memleaf 0.2.6__tar.gz → 0.2.8__tar.gz

{memleaf-0.2.6 → memleaf-0.2.8}/CHANGELOG.md

@@ -2,6 +2,31 @@
 
 All notable changes to memleaf are documented here.
 
+## 0.2.8 — 2026-09-01
+
+- Let the automatic summarizer independently reject an over-admitted gate
+  candidate with an exact `NO_CHANGE` response, without writing memory,
+  history, or Scope state; explicit remember remains write-oriented and does
+  not permit this response.
+- Ground every model-attributed project Scope in the candidate's own project
+  name or registered alias, and prevent the summarizer from drifting away
+  from the gate-selected Scope.
+- Register session and Vault Scopes only for committed memories or trusted
+  active duplicate targets, so temporary, deferred, and no-change mailbox
+  items cannot pollute Scope state.
+
+## 0.2.7 — 2026-09-01
+
+- Distinguish an invalid gate type enum from a candidate type that conflicts
+  with its selected active update target, while preserving the strict,
+  immutable target-type constraint.
+- Give each gate failure a bounded, actionable retry instruction: use a legal
+  memory type, preserve the target type for the same future use, or split a
+  genuinely different future use into an independent atomic candidate.
+- Preserve the dedicated `update_target_type_mismatch` diagnostic across Core,
+  MCP, and Hermes; repeated failures still write nothing and retain the inbox
+  for a later retry.
+
 ## 0.2.6 — 2026-09-01
 
 - Reject one worthy memory that spans multiple independent project scopes and

{memleaf-0.2.6/src/memleaf.egg-info → memleaf-0.2.8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: memleaf
-Version: 0.2.6
+Version: 0.2.8
 Summary: A local-first Markdown memory core for AI agents
 Author: memleaf contributors
 License-Expression: MIT
@@ -23,8 +23,8 @@ Dynamic: license-file
 
 [English](README.en.md) · [PyPI](https://pypi.org/project/memleaf/) · [GitHub](https://github.com/miffyblueboo/memleaf)
 
-> **当前版本：0.2.6。**
-> 核心库、Vault、stdio MCP Server、初始化 CLI、模型路由、提炼流程、受控检索协议和宿主适配器已经实现。memleaf 0.2.6 通过 PyPI 分发。
+> **当前版本：0.2.8。**
+> 核心库、Vault、stdio MCP Server、初始化 CLI、模型路由、提炼流程、受控检索协议和宿主适配器已经实现。memleaf 0.2.8 通过 PyPI 分发。
 > **当前版本支持 Hermes 和 Codex。** Antigravity（反重力）不检测、不安装、不配置。
 
 ## 项目定位

{memleaf-0.2.6 → memleaf-0.2.8}/README.md

@@ -4,8 +4,8 @@
 
 [English](README.en.md) · [PyPI](https://pypi.org/project/memleaf/) · [GitHub](https://github.com/miffyblueboo/memleaf)
 
-> **当前版本：0.2.6。**
-> 核心库、Vault、stdio MCP Server、初始化 CLI、模型路由、提炼流程、受控检索协议和宿主适配器已经实现。memleaf 0.2.6 通过 PyPI 分发。
+> **当前版本：0.2.8。**
+> 核心库、Vault、stdio MCP Server、初始化 CLI、模型路由、提炼流程、受控检索协议和宿主适配器已经实现。memleaf 0.2.8 通过 PyPI 分发。
 > **当前版本支持 Hermes 和 Codex。** Antigravity（反重力）不检测、不安装、不配置。
 
 ## 项目定位

{memleaf-0.2.6 → memleaf-0.2.8}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "memleaf"
-version = "0.2.6"
+version = "0.2.8"
 description = "A local-first Markdown memory core for AI agents"
 readme = "README.md"
 requires-python = ">=3.11"

{memleaf-0.2.6 → memleaf-0.2.8}/src/memleaf/__init__.py

@@ -1,6 +1,6 @@
 """Local-first Markdown memory core for AI agents."""
 
