@hunter-harness/workflow-harness 0.2.18 → 0.2.19

package/harness/bundles/general/cursor/scripts/harness_paths.py

@@ -20,7 +20,7 @@ import re
 import subprocess
 import sys
 from pathlib import Path
-from typing import Any
+from typing import Any, Sequence
 
 SCRIPTS_DIR = Path(__file__).resolve().parent
 if str(SCRIPTS_DIR) not in sys.path:
@@ -289,6 +289,175 @@ def assert_path_within(
     return resolved
 
 
+class CleanupTopologyError(ValueError):
+    """Raised when cleanup would delete or traverse into state/archive roots."""
+
+    code = "CLEANUP_TOPOLOGY_REFUSED"
+
+
+# Skip descending into these when scanning for junction/symlink escapes.
+# Matches integration HEAVY_WORKTREE_ROOTS plus .git (topology scan only).
+CLEANUP_WALK_SKIP_DIRS = frozenset(
+    {
+        "node_modules",
+        ".venv",
+        "venv",
+        "build",
+        "dist",
+        "target",
+        ".cache",
+        "__pycache__",
+        ".git",
+    }
+)
+
+
+def assert_cleanup_safe(
+    cleanup_root: Path,
+    state_roots: Sequence[Path] | None = None,
+    archive_roots: Sequence[Path] | None = None,
+) -> Path:
+    """Refuse cleanup when state/archive roots resolve inside the cleanup tree.
+
+    Also refuses when a junction/symlink under ``cleanup_root`` resolves into a
+    protected state/archive root (Windows junction / symlink escape).
+    """
+    cleanup = Path(cleanup_root).resolve()
+    protected: list[Path] = []
+    for raw in [*(state_roots or ()), *(archive_roots or ())]:
+        try:
+            protected.append(Path(raw).resolve())
+        except OSError as exc:
+            raise CleanupTopologyError(
+                f"CLEANUP_TOPOLOGY_REFUSED: cannot resolve protected path {raw}: {exc}"
+            ) from exc
+
+    for prot in protected:
+        if prot == cleanup or prot.is_relative_to(cleanup):
+            raise CleanupTopologyError(
+                "CLEANUP_TOPOLOGY_REFUSED: protected path "
+                f"{prot} is inside cleanup root {cleanup}"
+            )
+
+    if cleanup.is_dir():
+        for dirpath, dirnames, filenames in os.walk(cleanup, followlinks=False):
+            # Inspect this level (including heavy names) for link escapes, then
+            # prune descent so we do not walk install/cache trees.
+            entries = [*dirnames, *filenames]
+            dirnames[:] = [d for d in dirnames if d not in CLEANUP_WALK_SKIP_DIRS]
+            for name in entries:
+                child = Path(dirpath) / name
+                try:
+                    is_link = child.is_symlink()
+                except OSError:
+                    continue
+                # Junctions often report as directories; resolve and compare.
+                try:
+                    resolved_child = child.resolve()
+                except OSError:
+                    continue
+                escaped = not (
+                    resolved_child == cleanup or resolved_child.is_relative_to(cleanup)
+                )
+                if not (is_link or escaped):
+                    continue
+                for prot in protected:
+                    if (
+                        resolved_child == prot
+                        or resolved_child.is_relative_to(prot)
+                        or prot.is_relative_to(resolved_child)
+                    ):
+                        raise CleanupTopologyError(
+                            "CLEANUP_TOPOLOGY_REFUSED: cleanup path "
+                            f"{child} resolves to protected {resolved_child}"
+                        )
+    return cleanup
+
+
+_FORMAL_SNAPSHOT_DIRS = ("plans", "spec", "evidence", "reports", "meta")
+_FORMAL_SNAPSHOT_FILES = ("events.ndjson",)
+
+
+def snapshot_change_formal_layer(
+    project_root: Path,
+    change_id: str,
+    *,
+    change_dir: Path | None = None,
+) -> dict[str, Any]:
+    """Copy change formal layer into ``.harness/cache/change-snapshots/<change>/``.
+
+    Destination is always under the main project / common harness root — never
+    under a feature worktree path alone.
+    """
+    import shutil
+
+    project = resolve_main_project_root(Path(project_root))
+    src = Path(change_dir).resolve() if change_dir is not None else (
+        project / ".harness" / "changes" / change_id
+    ).resolve()
+    if not src.is_dir():
+        raise FileNotFoundError(f"change dir not found for snapshot: {src}")
+
+    dest = (project / ".harness" / "cache" / "change-snapshots" / change_id).resolve()
+    # Hard guard: never write the snapshot inside the source tree itself.
+    if dest == src or dest.is_relative_to(src):
+        raise ValueError(
+            f"snapshot destination must not be inside change dir: {dest}"
+        )
+    dest.mkdir(parents=True, exist_ok=True)
+
+    copied: list[str] = []
+    for name in _FORMAL_SNAPSHOT_DIRS:
+        source_dir = src / name
+        if not source_dir.is_dir():
+            continue
+        target_dir = dest / name
+        shutil.copytree(source_dir, target_dir, dirs_exist_ok=True)
+        copied.append(name + "/")
+    for name in _FORMAL_SNAPSHOT_FILES:
+        source_file = src / name
+        if not source_file.is_file():
+            continue
+        shutil.copy2(source_file, dest / name)
+        copied.append(name)
+
+    file_hashes: dict[str, str] = {}
+    for path in sorted(dest.rglob("*")):
+        if not path.is_file():
+            continue
+        rel = path.relative_to(dest).as_posix()
+        if rel == "manifest.json":
+            continue
+        digest = hashlib.sha256(path.read_bytes()).hexdigest()
+        file_hashes[rel] = "sha256:" + digest
+
+    manifest = {
+        "schemaVersion": 1,
+        "changeId": change_id,
+        "source": str(src),
+        "destination": str(dest),
+        "projectRoot": str(project),
+        "copied": copied,
+        "files": file_hashes,
+        "fileCount": len(file_hashes),
+    }
+    manifest_path = dest / "manifest.json"
+    manifest_path.write_text(
+        json.dumps(manifest, ensure_ascii=False, indent=2) + "\n",
+        encoding="utf-8",
+        newline="\n",
+    )
+    return {
+        "ok": True,
+        "code": "CHANGE_SNAPSHOT_WRITTEN",
+        "changeId": change_id,
+        "destination": str(dest),
+        "fileCount": len(file_hashes),
+        "copied": copied,
+        "manifestPath": str(manifest_path),
+    }
+
+
 def cmd_resolve_layout(args: argparse.Namespace) -> int:
     layout = resolve_change_layout(Path.cwd(), args.change)
     sys.stdout.write(json.dumps(layout, ensure_ascii=False, indent=2) + "\n")

