s{{0,3}}(?:`{{{fence[1]},}}|~{{{fence[1]},}})\s*",
+                    line,
+                )
+            ):
+                # Closing fences are marker-only (optional whitespace); trailing
+                # content such as a language tag must not end the fence.
+                fence = None
+            continue
+        if fence is None:
+            kept.append(line)
+    return "\n".join(kept)
+
+
 @dataclass
 class Report:
     ok: bool = True
@@ -248,7 +274,9 @@ def validate(body: str) -> Report:
     tasks_at = _find(body, REQUIRED["Tasks"])
     if tasks_at is not None:
         task_items = re.findall(
-            r"^\s*[-*]\s*\[[ xX]\]\s*(.+)$", _section_text(body, tasks_at), re.M
+            r"^\s*[-*]\s*\[[ xX]\]\s*(.+)$",
+            _without_fenced_code(_section_text(body, tasks_at)),
+            re.M,
         )
         if not task_items:
             report.problems.append(
@@ -261,7 +289,7 @@ def validate(body: str) -> Report:
 
     acceptance_at = _find(body, REQUIRED["Acceptance Criteria"])
     if acceptance_at is not None:
-        acceptance = _section_text(body, acceptance_at)
+        acceptance = _without_fenced_code(_section_text(body, acceptance_at))
         if not GATE.search(acceptance):
             report.problems.append(
                 "`Acceptance Criteria` names no test, runnable command or observable "

templates/consumer-repo/.github/workflows/agents-issue-intake.yml

@@ -114,7 +114,8 @@ jobs:
        (github.event.issue.state != 'closed' &&
         (contains(toJson(github.event.issue.labels.*.name), 'agent:') ||
          contains(toJson(github.event.issue.labels.*.name), 'agents:')))) &&
-      !contains(github.event.issue.labels.*.name, 'agents:auto-pilot')
+      (github.event_name != 'issues' ||
+      !contains(github.event.issue.labels.*.name, 'agents:auto-pilot'))
     runs-on: ubuntu-latest
     outputs:
       should_run: ${{ steps.check.outputs.should_run }}

tests/scripts/test_issue_format.py

@@ -40,6 +40,27 @@ def test_fence_with_language_marker_does_not_close_a_code_block() -> None:
     assert "Acceptance Criteria" in report.missing_required
 
 
+@pytest.mark.parametrize("marker", ["```", "~~~"])
+def test_fence_close_requires_marker_only_line(marker: str) -> None:
+    """Trailing text after fence markers must not end the fenced region."""
+    validator = _validator()
+    close_with_junk = f"{marker} not-a-close"
+    body = (
+        f"{marker}markdown\n"
+        "## Tasks\n- [ ] pretend\n"
+        f"{close_with_junk}\n"
+        "## Acceptance Criteria\n- pytest tests/test_x.py\n"
+        f"{marker}\n"
+    )
+    report = validator.validate(body)
+    assert not report.ok
+    assert set(report.missing_required) >= {"Tasks", "Acceptance Criteria"}
+    # Section body stripping must keep pretending the junk line is still inside.
+    stripped = validator._without_fenced_code(body)
+    assert "Acceptance Criteria" not in stripped
+    assert "pretend" not in stripped
+
+
 def test_heading_with_trailing_qualifier_matches_required_section() -> None:
     validator = _validator()
     report = validator.validate(

tests/scripts/test_issue_formatter.py

@@ -678,7 +678,7 @@ def test_strip_original_issue_blocks_removes_balanced_nested_details() -> None:
 <details>
 <summary>Original Issue</summary>
 
-```text
+````text
 outer
 <details>
 <summary>Original Issue</summary>
@@ -687,7 +687,7 @@ def test_strip_original_issue_blocks_removes_balanced_nested_details() -> None:
 inner
 ```
 </details>
-```
+````
 </details>
 
 ## Scope
@@ -703,6 +703,43 @@ def test_strip_original_issue_blocks_removes_balanced_nested_details() -> None:
     assert "</details>" not in stripped
 
 
+@pytest.mark.parametrize("marker", ["```", "~~~"])
+def test_strip_original_issue_ignores_details_when_fence_close_has_trailing_text(
+    marker: str,
+) -> None:
+    """A fence line with trailing junk must not resume HTML details counting."""
+    payload = "\n".join(
+        [
+            "## Why",
+            "",
+            "Keep this text.",
+            "",
+            "<details>",
+            "<summary>Original Issue</summary>",
+            "",
+            f"{marker}text",
+            "literal <details>",
+            f"{marker} still-inside",
+            "literal </details>",
+            marker,
+            "</details>",
+            "",
+            "## Scope",
+            "",
+            "Keep this too.",
+            "",
+        ]
+    )
+
+    stripped = issue_formatter._strip_original_issue_blocks(payload)
+
+    assert "Keep this text." in stripped
+    assert "Keep this too." in stripped
+    assert "Original Issue" not in stripped
+    assert "literal <details>" not in stripped
+    assert "</details>" not in stripped
+
+
 def test_reuse_sets_needs_refinement_when_validator_fails() -> None:
     """Structurally conformant but contract-invalid bodies must not claim ready."""
     conformant = "\n".join(
@@ -750,3 +787,17 @@ def test_bare_curl_is_not_a_safe_verify_command() -> None:
     assert (
         issue_formatter.SAFE_VERIFY_COMMAND_RE.match("curl https://example.test/health") is not None
     )
+
+
+def test_safe_verify_command_accepts_unittest_and_requires_gh_args() -> None:
+    assert (
+        issue_formatter.SAFE_VERIFY_COMMAND_RE.match("python -m unittest tests.test_x") is not None
+    )
+    assert issue_formatter.SAFE_VERIFY_COMMAND_RE.match("unittest") is not None
+    assert issue_formatter.SAFE_VERIFY_COMMAND_RE.match("gh run") is None
+    assert issue_formatter.SAFE_VERIFY_COMMAND_RE.match("gh workflow run") is None
+    assert issue_formatter.SAFE_VERIFY_COMMAND_RE.match("gh run watch") is not None
+    assert (
+        issue_formatter.SAFE_VERIFY_COMMAND_RE.match("gh workflow run agents-issue-optimizer.yml")
+        is not None
+    )