python-hwpx-automation 7.0.4__tar.gz → 7.2.0__tar.gz

{python_hwpx_automation-7.0.4/src/python_hwpx_automation.egg-info → python_hwpx_automation-7.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-hwpx-automation
-Version: 7.0.4
+Version: 7.2.0
 Summary: Task-oriented HWPX document automation for Python, with an optional MCP adapter.
 Author: Kohkyuhyun
 License-Expression: Apache-2.0
@@ -20,7 +20,7 @@ Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
 License-File: NOTICE
-Requires-Dist: python-hwpx<7,>=6.3.0
+Requires-Dist: python-hwpx<7,>=6.5.0
 Requires-Dist: pydantic<3,>=2.11
 Requires-Dist: cryptography>=42.0
 Requires-Dist: anyio>=4.0
@@ -74,10 +74,8 @@ Dynamic: license-file
 
 <!-- release-state: release-approved -->
 > [!NOTE]
-> 아직 공개되지 않은 7.0.4 후보입니다.
-> 후보 조합: `python-hwpx 6.3.0 → python-hwpx-automation 7.0.4 → hwpx-plugin 2.1.0`.
-> 공개 트레인: `python-hwpx 6.3.0 → python-hwpx-automation 7.0.3 → hwpx-plugin 2.0.3`.
-> 계약은 `8c278ebd5becba08`이며 도구 표면과 최소 호환 버전은 유지합니다.
+> 발행이 승인된 7.2.0 후보 (release-approved; 원격 발행 대기): `python-hwpx 6.5.0 → python-hwpx-automation 7.2.0 → hwpx-plugin 2.3.0` (`5e5c23651f92785a`).
+> 공개 트레인: `python-hwpx 6.4.0 → python-hwpx-automation 7.1.0 → hwpx-plugin 2.2.0` (`ba0211fc854a0a97`).
 > 공개 좌표는 core·automation PyPI와 plugin GitHub
 > Release·marketplace·실제 marketplace 설치 관찰 후에만 승격합니다.
 > [릴리스 runbook](docs/release-runbook.md)
@@ -134,7 +132,7 @@ Gemini CLI는 `~/.gemini/settings.json`, Cursor·Windsurf는 각 에디터의 MC
       "command": "uvx",
       "args": [
         "--from",
-        "python-hwpx-automation[mcp]==7.0.3",
+        "python-hwpx-automation[mcp]==7.1.0",
         "hwpx-automation-mcp"
       ],
       "env": {
@@ -241,3 +239,5 @@ python scripts/run_conformance.py run \
 ## License · Maintainer
 
 Apache-2.0 ([LICENSE](LICENSE) · [NOTICE](NOTICE)) — **Kohkyuhyun** [@airmang](https://github.com/airmang) · [kokyuhyun@hotmail.com](mailto:kokyuhyun@hotmail.com)
+
+기존 문서 수정은 [조회·대상 확정·보존 저장·검증 안내](docs/existing-edit-navigation.md)를 먼저 확인하세요. 새 문서는 document-plan, 에이전트 연결은 선택 MCP와 호스트 플러그인을 사용합니다.

