feat: 将项目从nano重命名为beaver并更新相关配置

- 将所有环境变量前缀从NANO_改为BEAVER_
- 更新README.md文档内容，包括项目介绍、组件说明和快速开始指南
- 修改.gitignore文件，添加auth-portal运行时路径排除规则
- 更新app-instance镜像标签从nano/app-instance改为beaver/app-instance
- 增强技能安全检查器，支持工具前缀白名单功能
- 添加技能草稿重新检查安全性API端点
- 扩展证据选择器，收集工具调用名称用于技能学习
- 改进技能合成器，基于实际调用的工具生成工具提示
- 优化路由超时处理机制，增加重试逻辑
- 更新后端架构文档，添加可视化入口和基础概念说明
- 实现在WebSocket消息中传递工具迭代次数信息

app-instance/README.md

@@ -45,14 +45,14 @@ runtime/registry/instances.json
 ### 1. 构建镜像
 
 ```bash
-docker build -t nano/app-instance:latest .
+docker build -t beaver/app-instance:latest .
 ```
 
 ### 2. 创建实例
 
 ```bash
 ./create-instance.sh \
-  --image nano/app-instance:latest \
+  --image beaver/app-instance:latest \
   --instance-id demo-001 \
   --auth-username admin \
   --auth-password 123456 \

app-instance/backend/beaver/engine/loader.py

@@ -260,7 +260,12 @@ class EngineLoader:
             review_service=review_service,
             publisher=skill_publisher,
             safety_checker=SkillDraftSafetyChecker(
-                allowed_tool_names={spec.name for spec in tool_registry.list_specs()}
+                allowed_tool_names={spec.name for spec in tool_registry.list_specs()},
+                allowed_tool_prefixes={
+                    f"mcp_{server_id}_"
+                    for server_id in self.config.tools.mcp_servers
+                    if str(server_id).strip()
+                },
             ),
             evaluator=SkillDraftEvaluator(run_memory_store),
         )

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

@@ -1437,6 +1437,15 @@ def create_app(
             raise HTTPException(status_code=404, detail="Safety report not found")
         return report.to_dict()
 
+    @app.post("/api/skills/{skill_name}/drafts/{draft_id}/safety")
+    async def recheck_skill_draft_safety(skill_name: str, draft_id: str, request: Request) -> dict[str, Any]:
+        loaded = get_agent_service(request).create_loop().boot()
+        try:
+            report = loaded.skill_learning_pipeline.check_safety(skill_name, draft_id)  # type: ignore[union-attr]
+        except ValueError as exc:
+            raise _skill_draft_http_error(exc) from exc
+        return report.to_dict()
+
     @app.get("/api/skills/{skill_name}/drafts/{draft_id}/eval")
     async def get_skill_draft_eval(skill_name: str, draft_id: str, request: Request) -> dict[str, Any]:
         loaded = get_agent_service(request).create_loop().boot()
@@ -1831,6 +1840,7 @@ def create_app(
                         "model": _clean_text(payload.get("model")) or None,
                         "provider_name": _clean_text(payload.get("provider_name")) or None,
                         "embedding_model": _clean_text(payload.get("embedding_model")) or None,
+                        "max_tool_iterations": _int_or_none(payload.get("max_tool_iterations")),
                     }
                     websocket_thinking_enabled = _bool_or_none(payload.get("thinking_enabled"))
                     if websocket_thinking_enabled is not None:
@@ -1844,6 +1854,7 @@ def create_app(
                         "content": f"Run failed before completion: {exc}",
                         "session_id": session_id,
                         "finish_reason": "error",
+                        "tool_iterations": 0,
                         "metadata": {
                             "error": str(exc),
                             "input_metadata": _websocket_input_metadata(payload),
@@ -2403,6 +2414,15 @@ def _bool_or_none(value: Any) -> bool | None:
     return None
 
 
+def _int_or_none(value: Any) -> int | None:
+    if value in (None, ""):
+        return None
+    try:
+        return int(value)
+    except (TypeError, ValueError):
+        return None
+
+
 def _websocket_message_payload(result: Any, *, input_payload: dict[str, Any]) -> dict[str, Any]:
     validation_result = getattr(result, "validation_result", None)
     task_id = getattr(result, "task_id", None)
@@ -2414,6 +2434,7 @@ def _websocket_message_payload(result: Any, *, input_payload: dict[str, Any]) ->
         "session_id": getattr(result, "session_id", None),
         "run_id": getattr(result, "run_id", None),
         "finish_reason": getattr(result, "finish_reason", None),
+        "tool_iterations": getattr(result, "tool_iterations", 0),
         "provider_name": getattr(result, "provider_name", None),
         "model": getattr(result, "model", None),
         "usage": dict(getattr(result, "usage", {}) or {}),

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

@@ -42,6 +42,8 @@ class EvidenceSelector:
         resolved_session_ids: list[str] = list(dict.fromkeys(session_ids or []))
         task_summaries: list[str] = []
         session_excerpts: list[str] = []
+        tool_names: list[str] = []
+        selected_tool_names: list[str] = []
         for run_id in run_ids:
             record = runs_by_id.get(run_id)
             if record is None:
@@ -56,12 +58,19 @@ class EvidenceSelector:
                 excerpt = self._session_excerpt(record.session_id, run_id)
                 if excerpt:
                     session_excerpts.append(excerpt)
+                run_tool_names, run_selected_tool_names = self._run_tool_names(record.session_id, run_id)
+                tool_names.extend(run_tool_names)
+                selected_tool_names.extend(run_selected_tool_names)
         return EvidencePacket(
             run_ids=resolved_run_ids,
             session_ids=resolved_session_ids,
             task_summaries=task_summaries[:8],
             session_excerpts=session_excerpts[:6],
-            metadata={"bounded": True},
+            metadata={
+                "bounded": True,
+                "tool_names": _unique_strings(tool_names),
+                "selected_tool_names": _unique_strings(selected_tool_names),
+            },
         )
 
     def _session_excerpt(self, session_id: str, run_id: str) -> str:
@@ -74,3 +83,37 @@ class EvidenceSelector:
                 continue
             visible.append(f"{event.role}: {event.content.strip()}")
         return "\n".join(visible[:12])[:2000]
+
+    def _run_tool_names(self, session_id: str, run_id: str) -> tuple[list[str], list[str]]:
+        if self.session_manager is None:
+            return [], []
+
+        names: list[str] = []
+        selected_names: list[str] = []
+        for event in self.session_manager.get_run_event_records(session_id, run_id):
+            if event.tool_name:
+                names.append(event.tool_name)
+            if event.tool_calls:
+                for call in event.tool_calls:
+                    if not isinstance(call, dict):
+                        continue
+                    name = call.get("name")
+                    function = call.get("function")
+                    if not name and isinstance(function, dict):
+                        name = function.get("name")