-__version__ = "0.2.6"
+__version__ = "0.2.8"
 
 from .config import DEFAULT_CONFIG, default_config, load_config, save_config
 from .frontmatter import FrontmatterError, dump_frontmatter, dump_yaml, load_yaml, parse_frontmatter

{memleaf-0.2.6 → memleaf-0.2.8}/src/memleaf/hermes_provider/__init__.py

@@ -77,6 +77,9 @@ _MODEL_VALIDATION_DETAILS = frozenset(
         "duplicate_candidate_id",
         "duplicate_update_target",
         "mixed_project_scopes",
+        "update_target_type_mismatch",
+        "scope_not_grounded",
+        "scope_drift",
         "invalid_evidence",
         "invalid_flags",
         "invalid_type",

{memleaf-0.2.6 → memleaf-0.2.8}/src/memleaf/hermes_provider/plugin.yaml

@@ -1,5 +1,5 @@
 name: memleaf
-version: 0.2.6
+version: 0.2.8
 description: "Hermes-native external MemoryProvider for local-first Markdown memory shared with the memleaf MCP server."
 hooks:
   - prefetch

{memleaf-0.2.6 → memleaf-0.2.8}/src/memleaf/processing.py

@@ -32,13 +32,17 @@ from .models import Memory, utc_now
 from .native_index import NativeIndexer
 from .prompts import (
     DUPLICATE_TARGET_CORRECTION,
+    GATE_TYPE_CORRECTION,
     JSON_CORRECTION,
     MIXED_PROJECT_SCOPES_CORRECTION,
     RELATIVE_TIME_CORRECTION,
     GATE_SYSTEM,
+    SCOPE_GROUNDING_CORRECTION,
+    SUMMARY_SCOPE_CORRECTION,
     SUMMARY_TARGET_CORRECTION,
     SUMMARY_TYPE_CORRECTION,
     SUMMARIZE_SYSTEM,
+    UPDATE_TARGET_TYPE_CORRECTION,
     gate_prompt,
     summarize_prompt,
 )
@@ -49,6 +53,7 @@ from .scope_maintenance import ScopeMaintainer, ScopeMaintenanceError, scope_reg
 from .validation import (
     MODEL_VALIDATION_DETAILS,
     ModelOutputError,
+    NO_CHANGE_DECISION,
     parse_gate_output,
     parse_summarize_output,
 )
@@ -482,6 +487,14 @@ class Processor:
             return DUPLICATE_TARGET_CORRECTION
         if hint == "mixed_project_scopes":
             return MIXED_PROJECT_SCOPES_CORRECTION
+        if stage == "gate" and hint == "update_target_type_mismatch":
+            return UPDATE_TARGET_TYPE_CORRECTION
+        if stage == "gate" and hint == "invalid_type":
+            return GATE_TYPE_CORRECTION
+        if stage == "gate" and hint == "scope_not_grounded":
+            return SCOPE_GROUNDING_CORRECTION
+        if stage == "summarize" and hint == "scope_drift":
+            return SUMMARY_SCOPE_CORRECTION
         if hint == "relative_time":
             return RELATIVE_TIME_CORRECTION
         if stage == "summarize" and hint == "invalid_update_target":
@@ -1608,6 +1621,7 @@ class Processor:
                     related_native_ids=related_native_ids,
                     related_memory_ids=related_memory_ids,
                     scope_registry=validation_scope_registry,
+                    allow_no_change=False,
                 ),
                 diagnostic_context={
                     "source": turn.source,
@@ -1646,6 +1660,7 @@ class Processor:
                 current_event_keys=turn.event_keys,
                 related_memory_ids=gate_related_memory_ids,
                 related_memory_types=gate_related_memory_types,
+                scope_registry=validation_scope_registry,
             ),
             diagnostic_context={
                 "source": turn.source,
@@ -1657,13 +1672,6 @@ class Processor:
         observed_scopes: list[str] = []
         for candidate in gate["candidates"]:
             candidate_scopes = list(candidate["scopes"])
-            if (
-                candidate_scopes != ["unscoped"]
-                and candidate.get("scope_source") != "insufficient_context"
-            ):
-                for observed_scope in candidate_scopes:
-                    if observed_scope not in observed_scopes:
-                        observed_scopes.append(observed_scope)
             # An automatic candidate with no reliable project attribution is
             # retained as a retryable inbox turn, never silently promoted to
             # global knowledge.  The processed ledger records only a compact
@@ -1752,6 +1760,27 @@ class Processor:
                             native_refs=candidate_native_refs,
                         )
                     )
+                    # Automatic duplicate observations are metadata no-ops;
+                    # only the already-active target's scopes are trustworthy
+                    # session context, never a transient model-provided scope.
+                    duplicate_scopes = next(
+                        (
+                            item.get("scopes")
+                            for item in candidate_related
+                            if isinstance(item, Mapping)
+                            and isinstance(item.get("memory_id"), str)
+                            and item["memory_id"].casefold() == duplicate_memory_id.casefold()
+                            and isinstance(item.get("scopes"), list)
+                        ),
+                        [],
+                    )
+                    for observed_scope in duplicate_scopes:
+                        if (
+                            isinstance(observed_scope, str)
+                            and observed_scope != "unscoped"
+                            and observed_scope not in observed_scopes
+                        ):
+                            observed_scopes.append(observed_scope)
                 continue
 
