```

feat(learning): 添加技能学习候选者合成锁定机制

添加了 DraftSynthesisInProgress 和 DraftHasNoChanges 异常来处理并发场景，
确保同一技能学习候选者的合成过程不会重复执行。实现了 claim_learning_candidate_for_synthesis
方法来原子性地锁定候选者进行合成。

fix(web): 为技能草案创建端点添加适当的HTTP状态码

当草案没有变化或正在合成时，现在正确返回409状态码而不是内部错误。

feat(skills): 实现技能修订内容比较以检测无变化情况

添加了 _is_noop_revision 方法来比较基础技能和提议的修订，
如果内容没有实际变化则抛出 NoDraftChanges 异常。

refactor(process): 修复任务证据记录后根运行状态更新逻辑

将任务证据记录事件后的状态从 waiting 更改为 done，并设置 finished_at 时间戳。

feat(tools): 防止在同一运行中重复执行外部写入操作

为邮件发送、日历创建等外部写入工具添加去重机制，避免重复的外部操作。

test: 添加技能学习和工具执行的单元测试

增加测试用例验证并发草案合成、重复外部写入抑制和无变化修订检测等功能。
```

app-instance/backend/beaver/interfaces/web/app.py

@@ -52,7 +52,13 @@ from beaver.services.user_file_resolver import (
 )
 from beaver.skills.authoring import canonical_skill_format_instructions, ensure_canonical_skill_body, normalize_skill_frontmatter
 from beaver.skills.authoring.format import parse_skill_rewrite_json
-from beaver.skills.learning import SkillLearningService, SkillLearningWorker, SkillLearningWorkerConfig
+from beaver.skills.learning import (
+    DraftHasNoChanges,
+    DraftSynthesisInProgress,
+    SkillLearningService,
+    SkillLearningWorker,
+    SkillLearningWorkerConfig,
+)
 from beaver.skills.learning.replay import ReplayRunner
 from beaver.skills.catalog.utils import extract_required_tool_names, parse_frontmatter
 
@@ -2236,6 +2242,10 @@ def create_app(
                 candidate_id,
                 provider_bundle=provider_bundle,
             )
+        except DraftHasNoChanges as exc:
+            raise HTTPException(status_code=409, detail=str(exc)) from exc
+        except DraftSynthesisInProgress as exc:
+            raise HTTPException(status_code=409, detail=str(exc)) from exc
         except ValueError as exc:
             raise HTTPException(status_code=404, detail=str(exc)) from exc
         return _skill_draft_payload(loaded, draft.skill_name, draft.draft_id)
@@ -2251,6 +2261,10 @@ def create_app(
                 candidate_id,
                 provider_bundle=provider_bundle,
             )
+        except DraftHasNoChanges as exc:
+            raise HTTPException(status_code=409, detail=str(exc)) from exc
+        except DraftSynthesisInProgress as exc:
+            raise HTTPException(status_code=409, detail=str(exc)) from exc
         except ValueError as exc:
             raise HTTPException(status_code=404, detail=str(exc)) from exc
         return _skill_draft_payload(loaded, draft.skill_name, draft.draft_id)

app-instance/backend/beaver/memory/skills/store.py

@@ -114,6 +114,52 @@ class SkillLearningStore:
             )
         return updated
 
+    def claim_learning_candidate_for_synthesis(
+        self,
+        candidate_id: str,
+        *,
+        force: bool = False,
+    ) -> SkillLearningCandidate | None:
+        """Atomically claim a candidate before the expensive draft synthesis step."""
+
+        with self._locked():
+            candidates = self.list_learning_candidates()
+            claimed: SkillLearningCandidate | None = None
+            for index, candidate in enumerate(candidates):
+                if candidate.candidate_id != candidate_id:
+                    continue
+                if candidate.status in {"queued", "synthesizing"}:
+                    return None
+                if not force and candidate.draft_skill_name and candidate.draft_id:
+                    return None
+                payload = candidate.to_dict()
+                payload.update(
+                    {
+                        "status": "synthesizing",
+                        "last_error": None,
+                        "updated_at": _utc_now(),
+                    }
+                )
+                claimed = SkillLearningCandidate.from_dict(payload)
+                candidates[index] = claimed
+                break
+            if claimed is None:
+                return None
+            self.candidates_path.parent.mkdir(parents=True, exist_ok=True)
+            self.candidates_path.write_text(
+                "".join(
+                    json.dumps(candidate.to_dict(), ensure_ascii=False, sort_keys=True) + "\n"
+                    for candidate in candidates
+                ),
+                encoding="utf-8",
+            )
+            self.append_audit_event(
+                candidate_id,
+                "draft_synthesis_started",
+                {"status": "synthesizing", "force": force},
+            )
+            return claimed
+
     def list_learning_candidates(self, status: str | None = None) -> list[SkillLearningCandidate]:
         results: list[SkillLearningCandidate] = []
         for payload in self._read_jsonl(self.candidates_path):