+                    if name:
+                        names.append(str(name))
+            if event.event_type == "tool_selection_snapshotted" and isinstance(event.event_payload, dict):
+                selected = event.event_payload.get("tool_names")
+                if isinstance(selected, list):
+                    selected_names.extend(str(item) for item in selected if str(item).strip())
+        return _unique_strings(names), _unique_strings(selected_names)
+
+
+def _unique_strings(values: list[str]) -> list[str]:
+    result: list[str] = []
+    for value in values:
+        cleaned = str(value).strip()
+        if cleaned and cleaned not in result:
+            result.append(cleaned)
+    return result

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

@@ -32,8 +32,14 @@ class SkillDraftSafetyChecker:
         "credentials",
     }
 
-    def __init__(self, *, allowed_tool_names: set[str] | None = None) -> None:
+    def __init__(
+        self,
+        *,
+        allowed_tool_names: set[str] | None = None,
+        allowed_tool_prefixes: set[str] | None = None,
+    ) -> None:
         self.allowed_tool_names = allowed_tool_names
+        self.allowed_tool_prefixes = allowed_tool_prefixes or set()
 
     def check(self, draft: SkillDraft) -> SkillDraftSafetyReport:
         issues: list[str] = []
@@ -50,7 +56,7 @@ class SkillDraftSafetyChecker:
 
         tool_hints = _tool_hints(frontmatter)
         if self.allowed_tool_names is not None:
-            unknown = [name for name in tool_hints if name not in self.allowed_tool_names]
+            unknown = [name for name in tool_hints if not self._is_allowed_tool_hint(name)]
             if unknown:
                 blocked.append(f"unknown tool hints: {', '.join(sorted(unknown))}")
         dangerous = sorted({name for name in tool_hints if name.lower() in self._DANGEROUS_TOOL_HINTS})
@@ -80,6 +86,11 @@ class SkillDraftSafetyChecker:
             created_at=_utc_now(),
         )
 
+    def _is_allowed_tool_hint(self, name: str) -> bool:
+        if self.allowed_tool_names is not None and name in self.allowed_tool_names:
+            return True
+        return any(name.startswith(prefix) and len(name) > len(prefix) for prefix in self.allowed_tool_prefixes)
+
 
 def _tool_hints(frontmatter: dict) -> list[str]:
     raw = frontmatter.get("tools")

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

@@ -65,19 +65,29 @@ class SkillDraftSynthesizer:
         )
         payload = self._parse_payload(response.content or "")
         if payload:
-            return payload
+            return self._normalize_payload(payload, evidence_packet)
         return self._fallback_payload(candidate, evidence_packet, action)
 
     @staticmethod
     def _build_prompt(candidate: SkillLearningCandidate, evidence_packet: EvidencePacket, action: str) -> str:
+        tool_names = _coerce_string_list(evidence_packet.metadata.get("tool_names"))
+        tool_section = ", ".join(tool_names) if tool_names else "none observed"
+        selected_tool_names = _coerce_string_list(evidence_packet.metadata.get("selected_tool_names"))
+        selected_tool_section = ", ".join(selected_tool_names) if selected_tool_names else "none recorded"
         return (
             f"Action: {action}\n"
             f"Candidate kind: {candidate.kind}\n"
             f"Reason: {candidate.reason}\n"
             f"Related skills: {candidate.related_skill_names}\n"
+            f"Called tool names: {tool_section}\n"
+            f"Run-selected tool names: {selected_tool_section}\n"
             f"Task summaries:\n- " + "\n- ".join(evidence_packet.task_summaries)
             + "\n\nSession excerpts:\n" + "\n\n".join(evidence_packet.session_excerpts)
-            + "\n\nReturn JSON only."
+            + "\n\nReturn JSON only. The frontmatter object must include:"
+            + "\n- description: a concise skill description"
+            + "\n- tools: an explicit JSON array of exact tool names this skill needs. "
+            + "Prefer called tool names when the workflow depends on them; use run-selected tool names only when clearly required. "
+            + "Use [] only when no tool is required."
         )
 
     @staticmethod
@@ -103,6 +113,19 @@ class SkillDraftSynthesizer:
             "change_reason": str(payload.get("change_reason") or ""),
         }
 
+    @staticmethod
+    def _normalize_payload(payload: dict[str, Any], evidence_packet: EvidencePacket) -> dict[str, Any]:
+        frontmatter = dict(payload.get("frontmatter") or {})
+        tool_hints = _coerce_string_list(frontmatter.get("tools"))
+        if not tool_hints:
+            tool_hints = _coerce_string_list(evidence_packet.metadata.get("tool_names"))
+        frontmatter["tools"] = tool_hints
+        return {
+            "frontmatter": frontmatter,
+            "content": str(payload.get("content") or "").strip(),
+            "change_reason": str(payload.get("change_reason") or ""),
+        }
+
     @staticmethod
     def _fallback_payload(candidate: SkillLearningCandidate, evidence_packet: EvidencePacket, action: str) -> dict[str, Any]:
         related = candidate.related_skill_names[0] if candidate.related_skill_names else "generated-skill"
@@ -111,8 +134,25 @@ class SkillDraftSynthesizer:
         return {
             "frontmatter": {
                 "description": candidate.reason or f"Auto-generated {action} draft for {title}.",
-                "tools": [],
+                "tools": _coerce_string_list(evidence_packet.metadata.get("tool_names")),
             },
             "content": f"# {title}\n\n## Evidence\n\n{content}\n",
             "change_reason": candidate.reason or f"Fallback {action} synthesis.",
         }
+
+
+def _coerce_string_list(value: Any) -> list[str]:
+    raw_items: list[Any]
+    if isinstance(value, list):
+        raw_items = value
+    elif isinstance(value, str):
+        raw_items = value.split(",")
+    else:
+        raw_items = []
+
+    result: list[str] = []
+    for item in raw_items:
+        cleaned = str(item).strip()
+        if cleaned and cleaned not in result:
+            result.append(cleaned)
+    return result

app-instance/backend/beaver/tasks/router.py

@@ -26,38 +26,42 @@ class MainAgentRouter:
     ) -> MainAgentDecision:
         if provider is None:
             return self._fallback(active_task=active_task, reason="router_provider_unavailable")
-        try:
-            chat_kwargs: dict[str, Any] = {
-                "messages": [
-                    {
-                        "role": "system",
-                        "content": (
-                            "You are Beaver's Intent Agent. Your only job is to route the user's "
-                            "message to simple chat or internal Task mode. Return only compact JSON. "
-                            "Do not answer the user. Do not explain."