             gate_update_target = candidate.get("update_memory_id")
@@ -1787,6 +1816,9 @@ class Processor:
                     related_native_ids=candidate_native_ids,
                     related_memory_ids=candidate_memory_ids,
                     scope_registry=validation_scope_registry,
+                    expected_scopes=candidate["scopes"],
+                    expected_scope_source=candidate["scope_source"],
+                    allow_no_change=True,
                     expected_type=(
                         candidate.get("type")
                         if (
@@ -1808,6 +1840,8 @@ class Processor:
                     "turn_index": turn.turn_index,
                 },
             )
+            if summary.get("decision") == NO_CHANGE_DECISION:
+                continue
             if gate_update_target is not None:
                 summary_update_target = summary.get("update_memory_id")
                 if summary_update_target is None:
@@ -1833,9 +1867,6 @@ class Processor:
                     scope_source=summary.get("scope_source"),
                 )
                 continue
-            for observed_scope in summary["scopes"]:
-                if observed_scope not in observed_scopes:
-                    observed_scopes.append(observed_scope)
             requests.append(
                 self._request(
                     summary,
@@ -1845,6 +1876,13 @@ class Processor:
                     native_refs=candidate_native_refs,
                 )
             )
+            for observed_scope in summary["scopes"]:
+                if (
+                    isinstance(observed_scope, str)
+                    and observed_scope != "unscoped"
+                    and observed_scope not in observed_scopes
+                ):
+                    observed_scopes.append(observed_scope)
         return requests, observed_scopes
 
     def _state_for_snapshot_unlocked(self, snapshot: _Snapshot, processed: Mapping[str, Any]) -> Mapping[str, Any]:

{memleaf-0.2.6 → memleaf-0.2.8}/src/memleaf/prompts.py

@@ -34,6 +34,20 @@ useful. Related active memories only identify a complete duplicate or valid
 state-update target; they are not current evidence and cannot make a non-worthy
 operational/test result worthy. Their item, body, and serialized-character
 context is bounded; ellipses and omissions do not prove a fact is absent.
+For a worthy candidate with scope_source=model, each selected project scope must
+be grounded by this candidate's own memory text: use the project's registered
+scope name or one of its registered aliases; for a new unregistered project
+scope, use that project's name itself. Do not borrow a project name from another
+event, the session background, related memories, or an aggregate mailbox turn.
+If the candidate text does not identify exactly one selected project, choose an
+evidence-supported project, use unscoped with insufficient_context, defer it, or
+set worth=false. user and session_context scope sources may rely on their
+authoritative scope context.
+An update target's existing type is immutable: when the same future use is being
+updated, candidate type must exactly equal the related active target type. If a
+fact serves a different future use, do not force it onto that target; omit
+update_memory_id and emit a separate atomic candidate, or use worth=false when
+it is already covered or only temporary.
 Within one gate response, each active memory target may appear at most once
 across all candidates, whether it is named by duplicate_memory_id or
 update_memory_id. If multiple facts share one future-use/update target, merge
@@ -151,6 +165,16 @@ When the gate candidate supplies update_memory_id, that target is immutable:
 omit update_memory_id or copy that exact active ID, never replace it with a
 different, guessed, native, or historical ID. Keep the summary type identical
 to the gate candidate type and to the active update target type.
+For automatic capture/process, independently re-check whether this candidate
+still has a concrete future-use fact or action. If not, return exactly
+{"decision":"NO_CHANGE"}; this is the only no-write summary response. Use it
+for temporary delegation, waiting for feedback, a one-off meeting arrangement with
+no durable outcome/decision/project constraint, or a subject/attachment-only item
+without a concrete impact or conclusion.
+Explicit remember mode never permits NO_CHANGE and must return the normal summary
+object. When producing a normal automatic summary, copy the gate candidate's
+scopes exactly. scope_source may be omitted to inherit the gate value, but an
+explicit value must match it; never drift to another project.
 For the same use, preserve still-valid facts, add confirmed progress, replace
 contradictions, and never reduce the memory to the latest operation or create a
 sibling with a new label.
@@ -241,6 +265,27 @@ RELATIVE_TIME_CORRECTION = (
 )
 