app-instance/backend/beaver/services/process_service.py

@@ -351,8 +351,8 @@ class SessionProcessProjector:
                     )
 
             elif record.event_type == "task_evidence_recorded":
-                root["status"] = "waiting"
-                root["finished_at"] = None
+                root["status"] = "done"
+                root["finished_at"] = created_at
                 add_event(
                     event_id=_event_id(record, "evidence"),
                     run_id=record.run_id or root_run_id,

app-instance/backend/beaver/skills/learning/__init__.py

@@ -9,7 +9,7 @@ from .missing_skill import (
     MissingSkillDraftResult,
     MissingSkillSynthesizer,
 )
-from .pipeline import SkillLearningPipelineService
+from .pipeline import DraftHasNoChanges, DraftSynthesisInProgress, SkillLearningPipelineService
 from .preservation import check_preservation
 from .replay import ReplayArmRequest, ReplayRunner, ReplayToolExecutor, ReplayToolPolicy, classify_tool_mode
 from .service import RunReceiptContext, SkillLearningService
@@ -27,6 +27,8 @@ __all__ = [
     "MissingSkillDraftResult",
     "MissingSkillSynthesizer",
     "RunReceiptContext",
+    "DraftHasNoChanges",
+    "DraftSynthesisInProgress",
     "SkillLearningPipelineService",
     "check_preservation",
     "ReplayToolExecutor",

app-instance/backend/beaver/skills/learning/pipeline.py

@@ -9,7 +9,7 @@ from beaver.memory.skills import SkillDraftEvalReport, SkillDraftSafetyReport, S
 from beaver.skills.drafts import DraftService
 from beaver.skills.learning.eval import SkillDraftEvaluator
 from beaver.skills.learning.replay import ReplayRunner
-from beaver.skills.learning.service import SkillLearningService
+from beaver.skills.learning.service import NoDraftChanges, SkillLearningService
 from beaver.skills.learning.safety import SkillDraftSafetyChecker
 from beaver.skills.publisher import SkillPublisher
 from beaver.skills.reviews import ReviewService
@@ -22,6 +22,14 @@ _REJECTABLE_DRAFT_STATUSES = {
 }
 
 
+class DraftSynthesisInProgress(RuntimeError):
+    """Raised when another request already claimed the candidate for synthesis."""
+
+
+class DraftHasNoChanges(RuntimeError):
+    """Raised when synthesis produced no effective changes from the base skill."""
+
+
 class SkillLearningPipelineService:
     """Coordinates candidate -> draft -> review -> publish lifecycle."""
 
@@ -60,8 +68,23 @@ class SkillLearningPipelineService:
         candidate_id: str,
         *,
         provider_bundle: ProviderBundle,
+        force: bool = False,
     ) -> SkillDraft:
-        draft = await self.learning_service.synthesize_draft(candidate_id, provider_bundle)
+        if not force:
+            existing = self._draft_for_candidate(candidate_id)
+            if existing is not None:
+                return existing
+        claimed = self.learning_store.claim_learning_candidate_for_synthesis(candidate_id, force=force)
+        if claimed is None:
+            existing = self._draft_for_candidate(candidate_id)
+            if existing is not None:
+                return existing
+            raise DraftSynthesisInProgress(f"Draft synthesis is already in progress for candidate: {candidate_id}")
+        try:
+            draft = await self.learning_service.synthesize_draft(candidate_id, provider_bundle)
+        except NoDraftChanges as exc:
+            self.mark_candidate_superseded(candidate_id, str(exc))
+            raise DraftHasNoChanges(str(exc)) from exc
         self.mark_draft_synthesized(candidate_id, draft)
         return draft
 