-                        ),
-                    },
-                    {
-                        "role": "user",
-                        "content": self._prompt(
-                            message=message,
-                            active_task=active_task,
-                            recent_messages=recent_messages or [],
-                            intent_skill=intent_skill,
-                        ),
-                    },
-                ],
-                "tools": None,
-                "model": model,
-                "max_tokens": 256,
-                "temperature": 0.0,
-            }
-            if thinking_enabled is not None:
-                chat_kwargs["thinking_enabled"] = thinking_enabled
-            response = await asyncio.wait_for(provider.chat(**chat_kwargs), timeout=timeout_seconds)
-            return self.from_json(response.content or "", active_task=active_task)
-        except Exception as exc:
-            return self._fallback(active_task=active_task, reason=f"router_failed: {exc}")
+        chat_kwargs: dict[str, Any] = {
+            "messages": [
+                {
+                    "role": "system",
+                    "content": (
+                        "You are Beaver's Intent Agent. Your only job is to route the user's "
+                        "message to simple chat or internal Task mode. Return only compact JSON. "
+                        "Do not answer the user. Do not explain."
+                    ),
+                },
+                {
+                    "role": "user",
+                    "content": self._prompt(
+                        message=message,
+                        active_task=active_task,
+                        recent_messages=recent_messages or [],
+                        intent_skill=intent_skill,
+                    ),
+                },
+            ],
+            "tools": None,
+            "model": model,
+            "max_tokens": 256,
+            "temperature": 0.0,
+        }
+        if thinking_enabled is not None:
+            chat_kwargs["thinking_enabled"] = thinking_enabled
+
+        last_error: Exception | None = None
+        for attempt_timeout in (timeout_seconds, 12.0):
+            try:
+                response = await asyncio.wait_for(provider.chat(**chat_kwargs), timeout=attempt_timeout)
+                return self.from_json(response.content or "", active_task=active_task)
+            except Exception as exc:
+                last_error = exc
+        return self._fallback(active_task=active_task, reason=f"router_failed: {last_error}")
 
     def from_json(self, text: str, *, active_task: TaskRecord | None = None) -> MainAgentDecision:
         payload = self._parse_json_object(text)

app-instance/backend/docs/architecture/backend-overview.md

@@ -2,10 +2,46 @@
 
 这是新 `Beaver` 后端的架构入口文档。
 
+可视化入口：
+
+- [Beaver Backend 可视化](backend-visualization.html)
+
+## 给零基础读者的版本
+
+可以先把这个后端理解成一个“帮用户完成任务的后台工厂”：
+
+1. 用户在前端发一句话。
+2. 后端入口把这句话接住。
+3. 服务层判断它是闲聊，还是一个需要执行和跟踪的任务。
+4. 如果是任务，系统会创建或继续一个 `Task`。
+5. 运行内核 `AgentLoop` 准备上下文、选择技能、选择工具、调用模型。
+6. 如果模型需要查文件、写文件、搜索或调用外部系统，就通过 `tools` 执行。
+7. 执行结果会写回会话、任务记录和运行记录，后续可以继续追踪、验证和学习。
+
+这套结构里最重要的原则是：**所有 agent 共用同一个运行内核 `engine`**。也就是说，主 agent 和被拆出去的小 agent 不是两套系统，它们最终都会回到 `AgentLoop`，使用同一套上下文、工具、技能和记录方式。
+
+## 先认识几个词
+
+- `interfaces`：入口层。负责接收 Web、CLI、Gateway、MCP 等不同来源的请求。
+- `services`：应用服务层。负责把入口请求转成系统内部要做的事情。
+- `engine`：运行内核。真正组织 prompt、调用模型、执行 tool loop 的地方。
+- `coordinator`：多 agent 编排层。负责把复杂任务拆成 sequence、parallel 或 DAG。
+- `skills`：技能层。可以理解成给 agent 的专项说明书。
+- `tools`：工具层。可以理解成 agent 能按需调用的动作，例如读文件、写文件、搜索、执行命令。
+- `memory`：记忆层。保存会话、任务结果、运行记录、反馈和技能学习数据。
+- `permissions`：权限与治理层。负责约束哪些能力能用、怎么用。
+
+## 一句话请求的流转
+
+典型路径是：
+
+`interfaces` -> `AgentService` -> `MainAgentRouter` -> `TaskService` / `TaskExecutionPlanner` -> `AgentLoop` -> `skills` / `tools` / `memory` -> 返回用户。
+
+如果任务很简单，可能只走单 agent。如果任务更复杂，`TaskExecutionPlanner` 可能先生成一个 team plan，让 `coordinator` 安排多个 sub-agent 分别处理，最后再由主 agent 综合输出。
+
 当前约束：
 
 1. 所有 agent 共用 `engine`。
 2. 多 agent 编排进入 `coordinator`。
 3. skills、memory、permissions 独立成能力层。
 4. `interfaces` 只做薄入口。
-

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

@@ -158,7 +158,7 @@ def test_load_config_reads_mcp_authz_identity(tmp_path) -> None:
                 },
                 "authz": {
                     "enabled": True,
-                    "baseUrl": "http://nano-authz-service:19090",
+                    "baseUrl": "http://beaver-authz-service:19090",
                 },
                 "backend_identity": {
                     "backend_id": "stevenli",
@@ -180,7 +180,7 @@ def test_load_config_reads_mcp_authz_identity(tmp_path) -> None:
     assert server.sensitive is True
 
     assert config.authz.enabled is True
-    assert config.authz.base_url == "http://nano-authz-service:19090"
+    assert config.authz.base_url == "http://beaver-authz-service:19090"
     assert config.backend_identity.backend_id == "stevenli"
     assert config.backend_identity.client_id == "stevenli"
 

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

@@ -38,6 +38,39 @@ class RouterProvider(LLMProvider):
         return "stub-model"
 
 
+class SequenceRouterProvider(LLMProvider):
+    def __init__(self, responses: list[str | Exception]) -> None:
+        super().__init__()
+        self.responses = list(responses)
+        self.calls: list[dict] = []
+
+    async def chat(
+        self,
+        messages: list[dict],
+        tools: list[dict] | None = None,