{python_hwpx_automation-7.0.4 → python_hwpx_automation-7.2.0}/README.md

@@ -14,10 +14,8 @@
 
 <!-- release-state: release-approved -->
 > [!NOTE]
-> 아직 공개되지 않은 7.0.4 후보입니다.
-> 후보 조합: `python-hwpx 6.3.0 → python-hwpx-automation 7.0.4 → hwpx-plugin 2.1.0`.
-> 공개 트레인: `python-hwpx 6.3.0 → python-hwpx-automation 7.0.3 → hwpx-plugin 2.0.3`.
-> 계약은 `8c278ebd5becba08`이며 도구 표면과 최소 호환 버전은 유지합니다.
+> 발행이 승인된 7.2.0 후보 (release-approved; 원격 발행 대기): `python-hwpx 6.5.0 → python-hwpx-automation 7.2.0 → hwpx-plugin 2.3.0` (`5e5c23651f92785a`).
+> 공개 트레인: `python-hwpx 6.4.0 → python-hwpx-automation 7.1.0 → hwpx-plugin 2.2.0` (`ba0211fc854a0a97`).
 > 공개 좌표는 core·automation PyPI와 plugin GitHub
 > Release·marketplace·실제 marketplace 설치 관찰 후에만 승격합니다.
 > [릴리스 runbook](docs/release-runbook.md)
@@ -74,7 +72,7 @@ Gemini CLI는 `~/.gemini/settings.json`, Cursor·Windsurf는 각 에디터의 MC
       "command": "uvx",
       "args": [
         "--from",
-        "python-hwpx-automation[mcp]==7.0.3",
+        "python-hwpx-automation[mcp]==7.1.0",
         "hwpx-automation-mcp"
       ],
       "env": {
@@ -181,3 +179,5 @@ python scripts/run_conformance.py run \
 ## License · Maintainer
 
 Apache-2.0 ([LICENSE](LICENSE) · [NOTICE](NOTICE)) — **Kohkyuhyun** [@airmang](https://github.com/airmang) · [kokyuhyun@hotmail.com](mailto:kokyuhyun@hotmail.com)
+
+기존 문서 수정은 [조회·대상 확정·보존 저장·검증 안내](docs/existing-edit-navigation.md)를 먼저 확인하세요. 새 문서는 document-plan, 에이전트 연결은 선택 MCP와 호스트 플러그인을 사용합니다.

{python_hwpx_automation-7.0.4 → python_hwpx_automation-7.2.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "python-hwpx-automation"
-version = "7.0.4"
+version = "7.2.0"
 description = "Task-oriented HWPX document automation for Python, with an optional MCP adapter."
 readme = "README.md"
 authors = [{ name = "Kohkyuhyun" }]
@@ -31,7 +31,7 @@ dependencies = [
     # breaking an already-published automation line. WP-F (engine surface 6.0):
     # this line adapted to the new namespace surface (design §9 WP-F) — the dev0
     # pre-release qualifier tracks core's own not-yet-tagged 6.0 identity.
-    "python-hwpx>=6.3.0,<7", # HWPX object model, OXML/OPC, format-native primitives
+    "python-hwpx>=6.5.0,<7", # HWPX object model, OXML/OPC, format-native primitives
 
     "pydantic>=2.11,<3",
     "cryptography>=42.0",

{python_hwpx_automation-7.0.4 → python_hwpx_automation-7.2.0}/src/hwpx_automation/handlers/content_edit.py

@@ -3,7 +3,9 @@
 
 from __future__ import annotations
 
+from pathlib import Path
 from typing import Any
+import zlib
 
 from .. import quality as quality_contract
 from ..core.content import (
@@ -805,9 +807,40 @@ def insert_picture(
     return _with_save_verification(result, verification)
 