@@ -71,13 +94,7 @@ class SkillLearningPipelineService:
         *,
         provider_bundle: ProviderBundle,
     ) -> SkillDraft:
-        self.learning_store.transition_learning_candidate(
-            candidate_id,
-            "synthesizing",
-            event_type="draft_synthesis_started",
-            last_error=None,
-        )
-        return await self.synthesize_draft(candidate_id, provider_bundle=provider_bundle)
+        return await self.synthesize_draft(candidate_id, provider_bundle=provider_bundle, force=True)
 
     def mark_candidate_queued(self, candidate_id: str) -> SkillLearningCandidate:
         return self._require_updated(
@@ -162,6 +179,12 @@ class SkillLearningPipelineService:
             raise ValueError(f"Draft not found: {skill_name}/{draft_id}")
         return draft
 
+    def _draft_for_candidate(self, candidate_id: str) -> SkillDraft | None:
+        candidate = self.get_candidate(candidate_id)
+        if not candidate.draft_skill_name or not candidate.draft_id:
+            return None
+        return self.draft_service.get_draft(candidate.draft_skill_name, candidate.draft_id)
+
     def submit_review(
         self,
         skill_name: str,

app-instance/backend/beaver/skills/learning/service.py

@@ -20,8 +20,9 @@ from beaver.plugins.tree_merge import merge_supporting_file_trees
 from beaver.skills.drafts.service import DraftService
 from beaver.skills.learning.evidence import EvidencePacket, EvidenceSelector
 from beaver.skills.learning.synthesizer import SkillDraftSynthesizer
-from beaver.skills.catalog.utils import parse_frontmatter
+from beaver.skills.catalog.utils import parse_frontmatter, strip_frontmatter
 from beaver.skills.specs import SkillActivationReceipt
+from beaver.skills.specs.serialization import normalize_frontmatter
 
 
 @dataclass(slots=True)
@@ -30,6 +31,10 @@ class RunReceiptContext:
     effect_records: list[SkillEffectRecord] = field(default_factory=list)
 
 
+class NoDraftChanges(ValueError):
+    """Raised when synthesis produces the same effective skill content as the base version."""
+
+
 class SkillLearningService:
     def __init__(
         self,
@@ -231,13 +236,18 @@ class SkillLearningService:
             )
         target_skill = candidate.related_skill_names[0]
         base_version = candidate.evidence.get("skill_version")
+        base_skill = self._base_skill_snapshot(target_skill, base_version)
         payload = await self.synthesizer.synthesize_revision(
             candidate,
             packet,
             provider,
             model,
-            base_skill=self._base_skill_snapshot(target_skill, base_version),
+            base_skill=base_skill,
         )
+        if self._is_noop_revision(payload, base_skill):
+            raise NoDraftChanges(
+                f"Synthesis produced no changes for {target_skill}/{base_version or 'current'}"
+            )
         return self.draft_service.create_revision_draft(
             skill_name=target_skill,
             base_version=base_version,
@@ -340,6 +350,16 @@ class SkillLearningService:
             "tool_hints": list(loaded.version.tool_hints),
         }
 
+    @staticmethod
+    def _is_noop_revision(payload: dict[str, Any], base_skill: dict[str, Any] | None) -> bool:
+        if base_skill is None:
+            return False
+        base_frontmatter = normalize_frontmatter(dict(base_skill.get("frontmatter") or {}))
+        proposed_frontmatter = normalize_frontmatter(dict(payload.get("frontmatter") or {}))
+        base_body = _normalize_skill_body(str(base_skill.get("content") or ""))
+        proposed_body = _normalize_skill_body(str(payload.get("content") or ""))
+        return base_frontmatter == proposed_frontmatter and base_body == proposed_body
+
     def _merged_base_skill_snapshot(self, skill_names: list[str]) -> dict[str, Any] | None:
         snapshots = [
             snapshot
@@ -602,6 +622,10 @@ class SkillLearningService:
         return parsed.astimezone(timezone.utc)
 