+        model: str | None = None,
+        max_tokens: int = 4096,
+        temperature: float = 0.7,
+        thinking_enabled: bool | None = None,
+    ) -> LLMResponse:
+        self.calls.append(
+            {
+                "messages": messages,
+                "max_tokens": max_tokens,
+                "temperature": temperature,
+                "model": model,
+                "thinking_enabled": thinking_enabled,
+            }
+        )
+        response = self.responses.pop(0)
+        if isinstance(response, Exception):
+            raise response
+        return LLMResponse(content=response, finish_reason="stop", provider_name="stub", model="stub-model")
+
+    def get_default_model(self) -> str:
+        return "stub-model"
+
+
 def _task() -> TaskRecord:
     return TaskRecord(
         task_id="task-1",
@@ -133,3 +166,38 @@ def test_router_fallback_keeps_active_task_but_not_new_task() -> None:
 
     assert active.is_task
     assert not inactive.is_task
+
+
+def test_router_retries_once_after_provider_failure() -> None:
+    provider = SequenceRouterProvider(
+        [
+            TimeoutError(),
+            '{"action":"new_task","reason":"needs search","short_title":"中美会面"}',
+        ]
+    )
+
+    decision = asyncio.run(
+        MainAgentRouter().classify(
+            "帮我看看昨天的中美会面都谈了什么？",
+            provider=provider,
+        )
+    )
+
+    assert decision.is_task
+    assert decision.action == "create_task"
+    assert len(provider.calls) == 2
+
+
+def test_router_fallback_after_two_provider_failures() -> None:
+    provider = SequenceRouterProvider([TimeoutError(), RuntimeError("provider down")])
+
+    decision = asyncio.run(
+        MainAgentRouter().classify(
+            "帮我看看昨天的中美会面都谈了什么？",
+            provider=provider,
+        )
+    )
+
+    assert not decision.is_task
+    assert decision.reason == "router_failed: provider down"
+    assert len(provider.calls) == 2

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

@@ -15,7 +15,12 @@ from beaver.skills.reviews import ReviewService
 from beaver.skills.specs import SkillSpecStore
 
 
-def _pipeline(tmp_path: Path, *, allowed_tools: set[str] | None = None) -> SkillLearningPipelineService:
+def _pipeline(
+    tmp_path: Path,
+    *,
+    allowed_tools: set[str] | None = None,
+    allowed_prefixes: set[str] | None = None,
+) -> SkillLearningPipelineService:
     spec_store = SkillSpecStore(tmp_path)
     run_store = RunMemoryStore(tmp_path / "memory" / "runs")
     learning_store = SkillLearningStore(tmp_path / "memory" / "skills")
@@ -32,7 +37,10 @@ def _pipeline(tmp_path: Path, *, allowed_tools: set[str] | None = None) -> Skill
         draft_service=drafts,
         review_service=ReviewService(spec_store),
         publisher=SkillPublisher(spec_store),
-        safety_checker=SkillDraftSafetyChecker(allowed_tool_names=allowed_tools),
+        safety_checker=SkillDraftSafetyChecker(
+            allowed_tool_names=allowed_tools,
+            allowed_tool_prefixes=allowed_prefixes,
+        ),
     )
 
 
@@ -106,3 +114,53 @@ def test_safety_blocks_unknown_tool_hint(tmp_path: Path) -> None:
 
     assert report.passed is False
     assert "unknown tool hints" in report.blocked_reasons[0]
+
+
+def test_safety_allows_configured_mcp_tool_prefix(tmp_path: Path) -> None:
+    pipeline = _pipeline(
+        tmp_path,
+        allowed_tools={"echo"},
+        allowed_prefixes={"mcp_officebench_"},
+    )
+    draft = pipeline.draft_service.create_new_skill_draft(
+        skill_name="officebench-excel",
+        proposed_content="# OfficeBench Excel\n\nUse the configured OfficeBench MCP tools.",
+        proposed_frontmatter={
+            "description": "officebench",
+            "tools": [
+                "mcp_officebench_shell_list_directory",
+                "mcp_officebench_excel_read_file",
+                "mcp_officebench_excel_set_cell",
+            ],
+        },
+        created_by="test",
+        reason="test",
+    )
+
+    report = pipeline.check_safety(draft.skill_name, draft.draft_id)
+
+    assert report.passed is True
+    assert report.blocked_reasons == []
+
+
+def test_safety_blocks_unconfigured_mcp_tool_prefix(tmp_path: Path) -> None:
+    pipeline = _pipeline(
+        tmp_path,
+        allowed_tools={"echo"},
+        allowed_prefixes={"mcp_outlook_mcp_"},
+    )
+    draft = pipeline.draft_service.create_new_skill_draft(
+        skill_name="wrong-mcp",
+        proposed_content="# Wrong MCP\n\nUse an unconfigured MCP namespace.",
+        proposed_frontmatter={
+            "description": "wrong mcp",
+            "tools": ["mcp_officebench_excel_set_cell"],
+        },
+        created_by="test",
+        reason="test",
+    )
+
+    report = pipeline.check_safety(draft.skill_name, draft.draft_id)
+
+    assert report.passed is False
+    assert "mcp_officebench_excel_set_cell" in report.blocked_reasons[0]

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

@@ -7,6 +7,7 @@ from types import SimpleNamespace
 
 from beaver.engine.providers.base import LLMProvider, LLMResponse
 from beaver.engine.providers.factory import ProviderBundle
+from beaver.engine.session import SessionManager
 from beaver.memory.runs import RunMemoryStore, RunRecord
 from beaver.memory.skills import SkillLearningCandidate, SkillLearningStore
 from beaver.skills.drafts import DraftService
@@ -125,6 +126,78 @@ def test_worker_retries_and_marks_failed_after_limit(tmp_path: Path) -> None:
     assert "provider failed" in (candidate.last_error or "")
 
 