 
+GATE_TYPE_CORRECTION = (
+    "Previous output violated: invalid_type. In gate output, whenever type is "
+    "non-null it must be exactly one of preference, fact, project, todo, event, "
+    "identity, or other. worth=true requires a non-null legal type; worth=false "
+    "may use null or any of those legal types (for example, a duplicate candidate "
+    "may retain its type). Do not invent enum values such as requirement, "
+    "decision, or task. Return only the strict gate JSON object."
+)
+
+
+UPDATE_TARGET_TYPE_CORRECTION = (
+    "Previous output violated: update_target_type_mismatch. An existing active "
+    "update target's type is immutable. For the same future-use topic, set the "
+    "candidate type exactly equal to the supplied related target type. For a "
+    "different future use, remove update_memory_id and emit a separate atomic "
+    "candidate, or set worth=false if it is already covered or only temporary. "
+    "Use only a supplied related active ID; never guess, use native/history IDs, "
+    "or force a cross-purpose update. Return only the strict gate JSON object."
+)
+
+
 MIXED_PROJECT_SCOPES_CORRECTION = (
     "Previous output violated: mixed_project_scopes. Keep each worthy "
     "candidate and summary atomic: it may contain at most one distinct "
@@ -280,6 +325,28 @@ SUMMARY_TYPE_CORRECTION = (
 )
 
 
+SUMMARY_SCOPE_CORRECTION = (
+    "Previous output violated: scope_drift. For a normal automatic summary, copy "
+    "the gate candidate scopes exactly and keep its scope_source unchanged. If "
+    "scope_source was omitted, inherit the gate value; an explicit value must "
+    "match it. Do not move the memory to another project. If the candidate has "
+    "no independent future-use fact or action, return exactly "
+    '{"decision":"NO_CHANGE"}. Return only the strict JSON object.'
+)
+
+
+SCOPE_GROUNDING_CORRECTION = (
+    "Previous output violated: scope_not_grounded. For each worthy candidate "
+    "with scope_source=model, choose a project scope named by that candidate's "
+    "own memory text: use the registered project name or alias, or the name "
+    "itself for a new scope. Do not borrow a name from another event, related "
+    "memory, session background, or an aggregate mailbox turn. If exactly one "
+    "project cannot be supported, choose the evidence-supported scope, use "
+    "unscoped with insufficient_context, defer it, or set worth=false. Return "
+    "only the strict gate JSON object."
+)
+
+
 COMPACT_SYSTEM = """You are memleaf's memory compactor. Return JSON only.
 Merge only the supplied low-priority memories when they express compatible
 information. Return an object with a memories array; [] is a safe no-op.
@@ -372,6 +439,14 @@ def summarize_prompt(
         f"Current scope registry (safe projection; no paths):\n"
         f"{_json(scope_registry if scope_registry is not None else [])}\n"
     )
+    if not explicit:
+        prompt += (
+            "Automatic admission re-check: if the candidate has no independent "
+            "future-use fact or action after reviewing the evidence, return exactly "
+            '{"decision":"NO_CHANGE"}; do not return an empty or partial memory object. '
+            "For a normal summary, copy Candidate scopes exactly; omit scope_source "
+            "to inherit it or repeat the same value, never choose another scope.\n"
+        )
     example_key = _first_event_key(events)
     if example_key is not None:
         candidate_type = candidate.get("type") if isinstance(candidate, dict) else None
@@ -405,6 +480,8 @@ def summarize_prompt(
                 }
             )
         )
+    if not explicit:
+        return prompt + '\nReturn one summary JSON object or exactly {"decision":"NO_CHANGE"}.'
     return prompt + "\nReturn one summary JSON object."
 