package/harness/bundles/java/claude-code/.harness-build.json

@@ -2,5 +2,5 @@
   "schemaVersion": 1,
   "overlay": "java",
   "agent": "claude-code",
-  "coreHash": "6e87bc67bd51cbe9"
+  "coreHash": "f01595d3142d34b7"
 }

package/harness/bundles/java/claude-code/harness-apidoc/SKILL.md

@@ -20,7 +20,7 @@ disallowed-tools:
   - Bash(awk *)
   - Bash(curl *)
 ---
-<!-- generated by harness_deploy.py; core=6e87bc67bd51cbe9; overlay=java; agent=claude-code; do not edit -->
+<!-- generated by harness_deploy.py; core=f01595d3142d34b7; overlay=java; agent=claude-code; do not edit -->
 # harness-apidoc — 前端接口文档（Java overlay）
 
 ## Purpose

package/harness/bundles/java/claude-code/harness-archive/SKILL.md

@@ -20,7 +20,7 @@ disallowed-tools:
   - Bash(awk *)
   - Bash(curl *)
 ---
-<!-- generated by harness_deploy.py; core=6e87bc67bd51cbe9; overlay=java; agent=claude-code; do not edit -->
+<!-- generated by harness_deploy.py; core=f01595d3142d34b7; overlay=java; agent=claude-code; do not edit -->
 # harness-archive — 变更归档
 