+def test_synthesizer_fills_missing_tools_from_evidence(tmp_path: Path) -> None:
+    pipeline = _pipeline(tmp_path)
+    candidate = pipeline.get_candidate("candidate-1")
+    provider = JsonProvider(
+        payload={
+            "frontmatter": {"description": "Generated skill"},
+            "content": "# Generated\n\nUse the observed workflow.",
+            "change_reason": "learned",
+        }
+    )
+    packet = EvidenceSelector(pipeline.learning_service.run_store).build_evidence_packet(
+        candidate.source_run_ids,
+        candidate.source_session_ids,
+    )
+    packet.metadata["tool_names"] = ["web_fetch", "memory"]
+
+    payload = asyncio.run(
+        SkillDraftSynthesizer().synthesize_new_skill(candidate, packet, provider, "stub")
+    )
+
+    assert payload["frontmatter"]["tools"] == ["web_fetch", "memory"]
+
+
+def test_evidence_selector_records_run_tool_names(tmp_path: Path) -> None:
+    run_store = RunMemoryStore(tmp_path / "memory" / "runs")
+    run_store.append_run_record(
+        RunRecord(
+            run_id="run-1",
+            session_id="session-1",
+            task_text="research latest docs",
+            started_at="start",
+            ended_at="end",
+            success=True,
+            finish_reason="stop",
+        )
+    )
+    session_manager = SessionManager(tmp_path)
+    session_manager.ensure_session("session-1")
+    session_manager.append_message(
+        "session-1",
+        run_id="run-1",
+        role="system",
+        event_type="tool_selection_snapshotted",
+        event_payload={"tool_names": ["memory", "web_fetch"]},
+        context_visible=False,
+    )
+    session_manager.append_message(
+        "session-1",
+        run_id="run-1",
+        role="assistant",
+        tool_calls=[{"id": "call-1", "function": {"name": "web_search"}}],
+    )
+    session_manager.append_message(
+        "session-1",
+        run_id="run-1",
+        role="tool",
+        tool_name="web_fetch",
+        content="ok",
+    )
+
+    try:
+        packet = EvidenceSelector(run_store, session_manager).build_evidence_packet(
+            ["run-1"],
+            ["session-1"],
+        )
+    finally:
+        session_manager.close()
+
+    assert packet.metadata["tool_names"] == ["web_search", "web_fetch"]
+    assert packet.metadata["selected_tool_names"] == ["memory", "web_fetch"]
+
+
 def test_worker_supersedes_candidate_when_active_draft_exists(tmp_path: Path) -> None:
     pipeline = _pipeline(tmp_path)
     pipeline.learning_store.record_learning_candidate(

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

@@ -78,6 +78,7 @@ def test_websocket_message_returns_chat_metadata_and_session_updated() -> None:
             "model": None,
             "provider_name": None,
             "embedding_model": None,
+            "max_tool_iterations": None,
         }
     ]
     assert message["type"] == "message"
@@ -128,5 +129,6 @@ def test_websocket_runtime_error_returns_assistant_error_message() -> None:
     assert message["role"] == "assistant"
     assert message["session_id"] == "web:alpha"
     assert message["finish_reason"] == "error"
+    assert message["tool_iterations"] == 0
     assert "boom" in message["content"]
     assert pong == {"type": "pong"}

app-instance/create-instance.sh

@@ -4,7 +4,7 @@ set -euo pipefail
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 REGISTRY_TOOL="${SCRIPT_DIR}/instance-registry.py"
 
-IMAGE_NAME="${IMAGE_NAME:-nano/app-instance:latest}"
+IMAGE_NAME="${IMAGE_NAME:-beaver/app-instance:latest}"
 INSTANCES_ROOT_DEFAULT="${SCRIPT_DIR}/runtime/instances"
 REGISTRY_PATH_DEFAULT="${SCRIPT_DIR}/runtime/registry/instances.json"
 KNOWN_PROVIDERS=" custom anthropic openai openrouter deepseek groq zhipu dashscope vllm gemini moonshot minimax aihubmix siliconflow volcengine "
@@ -25,6 +25,7 @@ MODEL="openai/gpt-5"
 PROVIDER="openai"
 API_KEY="${API_KEY:-}"
 API_BASE="${API_BASE:-}"
+SKIP_PROVIDER_CONFIG=0
 AUTH_USERNAME=""
 AUTH_PASSWORD=""
 USERNAME=""
@@ -42,22 +43,23 @@ REPLACE=0
 usage() {
   cat <<'EOF'
 Usage:
-  ./create-instance.sh --instance-id demo --auth-username admin --auth-password 123456 --api-key sk-xxx [options]
+  ./create-instance.sh --instance-id demo --auth-username admin --auth-password 123456 [options]
 
 Required:
   --instance-id <id>          Unique instance id.
   --auth-username <name>      Initial web login username.
   --auth-password <password>  Initial web login password.
-  --api-key <key>             Provider API key for Boardware Genius.
 
 Optional:
-  --image <name>              Docker image tag. Default: nano/app-instance:latest
+  --image <name>              Docker image tag. Default: beaver/app-instance:latest
   --container-name <name>     Docker container name. Default: app-instance-<slug>
   --host-port <port>          Host port to publish. Default: auto-pick from 20000-29999.
   --public-url <url>          Public URL exposed to users. Default: http://127.0.0.1:<host-port>
   --provider <name>           Provider key in config.json. Default: openai
   --api-base <url>            Optional custom provider base URL.
+  --api-key <key>             Provider API key for Boardware Genius.
   --model <name>              Model name. Default: openai/gpt-5
+  --skip-provider-config      Create the instance without model/provider/API key settings.
   --authz-base-url <url>      AuthZ service base URL.
   --authz-outlook-mcp-url <url>
                               Managed Outlook MCP URL for AuthZ mode.