 
+def _validate_workspace_image(data: bytes, image_format: str) -> None:
+    """Check bounded file input before inserting its bytes into a package."""
+    fmt = image_format.casefold().lstrip(".")
+    if fmt == "png":
+        if not data.startswith(b"\x89PNG\r\n\x1a\n"):
+            raise ValueError("image_filename content does not match image_format")
+        offset = 8
+        chunks: list[bytes] = []
+        while offset + 12 <= len(data):
+            length = int.from_bytes(data[offset:offset + 4], "big")
+            end = offset + 12 + length
+            if end > len(data):
+                break
+            kind = data[offset + 4:offset + 8]
+            payload = data[offset + 8:offset + 8 + length]
+            crc = int.from_bytes(data[end - 4:end], "big")
+            if zlib.crc32(kind + payload) != crc:
+                break
+            chunks.append(kind)
+            offset = end
+            if kind == b"IEND":
+                if length == 0 and offset == len(data) and chunks[0] == b"IHDR" and b"IDAT" in chunks:
+                    return
+                break
+    elif fmt in {"jpg", "jpeg"} and data.startswith(b"\xff\xd8\xff") and data.endswith(b"\xff\xd9"):
+        return
+    elif fmt == "gif" and data.startswith((b"GIF87a", b"GIF89a")) and data.endswith(b";"):
+        return
+    raise ValueError("image_filename content does not match image_format or is incomplete")
+
+
 def replace_picture(
     filename: str,
-    image_base64: str,
+    image_base64: str | None = None,
     image_format: str = "png",
     picture_index: int = 0,
     binary_item_id_ref: str | None = None,
@@ -815,15 +848,27 @@ def replace_picture(
     output: str | None = None,
     dry_run: bool = False,
     expected_revision: str = None,
+    image_filename: str | None = None,
 ) -> dict:
-    """그림 객체의 geometry를 유지하고 연결된 이미지 asset만 교체합니다."""
+    """그림 geometry를 유지하고 base64 또는 workspace 이미지 파일로 교체합니다."""
+    if (image_base64 is None) == (image_filename is None):
+        raise ValueError("provide exactly one of image_base64 or image_filename")
     path = resolve_path(filename)
     guard = _revision_guard(path, expected_revision)
     if guard is not None:
         return guard
     target_path = resolve_path(output) if output else path
+    if image_filename is not None:
+        image_path = Path(resolve_path(image_filename))
+        with image_path.open("rb") as image_file:
+            image_data = image_file.read(20 * 1024 * 1024 + 1)
+        if len(image_data) > 20 * 1024 * 1024:
+            raise ValueError("image_filename must be at most 20 MiB")
+        _validate_workspace_image(image_data, image_format)
+    else:
+        assert image_base64 is not None
+        image_data = _decode_image_base64(image_base64)
     doc = open_doc(path)
-    image_data = _decode_image_base64(image_base64)
     # media.replace_picture now returns a frozen PictureReplacement (design
     # §2.4) whose own .to_dict() uses different field names than this op's
     # established response contract (old_binaryItemIDRef/new_binaryItemIDRef/

{python_hwpx_automation-7.0.4 → python_hwpx_automation-7.2.0}/src/hwpx_automation/handlers/read_export.py

@@ -38,6 +38,7 @@ from ..upstream import (
     open_document,
 )
 from ..utils.helpers import default_max_chars, resolve_path, truncate_response
+from ..utils.read_budget import bound_document_summary
 from ._shared import _with_document_state
 
 _OUTPUT_MODES = {"full", "chunks"}
@@ -920,7 +921,8 @@ def get_document_map(
             "hwpx_extract_json",
         ],
     }
-    return _with_document_state(result, path)
+    result = _with_document_state(result, path)
+    return bound_document_summary(result) if summary_mode else result
 
 
 def hwpx_to_markdown(

{python_hwpx_automation-7.0.4 → python_hwpx_automation-7.2.0}/src/hwpx_automation/identity.json