 ## Purpose
@@ -50,6 +50,12 @@ disallowed-tools:
 - commit 已 push（`git log @{u}..HEAD` 输出为空）
 - **最终 hash 来源**：worktree 模式（`meta/worktree.json` requested=true）下读 ledger 的 `mergeFinalHash`（由 `/harness-submit` 合并段写入）；主目录模式读 submit 的 `final pushed hash`。当前 `git rev-parse HEAD` 须一致
 - test 报告存在或标记"跳过"；review 报告存在或标记"未运行 review"（review 不阻塞归档）
+- **最小必备集（H-4 / `harness_archive.py status` blockers）**：
+  - `plans/*-plan.md`
+  - 非空 `events.ndjson`
+  - `evidence/verification-ledger.json`
+  - 至少一个 test 或 review 报告/证据
+  - 缺失任一项 → `archivable=false`；可从 `.harness/cache/change-snapshots/<change>/` 恢复正式层后再重试
 
 ## Inputs
 

package/harness/bundles/java/claude-code/harness-archive/checklist.md

@@ -20,6 +20,12 @@ description: harness-archive 的归档前检查项和归档后验证项。仅在
 
 - [ ] 未在调用 finalize 前手工追加 archive 阶段边界
 - [ ] 只有一个未归档变更目录（多个时终止或让用户选择）
+- [ ] **最小必备集 blockers**（`harness_archive.py status`）：
+  - [ ] `plans/*-plan.md` 存在
+  - [ ] `events.ndjson` 存在且非空
+  - [ ] `evidence/verification-ledger.json` 存在
+  - [ ] 至少一个 test 或 review 报告/证据
+  - [ ] 若缺失：优先从 `.harness/cache/change-snapshots/<change>/` 恢复，禁止空壳归档
 - [ ] 变更目录下有 plans/ 子目录（至少有计划文件）
 - [ ] `events.ndjson` 存在；执行日志允许由 finalize 从事件流重新渲染（旧 archive 才兼容根目录 `execution-log.md`）
 - [ ] 准备生成 `archive-manifest-before.json`（path/size/sha256）

package/harness/bundles/java/claude-code/harness-codebase-map/SKILL.md

@@ -20,7 +20,7 @@ disallowed-tools:
   - Bash(awk *)
   - Bash(curl *)
 ---
-<!-- generated by harness_deploy.py; core=6e87bc67bd51cbe9; overlay=java; agent=claude-code; do not edit -->
+<!-- generated by harness_deploy.py; core=f01595d3142d34b7; overlay=java; agent=claude-code; do not edit -->
 # harness-codebase-map — 代码库地图
 
 ## Purpose

package/harness/bundles/java/claude-code/harness-knowledge-ingest/SKILL.md

@@ -20,7 +20,7 @@ disallowed-tools:
   - Bash(awk *)
   - Bash(curl *)
 ---
-<!-- generated by harness_deploy.py; core=6e87bc67bd51cbe9; overlay=java; agent=claude-code; do not edit -->
+<!-- generated by harness_deploy.py; core=f01595d3142d34b7; overlay=java; agent=claude-code; do not edit -->
 # harness-knowledge-ingest
 
 从 `.harness/archive/**/reports/final/summary-data.json` 抽取项目知识，生成并维护本地 `.harness/knowledge/` 索引。

package/harness/bundles/java/claude-code/harness-knowledge-query/SKILL.md

@@ -20,7 +20,7 @@ disallowed-tools:
   - Bash(awk *)
   - Bash(curl *)
 ---
-<!-- generated by harness_deploy.py; core=6e87bc67bd51cbe9; overlay=java; agent=claude-code; do not edit -->
+<!-- generated by harness_deploy.py; core=f01595d3142d34b7; overlay=java; agent=claude-code; do not edit -->
 # harness-knowledge-query
 