@@ -134,6 +136,7 @@ render_config_json() {
   PROVIDER="$PROVIDER" \
   API_KEY="$API_KEY" \
   API_BASE="$API_BASE" \
+  SKIP_PROVIDER_CONFIG="$SKIP_PROVIDER_CONFIG" \
   AUTHZ_BASE_URL="$AUTHZ_BASE_URL" \
   AUTHZ_OUTLOOK_MCP_URL="$AUTHZ_OUTLOOK_MCP_URL" \
   OUTLOOK_MCP_SERVER_ID="$OUTLOOK_MCP_SERVER_ID" \
@@ -151,11 +154,20 @@ target = Path(os.environ["TARGET_PATH"])
 provider = os.environ["PROVIDER"]
 outlook_mcp_url = os.environ["AUTHZ_OUTLOOK_MCP_URL"].strip()
 outlook_server_id = os.environ["OUTLOOK_MCP_SERVER_ID"].strip() or "outlook_mcp"
+skip_provider_config = os.environ["SKIP_PROVIDER_CONFIG"].strip() == "1"
 
-provider_cfg = {"apiKey": os.environ["API_KEY"]}
-api_base = os.environ["API_BASE"].strip()
-if api_base:
-    provider_cfg["apiBase"] = api_base
+providers = {}
+agent_defaults = {
+    "workspace": "/root/.beaver/workspace",
+}
+if not skip_provider_config:
+    provider_cfg = {"apiKey": os.environ["API_KEY"]}
+    api_base = os.environ["API_BASE"].strip()
+    if api_base:
+        provider_cfg["apiBase"] = api_base
+    providers[provider] = provider_cfg
+    agent_defaults["provider"] = provider
+    agent_defaults["model"] = os.environ["MODEL"]
 
 outlook_tool_names = [
     "auth_status",
@@ -193,14 +205,9 @@ if outlook_mcp_url:
 
 data = {
     "agents": {
-        "defaults": {
-            "workspace": "/root/.beaver/workspace",
-            "model": os.environ["MODEL"],
-        }
-    },
-    "providers": {
-        provider: provider_cfg,
+        "defaults": agent_defaults
     },
+    "providers": providers,
     "tools": {
         "restrictToWorkspace": True,
         "mcpServers": default_mcp_servers,
@@ -345,6 +352,10 @@ while [[ $# -gt 0 ]]; do
       MODEL="${2:-}"
       shift 2
       ;;
+    --skip-provider-config)
+      SKIP_PROVIDER_CONFIG=1
+      shift
+      ;;
     --auth-username)
       AUTH_USERNAME="${2:-}"
       shift 2
@@ -438,7 +449,9 @@ done
 [[ -n "$INSTANCE_ID" ]] || die "--instance-id is required"
 [[ -n "$AUTH_USERNAME" ]] || die "--auth-username is required"
 [[ -n "$AUTH_PASSWORD" ]] || die "--auth-password is required"
-[[ -n "$API_KEY" ]] || die "--api-key is required"
+if [[ "$SKIP_PROVIDER_CONFIG" -ne 1 ]]; then
+  [[ -n "$API_KEY" ]] || die "--api-key is required unless --skip-provider-config is set"
+fi
 
 INSTANCE_SLUG="$(slugify "$INSTANCE_ID")"
 USERNAME="${USERNAME:-$AUTH_USERNAME}"
@@ -469,10 +482,12 @@ if [[ -z "$INSTANCE_HOST" ]]; then
   INSTANCE_HOST="$(extract_url_host "$PUBLIC_URL")"
 fi
 
-case "$KNOWN_PROVIDERS" in
-  *" ${PROVIDER} "*) ;;
-  *) die "unsupported provider '${PROVIDER}'" ;;
-esac
+if [[ "$SKIP_PROVIDER_CONFIG" -ne 1 ]]; then
+  case "$KNOWN_PROVIDERS" in
+    *" ${PROVIDER} "*) ;;
+    *) die "unsupported provider '${PROVIDER}'" ;;
+  esac
+fi
 
 if [[ -n "$BACKEND_ID$CLIENT_ID$CLIENT_SECRET" ]]; then
   [[ -n "$BACKEND_ID" && -n "$CLIENT_ID" && -n "$CLIENT_SECRET" ]] || die "backend identity requires --backend-id, --client-id and --client-secret together"
@@ -550,9 +565,9 @@ RUN_ARGS=(
   -e "APP_FRONTEND_PORT=3000"
   -e "APP_BACKEND_PORT=18080"
   -e "BEAVER_OUTLOOK_MCP_SERVER_ID=${OUTLOOK_MCP_SERVER_ID}"
-  --label "nano.instance.id=${INSTANCE_ID}"
-  --label "nano.instance.slug=${INSTANCE_SLUG}"
-  --label "nano.instance.public_url=${PUBLIC_URL}"
+  --label "beaver.instance.id=${INSTANCE_ID}"
+  --label "beaver.instance.slug=${INSTANCE_SLUG}"
+  --label "beaver.instance.public_url=${PUBLIC_URL}"
 )
 
 if [[ -n "$NETWORK_NAME" ]]; then

app-instance/frontend/app/(app)/skills/page.tsx

@@ -41,6 +41,7 @@ import {
   listSkillDrafts,
   listSkills,
   publishSkillDraft,
+  recheckSkillDraftSafety,
   regenerateSkillDraft,
   rejectSkillDraft,
   rollbackPublishedSkill,
@@ -412,6 +413,11 @@ export default function SkillsPage() {
                           rejectSkillDraft(draft.skill_name, draft.draft_id)
                         )
                       }
+                      onRecheckSafety={() =>
+                        runAction(`safety:${draft.draft_id}`, () =>
+                          recheckSkillDraftSafety(draft.skill_name, draft.draft_id)
+                        )
+                      }
                       onPublish={(confirmHighRisk) =>
                         runAction(`publish:${draft.draft_id}`, () =>
                           publishSkillDraft(draft.skill_name, draft.draft_id, '', confirmHighRisk)
@@ -697,6 +703,7 @@ function DraftCard({
   onSubmit,
   onApprove,
   onReject,
+  onRecheckSafety,
   onPublish,
 }: {
   draft: SkillDraft;
@@ -704,6 +711,7 @@ function DraftCard({
   onSubmit: () => Promise<unknown>;
   onApprove: () => Promise<unknown>;
   onReject: () => Promise<unknown>;
+  onRecheckSafety: () => Promise<unknown>;
   onPublish: (confirmHighRisk: boolean) => Promise<unknown>;
 }) {
   const { locale } = useAppI18n();
@@ -814,6 +822,10 @@ function DraftCard({
             <XCircle className="mr-2 h-4 w-4" />
             {t('拒绝', 'Reject')}
           </Button>
+          <Button variant="outline" size="sm" disabled={busy || TERMINAL_DRAFT_STATUSES.has(draft.status)} onClick={() => void onRecheckSafety()}>
+            <ShieldCheck className="mr-2 h-4 w-4" />
+            {t('复检', 'Recheck')}
+          </Button>
           <Button size="sm" disabled={busy || publishBlocked} onClick={handlePublish}>
             <Rocket className="mr-2 h-4 w-4" />
             {t('发布', 'Publish')}

app-instance/frontend/app/(app)/tasks/[taskId]/page.tsx

@@ -3,7 +3,7 @@
 import Link from 'next/link';
 import { useParams, useRouter } from 'next/navigation';
 import React, { useMemo, useState } from 'react';
-import { AlertCircle, ArrowLeft, Bot, CheckCircle2, Download, FileText, HelpCircle, MessageSquare, RefreshCw, Trash2, User, XCircle } from 'lucide-react';
+import { AlertCircle, ArrowLeft, Bot, CheckCircle2, Download, FileText, HelpCircle, Loader2, MessageSquare, RefreshCw, ThumbsUp, Trash2, User, XCircle } from 'lucide-react';
 
 import { TaskRuntimeStatusBadge, formatTaskRuntimeDuration, formatTaskRuntimeTime, progressPercent } from '@/components/task-runtime/TaskRuntimeShared';
 import { Badge } from '@/components/ui/badge';
