specfuse-loop 0.3.12__tar.gz → 0.3.13__tar.gz

{specfuse_loop-0.3.12/specfuse_loop.egg-info → specfuse_loop-0.3.13}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: specfuse-loop
-Version: 0.3.12
+Version: 0.3.13
 Summary: Local-first executor for the Specfuse Plan + Work Unit gate-cycle methodology.
 Author: Specfuse contributors
 License: Apache-2.0

{specfuse_loop-0.3.12 → specfuse_loop-0.3.13}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "specfuse-loop"
-version = "0.3.12"
+version = "0.3.13"
 description = "Local-first executor for the Specfuse Plan + Work Unit gate-cycle methodology."
 readme = "README.md"
 requires-python = ">=3.10"

specfuse_loop-0.3.13/specfuse/loop/data/VERSION

@@ -0,0 +1 @@
+0.3.13

{specfuse_loop-0.3.12 → specfuse_loop-0.3.13}/specfuse/loop/loop.py

@@ -62,7 +62,7 @@ SPECFUSE_DIR = Path(".specfuse")
 REPO_ROOT = SPECFUSE_DIR.parent
 FEATURES_DIR = SPECFUSE_DIR / "features"
 VERIFICATION_PATH = SPECFUSE_DIR / "verification.yml"
-DRIVER_VERSION = "0.3.12"
+DRIVER_VERSION = "0.3.13"
 # Oldest scaffold layout this driver can drive. init.sh stamps the scaffold's own
 # version into `.specfuse/VERSION`; check_scaffold_version() fails loud at startup if
 # the consumer's scaffold is older than this, pointing at `specfuse upgrade`. Bump
@@ -2651,13 +2651,17 @@ _NO_FAILURES_SENTINEL = "### Failure-class breakdown\n\n(no non-passing attempts
 def summarize_attempt_failure_classes(
     feature_dir: Path,
     gate_n: int | None = None,
+    exclude_correlation_id: str | None = None,
 ) -> str:
     """Render a '### Failure-class breakdown' markdown table from events.jsonl.
 
     Reads attempt_outcome events whose outcome != 'passed'.  When gate_n is
     provided, restricts to events whose correlation_id belongs to that gate
-    (resolved via _gate_number_from_wu_id).  Returns _NO_FAILURES_SENTINEL when
-    no non-passing attempts match the filter.
+    (resolved via _gate_number_from_wu_id).  When exclude_correlation_id is
+    provided, drops events with that exact correlation_id — used to keep a
+    close WU's OWN non-passing attempts out of the breakdown it authors, so the
+    close's stumble does not arm the guard against itself (issue #145).  Returns
+    _NO_FAILURES_SENTINEL when no non-passing attempts match the filter.
 
     Pure function — reads events.jsonl; no writes, no side effects.
     Malformed JSONL lines are skipped (legacy-event tolerance, AC5).
@@ -2680,8 +2684,10 @@ def summarize_attempt_failure_classes(
         payload = evt.get("payload") or {}
         if payload.get("outcome") == "passed":
             continue
+        cid = evt.get("correlation_id", "")
+        if exclude_correlation_id is not None and cid == exclude_correlation_id:
+            continue
         if gate_n is not None:
-            cid = evt.get("correlation_id", "")
             if _gate_number_from_wu_id(cid) != gate_n:
                 continue
         non_passing.append(payload)
@@ -2745,7 +2751,14 @@ def assert_failure_class_breakdown_when_failures_present(
         return True, ""
 
     gate_n = _gate_number_from_wu_id(wu.wu_id)
-    summary = summarize_attempt_failure_classes(feature_dir, gate_n)
+    # Exclude the close WU's OWN non-passing attempts: the breakdown documents
+    # SUBSTANTIVE-WU failures, not the close's own stumble. Without this, a
+    # failed first close attempt retroactively requires a new subsection in the
+    # RETROSPECTIVE the close itself authors — and the between-attempt
+    # `reset --hard` wipes the partial each retry → spin → blocked_human on
+    # otherwise-done features (issue #145).
+    summary = summarize_attempt_failure_classes(
+        feature_dir, gate_n, exclude_correlation_id=wu.wu_id)
 
     if summary == _NO_FAILURES_SENTINEL:
         return True, ""
@@ -2753,7 +2766,7 @@ def assert_failure_class_breakdown_when_failures_present(
     if re.search(r"^#{3} Failure-class breakdown\b", retro.read_text(), re.MULTILINE):
         return True, ""
 
