memstack-skill-loader 4.13.0__tar.gz → 4.13.1__tar.gz

{memstack_skill_loader-4.13.0/src/memstack_skill_loader.egg-info → memstack_skill_loader-4.13.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: memstack-skill-loader
-Version: 4.13.0
+Version: 4.13.1
 Summary: MCP server that vector-indexes MemStack Pro skills for on-demand loading
 Requires-Python: >=3.10
 Requires-Dist: mcp>=1.0.0

{memstack_skill_loader-4.13.0 → memstack_skill_loader-4.13.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "memstack-skill-loader"
-version = "4.13.0"
+version = "4.13.1"
 description = "MCP server that vector-indexes MemStack Pro skills for on-demand loading"
 requires-python = ">=3.10"
 dependencies = [

{memstack_skill_loader-4.13.0 → memstack_skill_loader-4.13.1}/src/memstack_skill_loader/__init__.py

@@ -1,3 +1,3 @@
 """MemStack Skill Loader — MCP server for semantic skill search."""
 
-__version__ = "4.13.0"
+__version__ = "4.13.1"

{memstack_skill_loader-4.13.0 → memstack_skill_loader-4.13.1}/src/memstack_skill_loader/diary_ingest.py

@@ -7,7 +7,8 @@ module by path, exactly like the devlog-webhook:
     python -m memstack_skill_loader.diary_ingest "<markdown-backup-path>"
 
 Each fact line is stored through :class:`MemoryStore` with ``source_type='diary'``
-and ``source_ref`` set to the diary file path.
+and ``source_ref`` set to the diary file path, canonicalized to an absolute
+forward-slash path so the same file re-ingested from any cwd/arg form dedupes.
 
 FAIL-OPEN CONTRACT (do not "improve" away)
 ------------------------------------------
@@ -141,6 +142,25 @@ def _project_from_path(path: str) -> str:
     return slug or "default"
 
 
+def _canonical_source_ref(path: str) -> str:
+    """Canonicalize a diary path into a stable dedupe key.
+
+    The Memory Engine dedupe key is ``(source_ref, subject, claim)`` compared as
+    raw strings. A diary saved once via a relative path and once via an absolute
+    path therefore produced two different ``source_ref`` strings and re-stored
+    every fact as a duplicate live row. Resolving to a real absolute path
+    (collapsing ``..`` segments and following symlinks) and emitting it with
+    forward slashes makes every path form of the same file collapse to one key.
+
+    Fail-open: this runs in the never-break-the-diary-save path, so if resolution
+    fails for any reason, fall back to the raw string rather than raise.
+    """
+    try:
+        return Path(path).resolve().as_posix()
+    except OSError:
+        return str(path)
+
+
 def _warn(message: str) -> None:
     print(f"{_STDERR_PREFIX} {message}", file=sys.stderr)
 
@@ -169,7 +189,7 @@ def ingest_diary_file(path: str, store: MemoryStore | None = None) -> dict:
     if store is None:
         store = MemoryStore()
     project = _project_from_path(path)
-    source_ref = str(path)
+    source_ref = _canonical_source_ref(path)
 
     ingested = 0
     duplicates = 0

{memstack_skill_loader-4.13.0 → memstack_skill_loader-4.13.1/src/memstack_skill_loader.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: memstack-skill-loader
-Version: 4.13.0
+Version: 4.13.1
 Summary: MCP server that vector-indexes MemStack Pro skills for on-demand loading
 Requires-Python: >=3.10
 Requires-Dist: mcp>=1.0.0

{memstack_skill_loader-4.13.0 → memstack_skill_loader-4.13.1}/tests/test_diary_ingest.py