@@ -17,6 +17,14 @@ import { buildTaskRuntimeView, type TaskRuntimeNodeView } from '@/lib/task-runti
 import { useChatStore } from '@/lib/store';
 import type { BackendTask, BackendTaskRun, ProcessArtifact, ProcessEvent } from '@/types';
 
+type TaskFeedbackType = 'satisfied' | 'revise' | 'abandon';
+type TaskFeedbackItem = {
+  feedback_type?: unknown;
+  comment?: unknown;
+  created_at?: unknown;
+  run_id?: unknown;
+};
+
 function taskVisibleStatus(task: TaskRuntimeNodeView, locale: 'zh-CN' | 'en-US') {
   if (task.status === 'error') return pickAppText(locale, '任务失败', 'Task failed');
   if (task.status === 'cancelled') return pickAppText(locale, '已取消', 'Cancelled');
@@ -53,11 +61,13 @@ export default function TaskDetailPage() {
   const [backendTaskLoading, setBackendTaskLoading] = useState(false);
   const [selectedRunId, setSelectedRunId] = useState<string | null>(task?.rootRunId ?? null);
   const [revision, setRevision] = useState('');
+  const [runtimeFeedback, setRuntimeFeedback] = useState<TaskFeedbackItem | null>(null);
   const [actionError, setActionError] = useState<string | null>(null);
   const [actionBusy, setActionBusy] = useState<string | null>(null);
 
   React.useEffect(() => {
     setSelectedRunId(task?.rootRunId ?? null);
+    setRuntimeFeedback(null);
   }, [task?.rootRunId]);
 
   React.useEffect(() => {
@@ -138,6 +148,8 @@ export default function TaskDetailPage() {
     });
   };
 
+  const backendFeedbackRunId = backendTask ? pickFeedbackRunId(backendTask) : null;
+
   if (!task && backendTask) {
     const validation = backendTask.validation_result;
     const accepted = Boolean(validation?.accepted);
@@ -185,6 +197,26 @@ export default function TaskDetailPage() {
           </CardContent>
         </Card>
 
+        <TaskFeedbackPanel
+          sessionId={backendTask.session_id}
+          runId={backendFeedbackRunId}
+          taskStatus={backendTask.status}
+          feedbackItems={feedbackItems}
+          actionBusy={actionBusy}
+          onSubmit={(feedbackType, comment) =>
+            runAction(`backend-feedback-${feedbackType}`, async () => {
+              await submitChatFeedback({
+                sessionId: backendTask.session_id,
+                runId: backendFeedbackRunId!,
+                feedbackType,
+                comment,
+              });
+              const refreshed = await getBackendTask(backendTask.task_id);
+              setBackendTask(refreshed);
+            })
+          }
+        />
+
         <Card>
           <CardHeader>
             <CardTitle className="text-base">{pickAppText(locale, 'Agent 执行过程', 'Agent conversation process')}</CardTitle>
@@ -424,37 +456,33 @@ export default function TaskDetailPage() {
             </CardContent>
           </Card>
 
-          <Card>
-            <CardHeader>
-              <CardTitle className="text-base">{pickAppText(locale, '修订意见', 'Revision')}</CardTitle>
-            </CardHeader>
-            <CardContent className="space-y-3">
-              <Textarea
-                value={revision}
-                onChange={(event) => setRevision(event.target.value)}
-                placeholder={pickAppText(locale, '直接写下需要调整的地方...', 'Describe what should change...')}
-              />
-              <Button
-                className="w-full"
-                disabled={!revision.trim() || Boolean(actionBusy)}
-                onClick={() =>
-                  void runAction('revision', async () => {
-                    updateMessageFeedback(task.rootRunId, 'revise');
-                    await submitChatFeedback({
-                      sessionId: task.sessionId || 'web:default',
-                      runId: task.rootRunId,
-                      feedbackType: 'revise',
-                      comment: revision.trim(),
-                    });
-                    setRevision('');
-                  })
-                }
-              >
-                <RefreshCw className="mr-2 h-4 w-4" />
-                {pickAppText(locale, '提交修订', 'Submit revision')}
-              </Button>
-            </CardContent>
-          </Card>
+          <TaskFeedbackPanel
+            sessionId={task.sessionId || 'web:default'}
+            runId={task.rootRunId}
+            taskStatus={task.status}
+            feedbackItems={runtimeFeedback ? [runtimeFeedback] : []}
+            actionBusy={actionBusy}
+            revision={revision}
+            onRevisionChange={setRevision}
+            onSubmit={(feedbackType, comment) =>
+              runAction(`runtime-feedback-${feedbackType}`, async () => {
+                updateMessageFeedback(task.rootRunId, feedbackType);
+                await submitChatFeedback({
+                  sessionId: task.sessionId || 'web:default',
+                  runId: task.rootRunId,
+                  feedbackType,
+                  comment,
+                });
+                setRuntimeFeedback({
+                  feedback_type: feedbackType,
+                  comment: comment || '',
+                  created_at: new Date().toISOString(),
+                  run_id: task.rootRunId,
+                });
+                setRevision('');
+              })
+            }
+          />
 
           <Card>
             <CardHeader>
@@ -521,6 +549,136 @@ function Metric({ label, value }: { label: string; value: string }) {
   );
 }
 
+function TaskFeedbackPanel({
+  sessionId,
+  runId,
+  taskStatus,
+  feedbackItems,
+  actionBusy,
+  revision,
+  onRevisionChange,
+  onSubmit,
+}: {
+  sessionId: string;
+  runId: string | null;
+  taskStatus: string;
+  feedbackItems: TaskFeedbackItem[];
+  actionBusy: string | null;
+  revision?: string;
+  onRevisionChange?: (value: string) => void;
+  onSubmit: (feedbackType: TaskFeedbackType, comment?: string) => Promise<unknown>;
+}) {
+  const { locale } = useAppI18n();
+  const [localComment, setLocalComment] = React.useState('');
+  const comment = revision ?? localComment;
+  const setComment = onRevisionChange ?? setLocalComment;
+  const isFinalized = taskStatus === 'closed' || taskStatus === 'abandoned';