 并行 change 中查询必须传 `--change <change-id>`；成功后以该 change 的 `meta/knowledge-context.json` 为稳定指针。全局 `context-packs/latest.json` 仅向后兼容，不得用于判断多个 active change 的上下文归属。

package/harness/bundles/java/claude-code/harness-package/SKILL.md

@@ -20,7 +20,7 @@ disallowed-tools:
   - Bash(awk *)
   - Bash(curl *)
 ---
-<!-- generated by harness_deploy.py; core=6e87bc67bd51cbe9; overlay=java; agent=claude-code; do not edit -->
+<!-- generated by harness_deploy.py; core=f01595d3142d34b7; overlay=java; agent=claude-code; do not edit -->
 # harness-package — 增量模块打包（Java overlay）
 
 ## Purpose

package/harness/bundles/java/claude-code/harness-plan/SKILL.md

@@ -20,7 +20,7 @@ disallowed-tools:
   - Bash(awk *)
   - Bash(curl *)
 ---
-<!-- generated by harness_deploy.py; core=6e87bc67bd51cbe9; overlay=java; agent=claude-code; do not edit -->
+<!-- generated by harness_deploy.py; core=f01595d3142d34b7; overlay=java; agent=claude-code; do not edit -->
 # harness-plan — 需求规划
 
 ## Purpose

package/harness/bundles/java/claude-code/harness-review/SKILL.md

@@ -21,7 +21,7 @@ disallowed-tools:
   - Bash(curl *)
   - Bash(codegraph *)
 ---
-<!-- generated by harness_deploy.py; core=6e87bc67bd51cbe9; overlay=java; agent=claude-code; do not edit -->
+<!-- generated by harness_deploy.py; core=f01595d3142d34b7; overlay=java; agent=claude-code; do not edit -->
 # harness-review — 代码审查
 
 ## Purpose

package/harness/bundles/java/claude-code/harness-run/SKILL.md

@@ -21,7 +21,7 @@ disallowed-tools:
   - Bash(curl *)
   - Bash(codegraph *)
 ---
-<!-- generated by harness_deploy.py; core=6e87bc67bd51cbe9; overlay=java; agent=claude-code; do not edit -->
+<!-- generated by harness_deploy.py; core=f01595d3142d34b7; overlay=java; agent=claude-code; do not edit -->
 # harness-run — 需求编码
 
 ## Purpose

package/harness/bundles/java/claude-code/harness-submit/SKILL.md

@@ -20,7 +20,7 @@ disallowed-tools:
   - Bash(awk *)
   - Bash(curl *)
 ---
-<!-- generated by harness_deploy.py; core=6e87bc67bd51cbe9; overlay=java; agent=claude-code; do not edit -->
+<!-- generated by harness_deploy.py; core=f01595d3142d34b7; overlay=java; agent=claude-code; do not edit -->
 # harness-submit — 最终提交（含 worktree 合并）
 
 ## Purpose
@@ -96,12 +96,17 @@ M3. **merge** — `--no-ff` 合并 feature 分支；merge diff 出现其他 Chan
 M4. **verify** — 在 integration worktree 内执行组合态验证；他人提交引入或 ledger 不可复用时必跑
 M5. **push** — 仅在验证身份与远端基线仍匹配时 push；远端漂移 → `TARGET_MOVED` 结构化失败，不继续
 M6. **cleanup** — `git worktree remove --force` 精确路径 + 临时分支 + （push 成功后）保护 ref；释放 integration lock；失败保留 journal 与诊断证据；更新 `worktree.json`（`created=false` + removedAt）
+M6.5. **formal snapshot** — mergeFinalHash 写入后调用 `harness_paths.snapshot_change_formal_layer(projectRoot, changeId)`，备份到主仓 `.harness/cache/change-snapshots/<change>/`（不得写在 feature worktree 内）
 M7. **ledger + 收尾** — 经 `harness_ledger.py record` 写入 `mergeFinalHash`（= journal `pushedHead`）；**`harness_gate.py close --phase merge`**（禁止手工 phase.end）；提示 `/harness-archive`
 