@@ -116,9 +116,13 @@ def test_ingest_stores_facts_with_diary_provenance(tmp_path):
     store = MemoryStore()
     rows = store.list_facts(project="demoproj", limit=50)
     assert len(rows) == 4
+    # source_ref is stored in canonical form: an absolute, forward-slash path
+    # (resolve + as_posix), so re-ingests from any cwd/arg form dedupe.
+    expected_ref = Path(p).resolve().as_posix()
     for r in rows:
         assert r["source_type"] == "diary"
-        assert r["source_ref"] == str(p)
+        assert r["source_ref"] == expected_ref
+        assert "\\" not in r["source_ref"]
     # entities landed
     supa = [r for r in rows if r["subject"] == "adminstack.portal.auth"][0]
     assert set(store.get_entities(supa["id"])) == {"supabase", "auth"}
@@ -198,6 +202,70 @@ def test_fact_exists_dedupe_key(tmp_path):
     assert store.fact_exists("s.x", "the claim", "/other.md") is False
 
 
+# ---------------------------------------------------------------------------
+# source_ref canonicalization: the SAME file ingested via different path forms
+# (relative, absolute, ../-laden, symlinked) must dedupe to ONE fact. Regression
+# guard for the relative-vs-absolute duplicate-fact gap (4 dupes in the real
+# store on 2026-07-23). Genuinely different files must still stay separate.
+# ---------------------------------------------------------------------------
+
+def test_canonical_source_ref_is_absolute_forward_slash(tmp_path, monkeypatch):
+    p = _diary(tmp_path, "placeholder")
+    monkeypatch.chdir(tmp_path)
+    canon = diary_ingest._canonical_source_ref(p.name)  # relative input
+    assert canon == Path(p).resolve().as_posix()
+    assert Path(canon).is_absolute()
+    assert "\\" not in canon  # forward-slash only
+
+
+def test_relative_and_absolute_path_dedupe_to_one_fact(tmp_path, monkeypatch):
+    p = _diary(tmp_path, CANONICAL_EXAMPLES)
+    monkeypatch.chdir(tmp_path)
+    first = diary_ingest.ingest_diary_file(p.name)          # relative path
+    second = diary_ingest.ingest_diary_file(str(p.resolve()))  # absolute path
+    assert first["ingested"] == 4
+    assert second["ingested"] == 0
+    assert second["duplicates"] == 4
+    assert MemoryStore().stats(project="demoproj")["total"] == 4
+
+
+def test_dotdot_segments_canonicalize_to_same_fact(tmp_path):
+    p = _diary(tmp_path, CANONICAL_EXAMPLES)
+    sub = tmp_path / "sub"
+    sub.mkdir()
+    detour = sub / ".." / p.name  # resolves back to the same file
+    first = diary_ingest.ingest_diary_file(str(p))
+    second = diary_ingest.ingest_diary_file(str(detour))
+    assert first["ingested"] == 4
+    assert second["duplicates"] == 4
+    assert MemoryStore().stats(project="demoproj")["total"] == 4
+
+
+def test_symlinked_path_canonicalizes_to_same_fact(tmp_path):
+    p = _diary(tmp_path, CANONICAL_EXAMPLES)
+    link = tmp_path / "2026-07-20-demoproj-link.md"
+    try:
+        link.symlink_to(p)
+    except (OSError, NotImplementedError):
+        pytest.skip("symlink creation not permitted on this platform")
+    first = diary_ingest.ingest_diary_file(str(p))
+    second = diary_ingest.ingest_diary_file(str(link))  # follows to real path
+    assert first["ingested"] == 4
+    assert second["duplicates"] == 4
+    assert MemoryStore().stats(project="demoproj")["total"] == 4
+
+
+def test_two_different_diaries_store_as_separate_facts(tmp_path):
+    p1 = _diary(tmp_path, CANONICAL_EXAMPLES, name="2026-07-20-demoproj.md")
+    p2 = _diary(tmp_path, CANONICAL_EXAMPLES, name="2026-07-21-demoproj.md")
+    a = diary_ingest.ingest_diary_file(str(p1))
+    b = diary_ingest.ingest_diary_file(str(p2))
+    assert a["ingested"] == 4
+    assert b["ingested"] == 4  # different source_ref => not dupes
+    assert b["duplicates"] == 0
+    assert MemoryStore().stats(project="demoproj")["total"] == 8
+
+
 # ---------------------------------------------------------------------------
 # cross-repo drift guard: the diary SKILL.md examples must parse to exactly the
 # facts CANONICAL_EXAMPLES parses to. The skill lives in a SIBLING repo