 

{memleaf-0.2.6 → memleaf-0.2.8}/src/memleaf/scope_state.py

@@ -256,9 +256,20 @@ def _term_matches(query: str, term: str) -> bool:
 def resolve_query_project_scope(query: Any, config: Mapping[str, Any]) -> str | None:
     """Return one unambiguous project scope named or aliased by ``query``."""
 
+    matches = project_scope_matches_text(query, config)
+    return matches[0] if len(matches) == 1 else None
+
+
+def project_scope_matches_text(query: Any, config: Mapping[str, Any]) -> list[str]:
+    """Return registered project scopes named or aliased by text.
+
+    Matching uses the same token/substring rules as query scope resolution,
+    but retains every match so callers can reject ambiguous model attribution.
+    """
+
     query_value = _query_text(query)
     if not query_value:
-        return None
+        return []
     registry = validate_scope_registry(config.get("scopes", {}) if isinstance(config, Mapping) else {})
     matches: set[str] = set()
     for scope, node in registry.items():
@@ -271,7 +282,7 @@ def resolve_query_project_scope(query: Any, config: Mapping[str, Any]) -> str |
             terms.extend(item for item in aliases if isinstance(item, str))
         if any(_term_matches(query_value, term) for term in terms):
             matches.add(scope)
-    return next(iter(matches)) if len(matches) == 1 else None
+    return sorted(matches, key=str.casefold)
 
 
 def _resolved_path(value: str | Path, *, base_dir: Path | None = None) -> Path:
@@ -320,6 +331,7 @@ def resolve_project_path_scope(
 __all__ = [
     "ScopeError",
     "normalize_scopes",
+    "project_scope_matches_text",
     "register_scope_nodes",
     "resolve_project_path_scope",
     "resolve_query_project_scope",

{memleaf-0.2.6 → memleaf-0.2.8}/src/memleaf/validation.py

@@ -6,7 +6,12 @@ import json
 import re
 from typing import Any, Iterable, Mapping
 
-from .scope_state import ScopeError, validate_scope_key, validate_scope_registry
+from .scope_state import (
+    ScopeError,
+    project_scope_matches_text,
+    validate_scope_key,
+    validate_scope_registry,
+)
 
 
 class ModelOutputError(ValueError):
@@ -41,6 +46,7 @@ class ModelOutputError(ValueError):
 MEMORY_TYPES = frozenset(("preference", "fact", "project", "todo", "event", "identity", "other"))
 SCOPE_SOURCES = frozenset(("model", "user", "session_context", "insufficient_context"))
 TODO_STATUSES = frozenset(("active", "completed", "cancelled"))
+NO_CHANGE_DECISION = "NO_CHANGE"
 MODEL_VALIDATION_REASONS = frozenset(
     ("empty_content", "invalid_json", "schema_violation", "response_shape")
 )
@@ -53,6 +59,9 @@ MODEL_VALIDATION_DETAILS = frozenset(
         "duplicate_candidate_id",
         "duplicate_update_target",
         "mixed_project_scopes",
+        "update_target_type_mismatch",
+        "scope_not_grounded",
+        "scope_drift",
         "invalid_evidence",
         "invalid_flags",
         "invalid_type",
@@ -248,6 +257,78 @@ def _reject_mixed_project_scopes(scopes: Iterable[str]) -> None:
         )
 
