unique-user-memory 2026.36.0.dev2__tar.gz → 2026.36.0.dev3__tar.gz

{unique_user_memory-2026.36.0.dev2 → unique_user_memory-2026.36.0.dev3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: unique-user-memory
-Version: 2026.36.0.dev2
+Version: 2026.36.0.dev3
 Summary: 
 Author: Fabian Schläpfer
 Author-email: Fabian Schläpfer <fabian@unique.ch>

{unique_user_memory-2026.36.0.dev2 → unique_user_memory-2026.36.0.dev3}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "unique_user_memory"
-version = "2026.36.0.dev2"
+version = "2026.36.0.dev3"
 description = ""
 authors = [
     { name = "Fabian Schläpfer", email = "fabian@unique.ch" },

{unique_user_memory-2026.36.0.dev2 → unique_user_memory-2026.36.0.dev3}/unique_user_memory/tests/test_user_memory.py

@@ -26,6 +26,7 @@ from unique_user_memory.config import UserMemoryConfig
 from unique_user_memory.user_memory import (
     UserMemoryState,
     _gate_max_tokens,
+    _is_well_formed_profile,
     _sanitize_for_xml_context,
     condense_user_memory,
     consolidate_user_memory,
@@ -41,6 +42,7 @@ from unique_user_memory.user_memory import (
 from unique_user_memory.user_memory_message_log import UserMemoryMessageLogger
 from unique_user_memory.user_memory_postprocessor import UserMemoryPostprocessor
 from unique_user_memory.user_memory_prompts import (
+    condensation_system_prompt,
     consolidation_system_prompt,
     empty_profile,
     memory_gate_system_prompt,
@@ -49,6 +51,23 @@ from unique_user_memory.user_memory_prompts import (
 _TEST_LANGUAGE_MODEL = LanguageModelInfo.from_name(DEFAULT_LANGUAGE_MODEL)
 
 
+def _complete_profile_body(
+    identity: str = "- Prefers concise answers",
+    *,
+    legacy_title: bool = False,
+) -> str:
+    sections = (
+        f"## Identity\n{identity}",
+        "## Communication Preferences\n_(empty)_",
+        "## Work Context\n_(empty)_",
+        "## Skills & Expertise\n_(empty)_",
+        "## Follow-ups\n_(empty)_",
+        "## Recent Topics\n_(empty)_",
+    )
+    body = "\n\n".join(sections)
+    return f"# User Memory\n\n{body}" if legacy_title else body
+
+
 def _chat_event() -> ChatEvent:
     return ChatEvent(
         id="event_1",
@@ -111,6 +130,59 @@ def test_recent_topics_is_last_section_and_deduplicated() -> None:
     assert "do NOT also log that same information" in consolidation_prompt
 