-    # Count non-passing attempts for the error message.
+    # Count non-passing attempts (excluding the close's own) for the message.
     events_path = feature_dir / "events.jsonl"
     count = 0
     if events_path.exists():
@@ -2770,18 +2783,25 @@ def assert_failure_class_breakdown_when_failures_present(
             payload = evt.get("payload") or {}
             if payload.get("outcome") == "passed":
                 continue
+            cid = evt.get("correlation_id", "")
+            if cid == wu.wu_id:
+                continue
             if gate_n is not None:
-                cid = evt.get("correlation_id", "")
                 if _gate_number_from_wu_id(cid) != gate_n:
                     continue
             count += 1
 
     gate_label = f"gate {gate_n}" if gate_n is not None else "all gates"
+    # Actionable, self-contained retry note (flows into the next attempt's prompt
+    # via failure_note, issue #145 fix #2): name the moved bar AND embed the exact
+    # table to paste, so the agent doesn't rediscover the requirement.
     return (
         False,
         f"assert_failure_class_breakdown_when_failures_present: {count} "
-        f"non-passing attempt(s) in {gate_label} but '### Failure-class breakdown' "
-        f"subsection absent from RETROSPECTIVE.md",
+        f"substantive-WU attempt(s) in {gate_label} did not pass, so "
+        f"RETROSPECTIVE.md MUST include a '### Failure-class breakdown' subsection "
+        f"— it is absent. Add exactly this subsection to RETROSPECTIVE.md:\n\n"
+        f"{summary}",
     )
 
 

{specfuse_loop-0.3.12 → specfuse_loop-0.3.13/specfuse_loop.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: specfuse-loop
-Version: 0.3.12
+Version: 0.3.13
 Summary: Local-first executor for the Specfuse Plan + Work Unit gate-cycle methodology.
 Author: Specfuse contributors
 License: Apache-2.0

{specfuse_loop-0.3.12 → specfuse_loop-0.3.13}/specfuse_loop.egg-info/SOURCES.txt

@@ -51,6 +51,7 @@ tests/test_autosync_plugin.py
 tests/test_backend.py
 tests/test_bookkeeping_commit_crash_run.py
 tests/test_bookkeeping_commit_hook_crash.py
+tests/test_close_wu_self_reference.py
 tests/test_closing_deliverable_guard.py
 tests/test_cost_tracking.py
 tests/test_degraded_oracle.py