+  const recordedFeedback = feedbackForRun(feedbackItems, runId) ?? (isFinalized ? latestFeedback(feedbackItems) : null);
+  const canSubmit = Boolean(runId) && !recordedFeedback && !isFinalized && !actionBusy;
+
+  const submit = (feedbackType: TaskFeedbackType, nextComment?: string) => {
+    if (!runId || !canSubmit) return;
+    void onSubmit(feedbackType, nextComment);
+  };
+
+  return (
+    <Card>
+      <CardHeader>
+        <CardTitle className="text-base">{pickAppText(locale, '任务反馈', 'Task feedback')}</CardTitle>
+      </CardHeader>
+      <CardContent className="space-y-4">
+        {recordedFeedback ? (
+          <div className="rounded-md border border-border bg-muted/25 p-3 text-sm">
+            <div className="flex items-center gap-2 font-medium">
+              <CheckCircle2 className="h-4 w-4 text-[#657162]" />
+              {pickAppText(locale, '已提交反馈', 'Feedback submitted')}: {humanFeedback(String(recordedFeedback.feedback_type || ''), locale)}
+            </div>
+            {recordedFeedback.comment ? (
+              <p className="mt-2 text-muted-foreground">{String(recordedFeedback.comment)}</p>
+            ) : null}
+            {recordedFeedback.created_at ? (
+              <p className="mt-2 text-xs text-muted-foreground">{formatTaskRuntimeTime(String(recordedFeedback.created_at), locale)}</p>
+            ) : null}
+          </div>
+        ) : isFinalized ? (
+          <div className="rounded-md border border-border bg-muted/25 p-3 text-sm text-muted-foreground">
+            {pickAppText(locale, '任务已结束，不能再提交新的反馈。', 'This task is finalized and cannot accept new feedback.')}
+          </div>
+        ) : !runId ? (
+          <div className="rounded-md border border-border bg-muted/25 p-3 text-sm text-muted-foreground">
+            {pickAppText(locale, '暂无可反馈的运行记录。', 'No run is available for feedback yet.')}
+          </div>
+        ) : null}
+
+        <div className="grid gap-2 sm:grid-cols-3">
+          <FeedbackButton
+            type="satisfied"
+            icon={<ThumbsUp className="mr-2 h-4 w-4" />}
+            label={pickAppText(locale, '满意', 'Satisfied')}
+            actionBusy={actionBusy}
+            disabled={!canSubmit}
+            onClick={() => submit('satisfied', comment.trim() || undefined)}
+          />
+          <FeedbackButton
+            type="revise"
+            icon={<RefreshCw className="mr-2 h-4 w-4" />}
+            label={pickAppText(locale, '需要修改', 'Needs revision')}
+            actionBusy={actionBusy}
+            disabled={!canSubmit || !comment.trim()}
+            onClick={() => submit('revise', comment.trim())}
+          />
+          <FeedbackButton
+            type="abandon"
+            icon={<XCircle className="mr-2 h-4 w-4" />}
+            label={pickAppText(locale, '放弃', 'Abandon')}
+            actionBusy={actionBusy}
+            disabled={!canSubmit}
+            onClick={() => submit('abandon', comment.trim() || undefined)}
+          />
+        </div>
+
+        <Textarea
+          value={comment}
+          onChange={(event) => setComment(event.target.value)}
+          disabled={Boolean(recordedFeedback) || isFinalized || Boolean(actionBusy)}
+          placeholder={pickAppText(locale, '需要修改时写下具体要求；满意或放弃可选填说明。', 'Describe requested changes; notes are optional for satisfied or abandon.')}
+        />
+        <div className="text-xs text-muted-foreground">
+          {pickAppText(locale, '反馈将记录到当前任务运行：', 'Feedback will be recorded on run: ')}
+          <span className="font-mono">{runId || '-'}</span>
+          <span className="mx-1">·</span>
+          {pickAppText(locale, '会话：', 'Session: ')}
+          <span className="font-mono">{sessionId}</span>
+        </div>
+      </CardContent>
+    </Card>
+  );
+}
+
+function FeedbackButton({
+  type,
+  icon,
+  label,
+  actionBusy,
+  disabled,
+  onClick,
+}: {
+  type: TaskFeedbackType;
+  icon: React.ReactNode;
+  label: string;
+  actionBusy: string | null;
+  disabled: boolean;
+  onClick: () => void;
+}) {
+  const isBusy = Boolean(actionBusy?.endsWith(type));
+  return (
+    <Button type="button" variant="outline" className="w-full justify-center" disabled={disabled || Boolean(actionBusy)} onClick={onClick}>
+      {isBusy ? <Loader2 className="mr-2 h-4 w-4 animate-spin" /> : icon}
+      {label}
+    </Button>
+  );
+}
+
 function BackendRunConversation({ run, index }: { run: BackendTaskRun; index: number }) {
   const { locale } = useAppI18n();
   return (
@@ -597,6 +755,24 @@ function humanFinishReason(reason: string, locale: 'zh-CN' | 'en-US') {
   return reason;
 }
 
+function pickFeedbackRunId(task: BackendTask): string | null {
+  const runIds = task.run_ids.filter(Boolean);
+  if (runIds.length > 0) return runIds[runIds.length - 1];
+  const runs = task.runs ?? [];
+  if (runs.length > 0) return runs[runs.length - 1].run_id;
+  return null;
+}
+
+function feedbackForRun(items: TaskFeedbackItem[], runId: string | null): TaskFeedbackItem | null {
+  if (!runId) return null;
+  const ordered = [...items].reverse();
+  return ordered.find((item) => String(item.run_id || '') === runId) ?? null;
+}
+
+function latestFeedback(items: TaskFeedbackItem[]): TaskFeedbackItem | null {
+  return [...items].reverse()[0] ?? null;
+}
+
 function arrayOfStrings(value: unknown): string[] {
   return Array.isArray(value) ? value.map((item) => String(item)).filter(Boolean) : [];
 }

app-instance/frontend/lib/api.ts

@@ -777,6 +777,13 @@ export async function getSkillDraftSafety(skillName: string, draftId: string): P
   return fetchJSON(`/api/skills/${encodeURIComponent(skillName)}/drafts/${encodeURIComponent(draftId)}/safety`);
 }
 
+export async function recheckSkillDraftSafety(skillName: string, draftId: string): Promise<SkillDraftSafetyReport> {
+  return fetchJSON(`/api/skills/${encodeURIComponent(skillName)}/drafts/${encodeURIComponent(draftId)}/safety`, {
+    method: 'POST',
+    body: JSON.stringify({}),
+  });
+}
+
 export async function getSkillDraftEval(skillName: string, draftId: string): Promise<SkillDraftEvalReport> {
   return fetchJSON(`/api/skills/${encodeURIComponent(skillName)}/drafts/${encodeURIComponent(draftId)}/eval`);
 }