@@ -5,22 +5,30 @@
   "releaseState": {
     "status": "release-approved",
     "candidate": {
-      "pythonHwpx": "6.3.0",
+      "pythonHwpx": "6.5.0",
       "canonicalDistribution": "python-hwpx-automation",
-      "canonicalAutomation": "7.0.4",
+      "canonicalAutomation": "7.2.0",
       "compatibilityDistribution": "hwpx-mcp-server",
-      "compatibility": "7.0.4",
-      "plugin": "2.1.0",
-      "contractHash": "8c278ebd5becba08"
+      "compatibility": "7.2.0",
+      "plugin": "2.3.0",
+      "contractHash": "5e5c23651f92785a"
     },
     "currentPublic": {
-      "pythonHwpx": "6.3.0",
+      "pythonHwpx": "6.4.0",
       "primaryDistribution": "python-hwpx-automation",
-      "primaryApplication": "7.0.3",
-      "plugin": "2.0.3",
-      "contractHash": "8c278ebd5becba08"
+      "primaryApplication": "7.1.0",
+      "plugin": "2.2.0",
+      "contractHash": "ba0211fc854a0a97"
     },
-    "promotionGate": "Three states are mandatory: unreleased-candidate while auditing; release-approved only after separate owner approval and while currentPublic still names the previously observed coherent stack; released only in a follow-up commit after remote truth is observed for core, canonical automation, the compatibility distribution, the plugin GitHub release, the marketplace entry, and a real marketplace install. The automation tag workflow publishes only release-approved, leaves currentPublic unchanged, and hands an attached receipt to plugin publication."
+    "promotionGate": "Three states are mandatory: unreleased-candidate while auditing; release-approved only after separate owner approval and while currentPublic still names the previously observed coherent stack; released only in a follow-up commit after remote truth is observed for core, canonical automation, the compatibility distribution, the plugin GitHub release, the marketplace entry, and a real marketplace install. The automation tag workflow publishes only release-approved, leaves currentPublic unchanged, and hands an attached receipt to plugin publication.",
+    "previousPublic": {
+      "pythonHwpx": "6.4.0",
+      "primaryDistribution": "python-hwpx-automation",
+      "primaryApplication": "7.1.0",
+      "plugin": "2.2.0",
+      "contractHash": "ba0211fc854a0a97"
+    },
+    "publicationEvidence": null
   },
   "releaseFloors": {
     "description": "Frozen historical coordinates. Unlike releaseState these do not advance train by train; they pin the pre-6.0 legacy compatibility cap that every core 5 resolution must be observed against. They live here so the release path has no version literal that is not declared data.",

{python_hwpx_automation-7.0.4 → python_hwpx_automation-7.2.0}/src/hwpx_automation/mixed_form.py

@@ -29,7 +29,6 @@ from hwpx_automation.office.agent import (
     plan_mixed_form_fill,
     validate_mixed_form_plan,
 )
-from hwpx.quality import SavePipeline
 from pydantic import (
     BaseModel,
     ConfigDict,
@@ -40,6 +39,7 @@ from pydantic import (
 )
 
 from .execution_lock import PUBLIC_MUTATION_LOCK
+from .office.agent._batch_publication import GuardedSavePipeline
 from .office.rendering import resolve_hancom_backend
 from .storage import build_hwpx_open_safety_report
 from .utils.helpers import resolve_path
@@ -584,7 +584,7 @@ class _FailurePreimagePreserver:
         return True
 
 
-class _WorkspaceSavePipeline(SavePipeline):
+class _WorkspaceSavePipeline(GuardedSavePipeline):
     """Run the core quality gate but publish through an identity-bound path."""
 
     def __init__(
@@ -1652,7 +1652,7 @@ def _finalize_canonical_apply(
     verification = copy.deepcopy(core_verification)
     response_ok = bool(response.get("ok")) and bool(source_report["ok"])
     missing_fresh_publication = bool(
-        response_ok
+        (response_ok or core_verification.get("publication", {}).get("ownershipMissing"))
         and not dry_run
         and not replayed
         and publication is None

python_hwpx_automation-7.2.0/src/hwpx_automation/office/agent/_batch_publication.py