 
+def _reject_ungrounded_project_scope(
+    memory: str,
+    scopes: Iterable[str],
+    scope_registry: Mapping[str, Any] | None,
+) -> None:
+    """Require model-attributed project scopes to be named by this candidate.
+
+    The gate's memory text is the only candidate-local attribution evidence.
+    Session background, other events, and related-memory bodies are
+    intentionally excluded so an aggregate mailbox turn cannot lend a
+    project name to an unrelated candidate.
+    """
+
+    selected = {
+        scope.casefold(): scope
+        for scope in scopes
+        if isinstance(scope, str) and scope.partition(":")[0] == "project"
+    }
+    if not selected:
+        return
+    try:
+        registry = validate_scope_registry(scope_registry or {})
+    except ScopeError as error:
+        raise ModelOutputError("invalid scope registry", validation_detail="invalid_scope") from error
+
+    # Resolve a selected project key through a registered canonical scope when
+    # it is an alias (including the full ``project:...`` spelling).  A merged
+    # source may remain selectable as an alias of its target; treating both as
+    # separate projects would incorrectly reject that otherwise valid update.
+    selected_owners: dict[str, str] = {}
+    for selected_scope in selected.values():
+        selected_owner = selected_scope
+        for registered_scope, node in registry.items():
+            if not registered_scope.startswith("project:"):
+                continue
+            terms = {
+                registered_scope.casefold(),
+                registered_scope.split(":", 1)[1].casefold(),
+            }
+            aliases = node.get("aliases", [])
+            if isinstance(aliases, list):
+                for alias in aliases:
+                    if isinstance(alias, str):
+                        alias_key = alias.casefold()
+                        terms.add(alias_key)
+                        if ":" not in alias:
+                            terms.add(f"project:{alias_key}")
+            if selected_scope.casefold() in terms:
+                selected_owner = registered_scope
+                break
+        selected_owners[selected_scope.casefold()] = selected_owner.casefold()
+
+    # Include unregistered selections as temporary nodes: new scopes must be
+    # grounded by their own project name, while registered scopes also accept
+    # their configured aliases.  All known projects are checked to reject an
+    # otherwise ambiguous candidate that names multiple projects.
+    candidate_registry = dict(registry)
+    for scope in selected.values():
+        owner = selected_owners[scope.casefold()]
+        if owner not in {registered.casefold() for registered in registry}:
+            candidate_registry.setdefault(scope, {})
+    matches = {
+        scope.casefold()
+        for scope in project_scope_matches_text(memory, {"scopes": candidate_registry})
+    }
+    if len(matches) != 1 or not matches.intersection(selected_owners.values()):
+        raise ModelOutputError(
+            "model project scope is not grounded by this candidate",
+            validation_detail="scope_not_grounded",
+        )
+
+
 def _memory_id(value: Any, field: str) -> str:
     result = _string(value, field)
     if "/" in result or "\\" in result or result in (".", ".."):
@@ -419,6 +500,7 @@ def validate_gate_output(
     current_event_keys: Iterable[Any] | None = None,
     related_memory_ids: Iterable[Any] | None = None,
     related_memory_types: Mapping[str, Any] | None = None,
+    scope_registry: Mapping[str, Any] | None = None,
 ) -> dict[str, Any]:
     """Validate and return a normalized gate object without writing anything."""
 
@@ -540,7 +622,7 @@ def validate_gate_output(
             if target_type is not None and candidate_type != target_type:
                 raise ModelOutputError(
                     "candidate type does not match update target",
-                    validation_detail="invalid_type",
+                    validation_detail="update_target_type_mismatch",
                 )
             item["update_memory_id"] = update_id
         for target_field in ("duplicate_memory_id", "update_memory_id"):
@@ -558,6 +640,8 @@ def validate_gate_output(
         item["scopes"] = _scopes(item["scopes"], item["scope_source"])
         if item["worth"]:
             _reject_mixed_project_scopes(item["scopes"])
+            if item["scope_source"] == "model":
+                _reject_ungrounded_project_scope(item["memory"], item["scopes"], scope_registry)
         if "reason" in item:
             _string(item["reason"], "reason", nonempty=False)
             if len(item["reason"]) > 30:
@@ -574,6 +658,7 @@ def parse_gate_output(
     current_event_keys: Iterable[Any] | None = None,
     related_memory_ids: Iterable[Any] | None = None,
     related_memory_types: Mapping[str, Any] | None = None,
+    scope_registry: Mapping[str, Any] | None = None,
 ) -> dict[str, Any]:
     try:
         parsed = parse_strict_json(raw)
@@ -588,6 +673,7 @@ def parse_gate_output(
             current_event_keys=current_event_keys,
             related_memory_ids=related_memory_ids,
             related_memory_types=related_memory_types,
+            scope_registry=scope_registry,
         )
     except ModelOutputError as error:
         if error.validation_reason is None:
@@ -636,6 +722,9 @@ def validate_summarize_output(
     expected_type: str | None = None,
     expected_update_memory_id: str | None = None,
     expected_target_type: str | None = None,
+    expected_scopes: Iterable[Any] | None = None,
+    expected_scope_source: str | None = None,
+    allow_no_change: bool = False,
 ) -> dict[str, Any]:
     """Validate one atomic memory summary; this function has no filesystem effects."""
 