 
+def _normalize_skill_body(content: str) -> str:
+    return "\n".join(line.rstrip() for line in strip_frontmatter(content).strip().splitlines()).strip()
+
+
 def _digest_map(root: Path) -> dict[str, dict[str, Any]]:
     digest = hash_plugin_skill_tree(root)
     return {

app-instance/backend/beaver/skills/learning/worker.py

@@ -9,7 +9,7 @@ from typing import Callable
 
 from beaver.engine.providers import ProviderBundle
 from beaver.memory.skills import SkillLearningCandidate
-from beaver.skills.learning.pipeline import SkillLearningPipelineService
+from beaver.skills.learning.pipeline import DraftHasNoChanges, SkillLearningPipelineService
 from beaver.skills.learning.replay import ReplayRunner
 
 
@@ -114,13 +114,13 @@ class SkillLearningWorker:
         if self._has_active_draft(candidate):
             self.pipeline.mark_candidate_superseded(candidate.candidate_id, "active draft already exists for this skill")
             return False
-        self.pipeline.mark_candidate_queued(candidate.candidate_id)
-        self.pipeline.mark_candidate_synthesizing(candidate.candidate_id)
-        draft = await self.pipeline.synthesize_draft(
-            candidate.candidate_id,
-            provider_bundle=self.provider_bundle_factory(),
-        )
-        self.pipeline.mark_draft_synthesized(candidate.candidate_id, draft)
+        try:
+            draft = await self.pipeline.synthesize_draft(
+                candidate.candidate_id,
+                provider_bundle=self.provider_bundle_factory(),
+            )
+        except DraftHasNoChanges:
+            return False
         safety = self.pipeline.check_safety(draft.skill_name, draft.draft_id)
         if not safety.passed or safety.risk_level == "critical":
             return True

app-instance/backend/beaver/tools/runtime/executor.py

@@ -11,6 +11,7 @@
 
 from __future__ import annotations
 
+import hashlib
 import json
 from typing import TYPE_CHECKING, Any
 
@@ -44,7 +45,45 @@ class ToolExecutor:
                 tool_name=tool_name,
                 error="tool_not_found",
             )
-        return await tool.invoke(arguments or {}, context or ToolContext())
+        normalized_arguments = dict(arguments or {})
+        tool_context = context or ToolContext()
+        write_key = _external_write_key(tool_name, normalized_arguments)
+        if write_key is None:
+            return await tool.invoke(normalized_arguments, tool_context)
+
+        external_writes = _external_write_state(tool_context)
+        previous = external_writes.get(write_key)
+        if previous is not None:
+            previous_content = str(previous.get("content") or "").strip()
+            detail = f" Previous result: {previous_content}" if previous_content else ""
+            return ToolResult(
+                success=True,
+                content=(
+                    f"Duplicate external write suppressed for {tool_name}. "
+                    "A matching write was already attempted in this run."
+                    f"{detail}"
+                ),
+                tool_name=tool_name,
+                error="duplicate_external_write_suppressed",
+                raw_output={"duplicate": True, "previous": previous},
+            )
+
+        external_writes[write_key] = {
+            "tool_name": tool_name,
+            "arguments": normalized_arguments,
+            "status": "attempted",
+            "content": "",
+            "error": None,
+        }
+        result = await tool.invoke(normalized_arguments, tool_context)
+        external_writes[write_key] = {
+            "tool_name": tool_name,
+            "arguments": normalized_arguments,
+            "status": "done" if result.success else "error",
+            "content": result.content,
+            "error": result.error,
+        }
+        return result
 