-> Ledger v3（v2 契约 / split-v1 布局起）：`record` 强制顶层身份（`schemaVersion=3/repositoryId/baseCommit/currentHead/diffHash/ownershipHash`，缺失非零退出、不写账本）与 typed metrics；legacy 契约行为不变。详见 `../protocols/ledger-protocol.md` 第十节。
+> Ledger v3（v2 契约 / split-v1 布局起）：`record` 强制顶层身份（`schemaVersion=3/repositoryId/baseCommit/currentHead/diffHash/ownershipHash`，缺失非零退出、`LEDGER_IDENTITY_INVALID`，不写账本）与 typed metrics；legacy 契约行为不变。详见 `../protocols/ledger-protocol.md` 第十节。
 
 正常路径**禁止创建、应用或删除仓库级 stash**。中断恢复：`harness_integration.py status` 读 journal，`recover` 从首个未完成步骤续跑；protection refs 只在 push 成功后的 cleanup 删除。
 
+**失败 txn 回收（abandon）**：verify/push 未成功且 remote 不含 mergeCommit 时，运行 `harness_integration.py abandon --change … --run-id …`：清理 integrationRoot / temp branch / protection refs / lock；**绝不删除 feature worktree**。push 已成功或 remote 已含 merge → `ABANDON_REFUSED`，改走 cleanup。
+
+**清 feature worktree 迁根硬门槛（H-1/H-2）**：删除 feature worktree 前，若 Agent root / cwd 落在待删路径内，必须先 `move_agent_to_root(projectRoot)`（目标为主仓已存在的 `main`/`master`；**禁止**为迁根去 fetch 已删除的 feature 分支）。迁根失败 → **拒绝删除**。清理前调用 `harness_paths.assert_cleanup_safe(cleanupRoot, stateRoots, archiveRoots)`；state/archive（含 junction/symlink 解析后）落在 cleanup 内 → `CLEANUP_TOPOLOGY_REFUSED`。
+
 ### Java 提交前验证补充
 
 通用 `harness-submit` 的最终门禁已统一为 `unitTestFull`（见通用 SKILL.md 步骤 2 / checklist 步骤 2）。Java overlay **不重复定义门禁规则**，只补两点：
@@ -140,6 +145,10 @@ test-tracking manifest 是 ignored test 的唯一强制暂存授权：只允许
 - **`git merge --no-ff` 固定**；禁止 fast-forward
 - **冲突不自动解** → 停下 → 用户手动解 → 确认后继续
 - **`mergeFinalHash`** = 主分支 push 后 HEAD；archive 优先读此字段
+- **迁根后再删 feature WT**：Agent 仍在待删 root 内则拒绝删除；目标分支已删时切主仓 `main`/`master`，禁止强行 fetch
+- **snapshot 后再清**：正式层已写入 `.harness/cache/change-snapshots/<change>/`
+- **拓扑安全**：`assert_cleanup_safe` 拒绝 state/archive ⊆ cleanup（含 junction）
+- **abandon ≠ cleanup**：仅失败 txn；永不删 feature worktree
 
 ### 五、Shell 安全 / 敏感信息 / 证据化
 

package/harness/bundles/java/claude-code/harness-submit/checklist.md

@@ -386,11 +386,26 @@ python harness/scripts/harness_integration.py cleanup --change <change-name> --r
 - 失败保留 journal 与诊断证据，状态 🟡WARN，不得宣称清理完成
 - 若存在 test-tracking manifest，确认其路径已由合并后的目标分支跟踪（`git ls-files --error-unmatch -- <exact-path>`），缺失即停止
 
+**失败 txn 改走 abandon（勿强行 cleanup）**：
+
+```powershell
+python harness/scripts/harness_integration.py abandon --change <change-name> --run-id <run-id> --feature-branch harness/<change-name> --target-branch <主分支> --temp-root <task-temp>
+```
+
+仅当 push ≠ DONE 且 remote 不含 mergeCommit；清理 integration 产物后可换新 run-id 重开。push 成功 → `ABANDON_REFUSED`。
+
 更新 `meta/worktree.json`：`created=false` + `removedAt`/`removedBy`/`removalNote`。
 