@@ -643,6 +732,13 @@ def validate_summarize_output(
         value = parse_strict_json(value)
     if not isinstance(value, Mapping):
         raise ModelOutputError("summarize output must be an object", validation_detail="root_shape")
+    if "decision" in value:
+        if allow_no_change and dict(value) == {"decision": NO_CHANGE_DECISION}:
+            return {"decision": NO_CHANGE_DECISION}
+        raise ModelOutputError(
+            "NO_CHANGE is only valid as the exact automatic no-write response",
+            validation_detail="unknown_fields",
+        )
     allowed = {
         "memory_id",
         "update_memory_id",
@@ -680,8 +776,24 @@ def validate_summarize_output(
             "summary type does not match update target",
             validation_detail="invalid_type",
         )
-    item["scopes"] = _scopes(item["scopes"], item.get("scope_source"))
+    summary_scope_source = item.get("scope_source")
+    if "scope_source" not in item and expected_scope_source is not None:
+        summary_scope_source = expected_scope_source
+        item["scope_source"] = summary_scope_source
+    item["scopes"] = _scopes(item["scopes"], summary_scope_source)
     _reject_mixed_project_scopes(item["scopes"])
+    if expected_scopes is not None:
+        expected_scope_values = list(expected_scopes)
+        if item["scopes"] != expected_scope_values:
+            raise ModelOutputError(
+                "summary scopes differ from gate candidate",
+                validation_detail="scope_drift",
+            )
+    if expected_scope_source is not None and summary_scope_source != expected_scope_source:
+        raise ModelOutputError(
+            "summary scope source differs from gate candidate",
+            validation_detail="scope_drift",
+        )
     item["scope_operations"] = _scope_operations(
         item.get("scope_operations"),
         summary_scopes=item["scopes"],
@@ -769,6 +881,9 @@ def parse_summarize_output(
     expected_type: str | None = None,
     expected_update_memory_id: str | None = None,
     expected_target_type: str | None = None,
+    expected_scopes: Iterable[Any] | None = None,
+    expected_scope_source: str | None = None,
+    allow_no_change: bool = False,
 ) -> dict[str, Any]:
     try:
         parsed = parse_strict_json(raw)
@@ -786,6 +901,9 @@ def parse_summarize_output(
             expected_type=expected_type,
             expected_update_memory_id=expected_update_memory_id,
             expected_target_type=expected_target_type,
+            expected_scopes=expected_scopes,
+            expected_scope_source=expected_scope_source,
+            allow_no_change=allow_no_change,
         )
     except ModelOutputError as error:
         if error.validation_reason is None:
@@ -893,6 +1011,7 @@ __all__ = [
     "MODEL_VALIDATION_DETAILS",
     "MODEL_VALIDATION_REASONS",
     "ModelOutputError",
+    "NO_CHANGE_DECISION",
     "SCOPE_SOURCES",
     "TODO_STATUSES",
     "parse_gate",

{memleaf-0.2.6 → memleaf-0.2.8/src/memleaf.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: memleaf
-Version: 0.2.6
+Version: 0.2.8
 Summary: A local-first Markdown memory core for AI agents
 Author: memleaf contributors
 License-Expression: MIT
@@ -23,8 +23,8 @@ Dynamic: license-file
 
 [English](README.en.md) · [PyPI](https://pypi.org/project/memleaf/) · [GitHub](https://github.com/miffyblueboo/memleaf)
 
-> **当前版本：0.2.6。**
-> 核心库、Vault、stdio MCP Server、初始化 CLI、模型路由、提炼流程、受控检索协议和宿主适配器已经实现。memleaf 0.2.6 通过 PyPI 分发。
+> **当前版本：0.2.8。**
+> 核心库、Vault、stdio MCP Server、初始化 CLI、模型路由、提炼流程、受控检索协议和宿主适配器已经实现。memleaf 0.2.8 通过 PyPI 分发。
 > **当前版本支持 Hermes 和 Codex。** Antigravity（反重力）不检测、不安装、不配置。
 
 ## 项目定位