 
+@pytest.mark.ai
+def test_generated_profile_templates_start_with_identity_without_document_title() -> (
+    None
+):
+    """Purpose: Verify new profile output omits the obsolete document-level title.
+    Why this matters: Persistent memory should expose only its meaningful H2 sections.
+    Setup summary: Render empty and LLM prompts, then assert the required start format.
+    """
+    profile = empty_profile("user_1")
+    prompts = (
+        consolidation_system_prompt(2000),
+        condensation_system_prompt(
+            max_tokens=2000,
+            current_tokens=2200,
+            target_tokens=1800,
+        ),
+    )
+
+    assert profile_body(profile).startswith("## Identity")
+    assert "# User Memory" not in profile
+    assert all("starting with\n`## Identity`" in prompt for prompt in prompts)
+    assert all(
+        "Do NOT add a document-level `# User Memory` title" in prompt
+        for prompt in prompts
+    )
+
+
+@pytest.mark.ai
+@pytest.mark.parametrize("legacy_title", [False, True])
+def test_well_formed_profile_accepts_current_and_legacy_complete_profiles(
+    legacy_title: bool,
+) -> None:
+    """Purpose: Verify validation supports titleless output and stored legacy profiles.
+    Why this matters: New rewrites must migrate format without breaking old memories.
+    Setup summary: Validate complete profiles with and without the former H1 title.
+    """
+    assert _is_well_formed_profile(_complete_profile_body(legacy_title=legacy_title))
+
+
+@pytest.mark.ai
+def test_well_formed_profile_rejects_missing_required_section() -> None:
+    """Purpose: Verify rewritten profiles retain every required H2 section.
+    Why this matters: A rewrite must not silently erase a category of user memory.
+    Setup summary: Remove Work Context from a complete profile and validate rejection.
+    """
+    profile = _complete_profile_body().replace(
+        "## Work Context\n_(empty)_\n\n",
+        "",
+    )
+
+    assert not _is_well_formed_profile(profile)
+
+
 def test_profile_body_strips_frontmatter() -> None:
     # The orchestrator renders this into the system prompt, where the
     # bookkeeping fields are noise for the model.
@@ -190,7 +262,7 @@ async def test_fit_user_memory_condenses_before_hard_cut(
         f"- fact number {index} that is fairly wordy about the user"
         for index in range(400)
     )
-    condensed = "# User Memory\n\n## Identity\n- concise summary of the user"
+    condensed = _complete_profile_body("- concise summary of the user")
     condense = AsyncMock(return_value=condensed)
     monkeypatch.setattr(
         "unique_user_memory.user_memory.condense_user_memory",
@@ -264,10 +336,11 @@ async def test_condense_user_memory_accepts_frontmatter_output(
     monkeypatch: pytest.MonkeyPatch,
 ) -> None:
     """Accept legacy LLM output while returning only the condensed profile body."""
-    condensed = "# User Memory\n\n## Identity\n- concise summary"
+    condensed = _complete_profile_body("- concise summary")
     response = MagicMock()
     response.choices[0].message.content = (
-        "---\nuser_id: stale-user\nturn_count: 99\n---\n\n" + condensed
+        "---\nuser_id: stale-user\nturn_count: 99\n---\n\n"
+        + _complete_profile_body("- concise summary", legacy_title=True)
     )
     llm_service = MagicMock()
     llm_service.complete_async = AsyncMock(return_value=response)
@@ -296,8 +369,8 @@ async def test_user_memory_llm_paths_forward_event_attribution(
     Why this matters: Every user-memory ModelUsage row must link to its chat.
     Setup summary: Exercise all shared service paths and inspect toolkit calls.
     """
-    rewritten = "# User Memory\n\n## Identity\n- Prefers concise answers"
-    condensed = "# User Memory\n\n## Identity\n- Concise"
+    rewritten = _complete_profile_body()
+    condensed = _complete_profile_body("- Concise")
     complete_async = AsyncMock(
         side_effect=[
             _completion_response("UPDATE"),
@@ -462,7 +535,7 @@ async def test_consolidate_user_memory_runs_full_rewrite_when_gate_update(
     monkeypatch: pytest.MonkeyPatch,
 ) -> None:
     current = empty_profile("user_1")
-    rewritten = "# User Memory\n\n## Identity\n- Prefers concise answers"
+    rewritten = _complete_profile_body()
     response = MagicMock()
     response.choices[0].message.content = rewritten
     llm_service = MagicMock()
@@ -498,7 +571,7 @@ async def test_consolidate_user_memory_runs_full_rewrite_when_gate_update(
 async def test_consolidate_user_memory_adds_frontmatter_to_llm_body(
     monkeypatch: pytest.MonkeyPatch,
 ) -> None:
-    rewritten = "# User Memory\n\n## Identity\n- Prefers concise answers"
+    rewritten = _complete_profile_body()
     response = MagicMock()
     response.choices[0].message.content = rewritten
     llm_service = MagicMock()
@@ -534,10 +607,11 @@ async def test_consolidate_user_memory_replaces_llm_frontmatter(
     monkeypatch: pytest.MonkeyPatch,
 ) -> None:
     """Strip untrusted legacy metadata before assembling the updated profile."""
-    rewritten = "# User Memory\n\n## Identity\n- Prefers concise answers"
+    rewritten = _complete_profile_body()
     response = MagicMock()
     response.choices[0].message.content = (
-        "---\nuser_id: stale-user\nturn_count: 99\n---\n\n" + rewritten
+        "---\nuser_id: stale-user\nturn_count: 99\n---\n\n"
+        + _complete_profile_body(legacy_title=True)
     )
     llm_service = MagicMock()
     llm_service.complete_async = AsyncMock(return_value=response)
@@ -572,7 +646,7 @@ async def test_consolidate_user_memory_skips_gate_when_disabled(
     monkeypatch: pytest.MonkeyPatch,
 ) -> None:
     current = empty_profile("user_1")
-    rewritten = "# User Memory\n\n## Identity\n- Prefers concise answers"
+    rewritten = _complete_profile_body()
     response = MagicMock()
     response.choices[0].message.content = rewritten
     llm_service = MagicMock()
@@ -729,7 +803,7 @@ async def test_should_consolidate_memory_falls_back_to_true_on_error(
 async def test_consolidate_user_memory_invokes_update_callbacks_on_rewrite(
     monkeypatch: pytest.MonkeyPatch,
 ) -> None:
-    rewritten = "# User Memory\n\n## Identity\n- Prefers concise answers"
+    rewritten = _complete_profile_body()
     response = MagicMock()
     response.choices[0].message.content = rewritten
     llm_service = MagicMock()