     async def execute_tool_call(
         self,
@@ -115,3 +154,42 @@ class ToolExecutor:
         if tool_call.get("name"):
             return str(tool_call["name"])
         return "unknown"
+
+
+_EXTERNAL_WRITE_TOOL_TERMS = (
+    "mail_send_email",
+    "mail_reply_to_message",
+    "mail_forward_message",
+    "mail_move_message",
+    "calendar_create_event",
+    "calendar_update_event",
+)
+
+
+def _external_write_state(context: ToolContext) -> dict[str, dict[str, Any]]:
+    state = context.metadata.setdefault("external_write_attempts", {})
+    if not isinstance(state, dict):
+        state = {}
+        context.metadata["external_write_attempts"] = state
+    return state
+
+
+def _external_write_key(tool_name: str, arguments: dict[str, Any]) -> str | None:
+    lowered = tool_name.lower()
+    if not any(term in lowered for term in _EXTERNAL_WRITE_TOOL_TERMS):
+        return None
+    payload = json.dumps(_normalize_for_key(arguments), ensure_ascii=False, sort_keys=True, separators=(",", ":"))
+    digest = hashlib.sha256(payload.encode("utf-8")).hexdigest()
+    return f"{lowered}:{digest}"
+
+
+def _normalize_for_key(value: Any) -> Any:
+    if isinstance(value, dict):
+        return {str(key): _normalize_for_key(value[key]) for key in sorted(value, key=str)}
+    if isinstance(value, list):
+        return [_normalize_for_key(item) for item in value]
+    if isinstance(value, tuple):
+        return [_normalize_for_key(item) for item in value]
+    if isinstance(value, (str, int, float, bool)) or value is None:
+        return value
+    return str(value)

app-instance/backend/tests/unit/test_create_instance_script.py

@@ -0,0 +1,69 @@
+import json
+import os
+import subprocess
+from pathlib import Path
+
+
+def test_create_instance_writes_default_max_tool_iterations(tmp_path) -> None:
+    app_instance_dir = Path(__file__).resolve().parents[3]
+    fake_bin = tmp_path / "bin"
+    fake_bin.mkdir()
+    docker = fake_bin / "docker"
+    docker.write_text(
+        """#!/usr/bin/env bash
+set -euo pipefail
+case "${1:-}" in
+  image)
+    [[ "${2:-}" == "inspect" ]]
+    exit 0
+    ;;
+  container)
+    [[ "${2:-}" == "inspect" ]]
+    exit 1
+    ;;
+  run)
+    exit 0
+    ;;
+  *)
+    echo "unexpected docker command: $*" >&2
+    exit 1
+    ;;
+esac
+""",
+        encoding="utf-8",
+    )
+    docker.chmod(0o755)
+
+    env = os.environ.copy()
+    env["PATH"] = f"{fake_bin}:{env['PATH']}"
+    instances_root = tmp_path / "instances"
+    result = subprocess.run(
+        [
+            str(app_instance_dir / "create-instance.sh"),
+            "--instance-id",
+            "default-tools",
+            "--auth-username",
+            "steven",
+            "--auth-password",
+            "secret",
+            "--skip-provider-config",
+            "--host-port",
+            "29001",
+            "--instances-root",
+            str(instances_root),
+            "--registry",
+            str(tmp_path / "registry.json"),
+            "--skip-initial-skills",
+        ],
+        cwd=app_instance_dir,
+        env=env,
+        text=True,
+        capture_output=True,
+        check=False,
+    )
+
+    assert result.returncode == 0, result.stderr
+    config_path = instances_root / "default-tools" / "beaver-home" / "config.json"
+    config = json.loads(config_path.read_text(encoding="utf-8"))
+
+    assert config["agents"]["defaults"]["maxToolIterations"] == 100

app-instance/backend/tests/unit/test_process_projection.py

@@ -363,6 +363,52 @@ def test_process_projection_emits_tool_cards_from_run_messages(tmp_path: Path) -
     assert tool_result["metadata"]["success"] is True
 