+**删除 feature worktree 前（硬门槛）**：
+
+1. 若 Agent root / cwd 落在待删 worktree → 先 `move_agent_to_root(<projectRoot>)`；目标为主仓，feature 分支已删则切 `main`/`master`，禁止 fetch 已不存在分支
+2. 迁根失败 → **停止删除**
+3. `assert_cleanup_safe(cleanupRoot, [.harness/changes/<id>, .harness/state/...], [.harness/archive])`；`CLEANUP_TOPOLOGY_REFUSED` → 停止
+4. mergeFinalHash 已写后先 `snapshot_change_formal_layer`，再清 feature WT
+
 ### 步骤 M7：更新 ledger + 收尾
 
-ledger 顶层写入 `"mergeFinalHash": "<journal pushedHead>"`；经 `harness_gate.py close --phase merge` 关闭（禁止手工 phase.end）。
+ledger 顶层写入 `"mergeFinalHash": "<journal pushedHead>"`；经 `harness_gate.py close --phase merge` 关闭（禁止手工 phase.end）。正式层快照确认存在于 `.harness/cache/change-snapshots/<change-name>/`。
 
 ```markdown
 ## 合并完成 — <change-name>
@@ -399,6 +414,7 @@ ledger 顶层写入 `"mergeFinalHash": "<journal pushedHead>"`；经 `harness_ga
 - 主分支 push: ✅（<old>..<new> → origin/<主分支>）
 - mergeFinalHash: `<merge-final-hash>`
 - 验证策略: 🔁REUSED / 🔄已重跑
-- worktree: 已清理
+- snapshot: `.harness/cache/change-snapshots/<change-name>/`
+- worktree: 已清理（迁根后）
 - 下一步: `/harness-archive`
 ```

package/harness/bundles/java/claude-code/harness-sync/SKILL.md

@@ -19,7 +19,7 @@ disallowed-tools:
   - Bash(awk *)
   - Bash(curl *)
 ---
-<!-- generated by harness_deploy.py; core=6e87bc67bd51cbe9; overlay=java; agent=claude-code; do not edit -->
+<!-- generated by harness_deploy.py; core=f01595d3142d34b7; overlay=java; agent=claude-code; do not edit -->
 # harness-sync — 元数据同步
 
 ## Purpose

package/harness/bundles/java/claude-code/harness-test/SKILL.md

@@ -22,7 +22,7 @@ disallowed-tools:
   - Bash(node *)
   - Bash(codegraph *)
 ---
-<!-- generated by harness_deploy.py; core=6e87bc67bd51cbe9; overlay=java; agent=claude-code; do not edit -->
+<!-- generated by harness_deploy.py; core=f01595d3142d34b7; overlay=java; agent=claude-code; do not edit -->
 # harness-test — 测试执行
 
 ## Purpose

package/harness/bundles/java/claude-code/protocols/report-pipeline-protocol.md

@@ -43,6 +43,8 @@ python <skills-root>/scripts/harness_events.py append --change-dir ".harness/cha
 
 `append` 写入契约（Task 4 §6.1）：普通 append = 加锁 -> 追加一行 -> fsync -> 解锁，**不 load 历史、不渲染**（O(1)，跨进程锁 `events.ndjson.lock`，UUID 用完整 `uuid4().hex`）；仅 `--type phase.end` append 成功后渲染一次 `logs/execution-log.md`；显式 `harness_events.py render` 随时从完整 events 重建；`harness_archive.py finalize` 在 collect 前强制 render 一次。人类可读摘要（触发指令、降级原因、阶段结论）写入事件的 `note` 字段；阶段开始/结束、命令、验证、artifact、问题、决策各写对应 `type` 事件。旧 archive 缺少 events 可回放兼容；新变更不得以 execution-log 已存在为理由跳过 events。
 
+**artifact 硬门禁（H-8）**：`--type artifact` **必须**带非空 `--path`（change 相对路径）。知识探索/预览类说明改用 `--type issue` 或 `--type decision`，禁止再用 `--kind informational` 伪装无 path 的 artifact。
+
 基础事件结构（schema_version 3；兼容读取 v1/v2）：
 
 ```json