specfuse_loop-0.3.13/tests/test_close_wu_self_reference.py

@@ -0,0 +1,115 @@
+#
+# Copyright 2026 Specfuse contributors
+# Licensed under the Apache License, Version 2.0. See LICENSE.
+#
+"""Issue #145 — a close WU's own failed attempt must not arm the
+failure-class-breakdown guard against that same close WU.
+
+`assert_failure_class_breakdown_when_failures_present` requires a
+`### Failure-class breakdown` subsection in RETROSPECTIVE.md once non-passing
+attempts exist in the gate. It counted ALL non-passing attempts — including the
+close WU's OWN. Since the close WU is the one that authors the RETROSPECTIVE the
+guard inspects, its malformed first attempt retroactively added a required
+subsection to its own output, and the between-attempt `reset --hard` wiped the
+partial each retry → spin → blocked_human on otherwise-done features.
+
+Fix #3: exclude the close WU's own attempts (correlation_id == its wu_id) from
+the failure set that arms the guard against it. Substantive-WU failures still
+require the breakdown. Fix #2: when the guard does fire, its message names the
+required subsection and embeds the exact expected table (it flows into the retry
+prompt via failure_note), so the agent doesn't rediscover the moved bar.
+"""
+
+from __future__ import annotations
+
+import json
+import tempfile
+import unittest
+from pathlib import Path
+
+from tests._loop_loader import load_loop
+
+loop = load_loop()
+
+CLOSE_ID = "FEAT-2026-0099/G1-CLOSE"
+SUBSTANTIVE_ID = "FEAT-2026-0099/G1-T01"
+
+
+def _close_wu(wu_file: Path) -> "loop.WorkUnit":
+    return loop.WorkUnit(
+        wu_id=CLOSE_ID, file=wu_file, depends_on=[], type="close",
+        model="opus", effort="high", status="pending", attempts=1,
+        title="Close gate 1", body="body",
+    )
+
+
+def _attempt_event(correlation_id: str, outcome: str = "failed",
+                   failure_class: str = "other", sig: str = "verdict_malformed") -> str:
+    return json.dumps({
+        "event_type": "attempt_outcome",
+        "correlation_id": correlation_id,
+        "timestamp": "2026-07-12T00:00:00Z",
+        "payload": {"outcome": outcome, "failure_class": failure_class,
+                    "failure_signature": sig},
+    })
+
+
+class TestCloseSelfReference(unittest.TestCase):
+    def _feature_dir(self, tmp: str, events: list[str], retro: str = "# Retro\n") -> Path:
+        d = Path(tmp)
+        (d / "events.jsonl").write_text("\n".join(events) + "\n")
+        (d / "RETROSPECTIVE.md").write_text(retro)
+        return d
+
+    def test_close_own_failure_does_not_arm_guard(self):
+        # Only the close WU's OWN attempts are non-passing (the observed run).
+        with tempfile.TemporaryDirectory() as tmp:
+            d = self._feature_dir(tmp, [_attempt_event(CLOSE_ID)])
+            wu = _close_wu(d / "WU.md")
+            ok, msg = loop.assert_failure_class_breakdown_when_failures_present(
+                wu, d, Path(tmp), "HEAD~1")
+        self.assertTrue(ok, f"close's own failure must not require the breakdown: {msg}")
+
+    def test_substantive_failure_still_requires_breakdown(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            d = self._feature_dir(
+                tmp, [_attempt_event(SUBSTANTIVE_ID), _attempt_event(CLOSE_ID)])
+            wu = _close_wu(d / "WU.md")
+            ok, msg = loop.assert_failure_class_breakdown_when_failures_present(
+                wu, d, Path(tmp), "HEAD~1")
+        self.assertFalse(ok, "a substantive-WU failure still requires the breakdown")
+        # Fix #2 — message is actionable and embeds the exact table to include.
+        self.assertIn("### Failure-class breakdown", msg)
+        self.assertIn("| failure_class |", msg)
+
+    def test_breakdown_table_excludes_close_own_attempts(self):
+        # Two substantive failures + one close-own failure; the rendered table
+        # the guard offers must count only the two substantive ones.
+        with tempfile.TemporaryDirectory() as tmp:
+            d = self._feature_dir(
+                tmp,
+                [_attempt_event(SUBSTANTIVE_ID), _attempt_event(SUBSTANTIVE_ID),
+                 _attempt_event(CLOSE_ID)])
+            table = loop.summarize_attempt_failure_classes(
+                d, gate_n=1, exclude_correlation_id=CLOSE_ID)
+        self.assertIn("**total** | **2**", table)
+
+    def test_summarize_all_self_returns_sentinel(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            d = self._feature_dir(tmp, [_attempt_event(CLOSE_ID)])
+            table = loop.summarize_attempt_failure_classes(
+                d, gate_n=1, exclude_correlation_id=CLOSE_ID)
+        self.assertEqual(table, loop._NO_FAILURES_SENTINEL)
+
+    def test_breakdown_present_passes(self):
+        retro = "# Retro\n\n### Failure-class breakdown\n\n| x | 1 | y |\n"
+        with tempfile.TemporaryDirectory() as tmp:
+            d = self._feature_dir(tmp, [_attempt_event(SUBSTANTIVE_ID)], retro=retro)
+            wu = _close_wu(d / "WU.md")
+            ok, _ = loop.assert_failure_class_breakdown_when_failures_present(
+                wu, d, Path(tmp), "HEAD~1")
+        self.assertTrue(ok)
+
+
+if __name__ == "__main__":
+    unittest.main()

specfuse_loop-0.3.12/specfuse/loop/data/VERSION

@@ -1 +0,0 @@
-0.3.12