 
+def test_process_projection_marks_root_done_when_result_is_ready(tmp_path: Path) -> None:
+    session = SessionManager(tmp_path)
+    run_store = RunMemoryStore(tmp_path / "memory" / "runs")
+    run_store.append_run_record(
+        RunRecord(
+            run_id="main-run",
+            session_id="web:test",
+            task_id="task-1",
+            attempt_index=1,
+            task_text="send email",
+            started_at="2026-01-01T00:00:03+00:00",
+            ended_at="2026-01-01T00:00:04+00:00",
+            success=True,
+            finish_reason="stop",
+        )
+    )
+    session.append_message(
+        "web:test",
+        role="system",
+        event_type="task_execution_planned",
+        event_payload={"task_id": "task-1", "attempt_index": 1, "plan_mode": "single", "strategy": "single"},
+        context_visible=False,
+    )
+    session.append_message(
+        "web:test",
+        role="system",
+        event_type="task_synthesis_completed",
+        event_payload={"task_id": "task-1", "attempt_index": 1, "main_run_id": "main-run"},
+        context_visible=False,
+    )
+    session.append_message(
+        "web:test",
+        run_id="main-run",
+        role="system",
+        event_type="task_evidence_recorded",
+        event_payload={"task_id": "task-1", "attempt_index": 1, "evidence_status": "recorded"},
+        context_visible=False,
+    )
+
+    projection = SessionProcessProjector(session, run_store).project("web:test")
+
+    root_run = next(run for run in projection["runs"] if run["run_id"] == "task:task-1:attempt:1")
+    assert root_run["status"] == "done"
+    assert root_run["finished_at"] is not None
+
+
 def test_process_projection_exposes_ephemeral_guidance_artifacts(tmp_path: Path) -> None:
     session = SessionManager(tmp_path)
     run_store = RunMemoryStore(tmp_path / "memory" / "runs")

app-instance/backend/tests/unit/test_skill_learning_worker.py

@@ -5,6 +5,8 @@ import json
 from pathlib import Path
 from types import SimpleNamespace
 
+import pytest
+
 from beaver.engine.providers.base import LLMProvider, LLMResponse
 from beaver.engine.providers.factory import ProviderBundle
 from beaver.engine.session import SessionManager
@@ -13,6 +15,8 @@ from beaver.memory.skills import SkillLearningCandidate, SkillLearningStore
 from beaver.skills.authoring.format import is_canonical_skill_body
 from beaver.skills.drafts import DraftService
 from beaver.skills.learning import (
+    DraftHasNoChanges,
+    DraftSynthesisInProgress,
     EvidenceSelector,
     SkillDraftSynthesizer,
     SkillLearningPipelineService,
@@ -22,7 +26,7 @@ from beaver.skills.learning import (
 )
 from beaver.skills.publisher import SkillPublisher
 from beaver.skills.reviews import ReviewService
-from beaver.skills.specs import SkillSpecStore
+from beaver.skills.specs import SkillSpecStore, SkillVersion
 
 
 class JsonProvider(LLMProvider):
@@ -44,6 +48,20 @@ class JsonProvider(LLMProvider):
         return "stub"
 
 
+class BlockingJsonProvider(JsonProvider):
+    def __init__(self, *, started: asyncio.Event, release: asyncio.Event) -> None:
+        super().__init__()
+        self.started = started
+        self.release = release
+        self.calls = 0
+
+    async def chat(self, messages: list[dict], tools: list[dict] | None = None, model: str | None = None, max_tokens: int = 4096, temperature: float = 0.7) -> LLMResponse:
+        self.calls += 1
+        self.started.set()
+        await self.release.wait()
+        return await super().chat(messages, tools=tools, model=model, max_tokens=max_tokens, temperature=temperature)
+
+
 def _bundle(provider: LLMProvider) -> ProviderBundle:
     runtime = SimpleNamespace(model="stub", provider_name="stub")
     return ProviderBundle(main_runtime=runtime, main_provider=provider)  # type: ignore[arg-type]
@@ -120,6 +138,69 @@ def _pipeline(tmp_path: Path) -> SkillLearningPipelineService:
     )
 