{unique_user_memory-2026.36.0.dev2 → unique_user_memory-2026.36.0.dev3}/unique_user_memory/user_memory.py

@@ -69,6 +69,8 @@ _TRUNCATION_MARKER = "\n\n<!-- truncated to fit memory budget -->"
 _DEFAULT_LANGUAGE_MODEL = LanguageModelInfo.from_name(DEFAULT_GPT_4o)
 _FRONTMATTER_RE = re.compile(r"^---\n.*?\n---\n", re.DOTALL)
 _TURN_COUNT_RE = re.compile(r"^turn_count:\s*(\d+)\s*$", re.MULTILINE)
+_LEGACY_PROFILE_TITLE_RE = re.compile(r"^# User Memory[ \t]*(?:\n+|$)")
+_H2_HEADING_RE = re.compile(r"^## (.+?)[ \t]*$", re.MULTILINE)
 
 
 def profile_body(content: str) -> str:
@@ -286,7 +288,7 @@ async def condense_user_memory(
         )
         return None
 
-    candidate = profile_body(_strip_code_fences(raw))
+    candidate = _without_legacy_profile_title(profile_body(_strip_code_fences(raw)))
     if not _is_well_formed_profile(candidate):
         logger.warning(
             "[user-memory] condense output did not look like a profile (%d chars)",
@@ -906,7 +908,9 @@ async def _rewrite_user_memory(
         logger.info("[user-memory] consolidation NOOP - keeping existing memory")
         return safe_current
 
-    candidate_body = profile_body(_strip_code_fences(raw))
+    candidate_body = _without_legacy_profile_title(
+        profile_body(_strip_code_fences(raw))
+    )
     if not _is_well_formed_profile(candidate_body):
         logger.warning(
             "[user-memory] LLM output did not look like a profile (%d chars)",
@@ -947,7 +951,14 @@ def _sanitize_for_xml_context(text: str) -> str:
 def _is_well_formed_profile(content: str) -> bool:
     if not content or len(content.strip()) < 20:
         return False
-    return content.startswith("# User Memory") and "## Identity" in content
+    body = _without_legacy_profile_title(content.strip())
+    headings = tuple(_H2_HEADING_RE.findall(body))
+    return body.startswith("## Identity") and headings == SECTION_HEADINGS
+
+
+def _without_legacy_profile_title(content: str) -> str:
+    """Remove the former document title while preserving the H2 profile body."""
+    return _LEGACY_PROFILE_TITLE_RE.sub("", content.strip(), count=1).lstrip()
 
 
 def _strip_code_fences(text: str) -> str:

{unique_user_memory-2026.36.0.dev2 → unique_user_memory-2026.36.0.dev3}/unique_user_memory/user_memory_prompts.py

@@ -19,8 +19,6 @@ last_updated: {{ timestamp }}
 turn_count: 0
 ---
 
-# User Memory
-
 {% for heading in section_headings -%}
 ## {{ heading }}
 _(empty)_
@@ -60,8 +58,9 @@ You receive two XML blocks:
 # Output
 
 Return the complete, rewritten Markdown profile body, starting with
-`# User Memory`. Do NOT emit a diff. Do NOT wrap the output in
-``` fences. Do NOT add commentary before or after the body.
+`## Identity`. Do NOT add a document-level `# User Memory` title. Do NOT
+emit a diff. Do NOT wrap the output in ``` fences. Do NOT add commentary
+before or after the body.
 
 The body MUST contain exactly these section headings, in this order, even
 when a section is empty (use the literal string `_(empty)_` as a placeholder):
@@ -256,8 +255,8 @@ fact about the user. This is lossy compression, not deletion of meaning.
 # Output
 
 Return the complete rewritten profile body, starting with
-`# User Memory`. Do NOT emit a diff, or commentary,
-and do NOT wrap the output in ``` fences.
+`## Identity`. Do NOT add a document-level `# User Memory` title. Do NOT
+emit a diff or commentary, and do NOT wrap the output in ``` fences.
 """