@@ -0,0 +1,185 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Revision-bound publication for the existing agent command executor.
+
+Quality remains in core SavePipeline; identity/CAS and rollback reuse the
+application workspace writer. Explicit Python paths do not acquire a cwd policy.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+from typing import Any
+
+from hwpx.quality import SavePipeline
+
+from ...workspace import WorkspaceOutputGuard, WorkspaceResolver
+from .model import AgentContractError
+
+
+class GuardedSavePipeline(SavePipeline):
+    """Application publisher carrying the identity of its own publication.
+
+    Existing domain publishers subclass this seam so the batch can retain their
+    guarded publication and recovery receipts after running their quality gate.
+    """
+
+    publication: WorkspaceOutputGuard | None = None
+
+
+class _BatchSavePipeline(GuardedSavePipeline):
+    def __init__(self, binding: BatchPublication) -> None:
+        super().__init__()
+        self.binding = binding
+        self.publication = None
+
+    def _publish(self, data: bytes, output_path: Any, output_stream: Any) -> str:
+        if (
+            output_stream is not None
+            or Path(output_path).resolve() != self.binding.output
+        ):
+            raise AgentContractError(
+                "verification_failed", "unexpected batch publication destination"
+            )
+        guard = self.binding.workspace.materialize_output_guard(
+            self.binding.output_guard
+        )
+        try:
+            self.publication = self.binding.workspace.atomic_publish_bytes(guard, data)
+        except BaseException:
+            self.binding.workspace.cleanup_owned_parent_directories(guard)
+            raise
+        return str(self.publication.path)
+
+
+def _existing_parent(path: Path) -> Path:
+    parent = path.parent
+    while not parent.exists() and parent != parent.parent:
+        parent = parent.parent
+    return parent
+
+
+class BatchPublication:
+    """One disposable request's input snapshot and destination precondition."""
+
+    def __init__(self, source: Path, output: Path) -> None:
+        self.source_argument = source.absolute()
+        self.source = source.resolve()
+        self.output = output.resolve()
+        self.workspace = WorkspaceResolver.from_roots(
+            [self.source.parent, _existing_parent(self.output)],
+            source="agent-explicit-paths",
+        )
+        self.source_guard = self.workspace.capture_output(self.source)
+        self.input_data = self.workspace.read_guarded_bytes(self.source_guard)
+        self.output_guard = self.workspace.capture_output_precondition(self.output)
+        self.output_data = (
+            self.workspace.read_guarded_bytes(self.output_guard)
+            if isinstance(self.output_guard, WorkspaceOutputGuard)
+            and self.output_guard.target_existed
+            else None
+        )
+        self.publisher: GuardedSavePipeline | None = None
+        self.publication_attempted = False
+
+    def assert_current_source(self) -> None:
+        try:
+            if self.source_argument.resolve() != self.source:
+                raise OSError("source path was rebound")
+            current = self.workspace.read_guarded_bytes(self.source_guard)
+            if current != self.input_data:
+                raise OSError("source bytes changed")
+        except OSError as exc:
+            raise AgentContractError(
+                "stale_revision",
+                "input changed during editing; read its current revision before retrying",
+                target="batch.input.filename",
+            ) from exc
+
+    def publish(
+        self, data: bytes, pipeline: SavePipeline, verification: dict[str, Any]
+    ) -> str:
+        self.assert_current_source()
+        if not isinstance(pipeline, GuardedSavePipeline) and type(pipeline)._publish is not SavePipeline._publish:
+            raise AgentContractError(
+                "unsupported_operation",
+                "custom publishers must provide an identity-bound publication receipt",
+                target="savePipeline",
+            )
+        self.publisher = (
+            pipeline
+            if isinstance(pipeline, GuardedSavePipeline)
+            else _BatchSavePipeline(self)
+        )
+        self.publisher.publication = None
+        self.publication_attempted = True
+        written = self.publisher._publish(data, self.output, None)
+        publication = self.publisher.publication
+        if publication is None:
+            verification["publication"] = {
+                "ok": False,
+                "ownershipMissing": True,
+                "rolledBack": False,
+            }
+            raise AgentContractError(
+                "verification_failed",
+                "publisher did not provide an owned output receipt",
+            )
+        if self.source != self.output:
+            self.assert_current_source()
+        if self.workspace.read_guarded_bytes(publication) != data:
+            raise AgentContractError(
+                "verification_failed",
+                "published bytes differ from the verified candidate",
+            )
+        verification["publication"] = {
+            "ok": True,
+            "inputRechecked": True,
+            "outputPreconditionChecked": True,
+        }
+        return str(written)
+
+    def rollback(self, verification: dict[str, Any]) -> None:
+        publication = self.publisher.publication if self.publisher is not None else None
+        if publication is None:
+            unchanged = not self.publication_attempted
+            if not unchanged:
+                try:
+                    unchanged = (
+                        self.workspace.capture_output_precondition(self.output)
+                        == self.output_guard
+                    )
+                except OSError:
+                    unchanged = False
+            verification.setdefault(
+                "publication",
+                {
+                    "ok": False,
+                    "candidatePublished": None if self.publication_attempted else False,
+                    "rolledBack": unchanged,
+                },
+            )
+            return
+        restored = False
+        try:
+            if self.output_data is None:
+                self.workspace.remove_output(publication)
+                restored = self.workspace.cleanup_owned_parent_directories(publication)
+            else:
+                mode = (
+                    self.output_guard.target_mode
+                    if isinstance(self.output_guard, WorkspaceOutputGuard)
+                    else None
+                )
+                self.workspace.atomic_publish_bytes(
+                    publication, self.output_data, mode=mode
+                )
+                restored = True
+        except (OSError, RuntimeError):
+            # The publication capability never authorizes replacing another
+            # writer's newer file. Do not claim that this request rolled it back.
+            pass
+        verification["publication"] = {
+            "ok": False,
+            "candidatePublished": True,
+            "rolledBack": restored,
+        }