 
+def _revision_pipeline(tmp_path: Path, content: str, frontmatter: dict) -> SkillLearningPipelineService:
+    spec_store = SkillSpecStore(tmp_path)
+    spec_store.write_skill_version(
+        SkillVersion(
+            skill_name="web-operation",
+            version="v0001",
+            content_hash="hash-v1",
+            summary_hash="summary-v1",
+            created_at="2026-06-01T00:00:00+00:00",
+            created_by="test",
+            change_reason="initial",
+            parent_version=None,
+            review_state="published",
+            frontmatter=frontmatter,
+            summary="web operation",
+            tool_hints=list(frontmatter.get("tools") or []),
+        ),
+        content,
+    )
+    spec_store.set_current_version("web-operation", "v0001")
+    run_store = RunMemoryStore(tmp_path / "memory" / "runs")
+    learning_store = SkillLearningStore(tmp_path / "memory" / "skills")
+    run_store.append_run_record(
+        RunRecord(
+            run_id="run-1",
+            session_id="session-1",
+            task_text="check detailed weather",
+            started_at="start",
+            ended_at="end",
+            success=True,
+            finish_reason="stop",
+        )
+    )
+    learning_store.record_learning_candidate(
+        SkillLearningCandidate(
+            candidate_id="candidate-revision",
+            kind="revise_skill",
+            source_run_ids=["run-1"],
+            source_session_ids=["session-1"],
+            related_skill_names=["web-operation"],
+            reason="revise web guidance",
+            evidence={"skill_version": "v0001"},
+            priority=10,
+            confidence=0.9,
+        )
+    )
+    draft_service = DraftService(spec_store)
+    learning_service = SkillLearningService(
+        run_store=run_store,
+        learning_store=learning_store,
+        draft_service=draft_service,
+        evidence_selector=EvidenceSelector(run_store),
+        synthesizer=SkillDraftSynthesizer(),
+    )
+    return SkillLearningPipelineService(
+        learning_store=learning_store,
+        learning_service=learning_service,
+        draft_service=draft_service,
+        review_service=ReviewService(spec_store),
+        publisher=SkillPublisher(spec_store),
+    )
+
+
 def test_worker_synthesizes_open_candidate_without_publish(tmp_path: Path) -> None:
     pipeline = _pipeline(tmp_path)
     worker = SkillLearningWorker(
@@ -137,6 +218,104 @@ def test_worker_synthesizes_open_candidate_without_publish(tmp_path: Path) -> No
     assert pipeline.list_drafts(candidate.draft_skill_name)[0].status == "draft"
 
 
+def test_concurrent_draft_synthesis_is_claimed_once(tmp_path: Path) -> None:
+    pipeline = _pipeline(tmp_path)
+
+    async def scenario():
+        started = asyncio.Event()
+        release = asyncio.Event()
+        provider = BlockingJsonProvider(started=started, release=release)
+        first = asyncio.create_task(
+            pipeline.synthesize_draft("candidate-1", provider_bundle=_bundle(provider))
+        )
+        await asyncio.wait_for(started.wait(), timeout=1)
+        with pytest.raises(DraftSynthesisInProgress):
+            await pipeline.synthesize_draft("candidate-1", provider_bundle=_bundle(JsonProvider()))
+        release.set()
+        return await first, provider
+
+    draft, provider = asyncio.run(scenario())
+    candidate = pipeline.get_candidate("candidate-1")
+
+    assert provider.calls == 1
+    assert candidate.status == "draft_ready"
+    assert candidate.draft_id == draft.draft_id
+    assert len(pipeline.list_drafts(candidate.draft_skill_name)) == 1
+
+
+def test_existing_draft_synthesis_request_returns_same_draft(tmp_path: Path) -> None:
+    pipeline = _pipeline(tmp_path)
+    first = asyncio.run(pipeline.synthesize_draft("candidate-1", provider_bundle=_bundle(JsonProvider())))
+    second = asyncio.run(pipeline.synthesize_draft("candidate-1", provider_bundle=_bundle(JsonProvider(fail=True))))
+
+    assert second.draft_id == first.draft_id
+    assert len(pipeline.list_drafts(first.skill_name)) == 1
+
+
+def test_revision_synthesis_with_no_content_changes_supersedes_candidate(tmp_path: Path) -> None:
+    content = (
+        "---\n"
+        "name: web-operation\n"
+        "description: Web search and fetch.\n"
+        "tools:\n"
+        "  - web_fetch\n"
+        "  - web_search\n"
+        "---\n"
+        "\n"
+        "# Web Operation\n"
+        "\n"
+        "## Overview\n"
+        "\n"
+        "Web search and fetch.\n"
+        "\n"
+        "## When to Use\n"
+        "\n"
+        "- Use when web information is required.\n"
+        "\n"
+        "## Required Tools\n"
+        "\n"
+        "- `web_fetch`\n"
+        "- `web_search`\n"
+        "\n"
+        "## Workflow\n"
+        "\n"
+        "- Use web_search, then web_fetch.\n"
+        "\n"
+        "## Validation\n"
+        "\n"
+        "- Verify sources.\n"
+        "\n"
+        "## Boundaries\n"
+        "\n"
+        "- Stay within the request.\n"
+        "\n"
+        "## Anti-Patterns\n"
+        "\n"
+        "- Do not cite unsupported claims.\n"
+    )
+    frontmatter = {
+        "name": "web-operation",
+        "description": "Web search and fetch.",
+        "tools": ["web_fetch", "web_search"],
+    }
+    pipeline = _revision_pipeline(tmp_path, content, frontmatter)
+    provider = JsonProvider(
+        payload={
+            "frontmatter": frontmatter,
+            "content": content,
+            "change_reason": "No changes are required.",
+        }
+    )
+
+    with pytest.raises(DraftHasNoChanges):
+        asyncio.run(pipeline.synthesize_draft("candidate-revision", provider_bundle=_bundle(provider)))
+    candidate = pipeline.get_candidate("candidate-revision")
+
+    assert candidate.status == "superseded"
+    assert "no changes" in (candidate.last_error or "").lower()
+    assert pipeline.list_drafts("web-operation") == []
+
+
 def test_worker_evaluates_draft_with_replay_runner_when_available(tmp_path: Path) -> None:
     pipeline = _pipeline(tmp_path)
     replay_runner = FakeReplayRunner()

app-instance/backend/tests/unit/test_tool_assembler.py

@@ -28,12 +28,14 @@ class DummyTool(BaseTool):
             toolset=toolset,
             always_available=always_available,
         )
+        self.calls: list[dict] = []
 
     @property
     def spec(self) -> ToolSpec:
         return self._spec
 
     async def invoke(self, arguments: dict, context: ToolContext) -> ToolResult:
+        self.calls.append(dict(arguments))
         return ToolResult(success=True, content="ok", tool_name=self.spec.name)
 
 
@@ -198,3 +200,30 @@ def test_tool_executor_parses_object_tool_call_string_arguments() -> None:
 
     assert name == "echo"
     assert arguments == {"text": "hello"}
+
+
+def test_tool_executor_suppresses_duplicate_external_write_in_same_run() -> None:
+    registry = ToolRegistry()
+    send_tool = DummyTool("mcp_outlook_mcp_mail_send_email", toolset="mcp")
+    registry.register(send_tool)
+    executor = ToolExecutor(registry)
+    context = ToolContext(
+        metadata={
+            "task_id": "task-1",
+            "run_id": "run-1",
+        }
+    )
+    arguments = {
+        "to_recipients": ["jay.chen@boardware.com"],
+        "subject": "请回复今天下午的日程安排",
+        "body": "Hi Jay",
+    }
+
+    first = asyncio.run(executor.execute("mcp_outlook_mcp_mail_send_email", arguments, context=context))
+    second = asyncio.run(executor.execute("mcp_outlook_mcp_mail_send_email", dict(arguments), context=context))
+
+    assert first.success is True
+    assert second.success is True
+    assert second.error == "duplicate_external_write_suppressed"
+    assert "Duplicate external write suppressed" in second.content
+    assert len(send_tool.calls) == 1

app-instance/create-instance.sh

@@ -187,6 +187,7 @@ skip_provider_config = os.environ["SKIP_PROVIDER_CONFIG"].strip() == "1"
 providers = {}
 agent_defaults = {
     "workspace": "/root/.beaver/workspace",
+    "maxToolIterations": 100,
 }
 if not skip_provider_config:
     provider_cfg = {"apiKey": os.environ["API_KEY"]}