{python_hwpx_automation-7.0.4 → python_hwpx_automation-7.2.0}/src/hwpx_automation/office/agent/_batch_verification.py

@@ -22,6 +22,7 @@ from hwpx.mutation_report import member_diff_bytes
 from hwpx.quality import QualityPolicy, SavePipeline
 from hwpx.tools.package_validator import validate_editor_open_safety
 
+from ._batch_publication import BatchPublication
 from .document import HwpxAgentDocument
 from .model import AgentBatchResult, AgentContractError, AgentError
 
@@ -82,7 +83,7 @@ def _failure_result(
         output_filename = str(output.get("filename") or "")
     return AgentBatchResult(
         ok=False,
-        rolled_back=True,
+        rolled_back=bool((verification or {}).get("publication", {}).get("rolledBack", True)),
         dry_run=bool(raw.get("dryRun", False)),
         input_revision=input_revision,
         document_revision=input_revision,
@@ -291,7 +292,7 @@ def _apply_commands_build_candidate_report(
             "package": safety_dict["validatePackage"],
             "reopen": safety_dict["reopen"],
             "openSafety": safety_dict,
-            "semanticDiff": {"ok": True, "changeCount": len(semantic_changes)},
+            "semanticDiff": {"ok": None, "changeCount": len(semantic_changes), "basis": "declared-command-log"},
             "bytePreservation": byte_report,
         }
     )
@@ -343,6 +344,7 @@ def _apply_commands_run_save_pipeline(
     requirements: set[str],
     save_pipeline: SavePipeline | None,
     verification: dict[str, Any],
+    publication: BatchPublication,
 ) -> Any:
     pipeline = save_pipeline or SavePipeline()
     quality_policy = _quality_policy(normalized["quality"])
@@ -356,11 +358,11 @@ def _apply_commands_run_save_pipeline(
         )
     quality_report = pipeline.run(
         candidate_data,
-        output_path=None if normalized["dryRun"] else output_path,
+        output_path=None,
         quality=quality_policy,
         before=input_path,
         reference_document=document,
-        publish="never" if normalized["dryRun"] else "on_pass",
+        publish="never",
         source_label="agent.apply_document_commands",
     )
     verification["savePipeline"] = quality_report.to_dict()
@@ -380,4 +382,8 @@ def _apply_commands_run_save_pipeline(
         raise AgentContractError(
             "verification_failed", "SavePipeline rejected the candidate", target="savePipeline"
         )
+    publication.assert_current_source()
+    if not normalized["dryRun"]:
+        quality_report.output_path = publication.publish(candidate_data, pipeline, verification)
+        verification["savePipeline"] = quality_report.to_dict()
     return quality_report