重构主控编排与运行时预热链路，统一工作区提示词/专家调度协议并补齐 wiki 记忆注入与写回闭环。

同时收敛启动与运维脚本默认行为（含 wiki worker）、更新 Admin 可观测性与相关测试，降低首轮时延并提高运行稳定性。

Made-with: Cursor

openclaw/docs/tools/llm-task.md

@@ -0,0 +1,119 @@
+---
+summary: "JSON-only LLM tasks for workflows (optional plugin tool)"
+read_when:
+  - You want a JSON-only LLM step inside workflows
+  - You need schema-validated LLM output for automation
+title: "LLM Task"
+---
+
+# LLM Task
+
+`llm-task` is an **optional plugin tool** that runs a JSON-only LLM task and
+returns structured output (optionally validated against JSON Schema).
+
+This is ideal for workflow engines like Lobster: you can add a single LLM step
+without writing custom OpenClaw code for each workflow.
+
+## Enable the plugin
+
+1. Enable the plugin:
+
+```json
+{
+  "plugins": {
+    "entries": {
+      "llm-task": { "enabled": true }
+    }
+  }
+}
+```
+
+2. Allowlist the tool (it is registered with `optional: true`):
+
+```json
+{
+  "agents": {
+    "list": [
+      {
+        "id": "main",
+        "tools": { "allow": ["llm-task"] }
+      }
+    ]
+  }
+}
+```
+
+## Config (optional)
+
+```json
+{
+  "plugins": {
+    "entries": {
+      "llm-task": {
+        "enabled": true,
+        "config": {
+          "defaultProvider": "openai-codex",
+          "defaultModel": "gpt-5.4",
+          "defaultAuthProfileId": "main",
+          "allowedModels": ["openai-codex/gpt-5.4"],
+          "maxTokens": 800,
+          "timeoutMs": 30000
+        }
+      }
+    }
+  }
+}
+```
+
+`allowedModels` is an allowlist of `provider/model` strings. If set, any request
+outside the list is rejected.
+
+## Tool parameters
+
+- `prompt` (string, required)
+- `input` (any, optional)
+- `schema` (object, optional JSON Schema)
+- `provider` (string, optional)
+- `model` (string, optional)
+- `thinking` (string, optional)
+- `authProfileId` (string, optional)
+- `temperature` (number, optional)
+- `maxTokens` (number, optional)
+- `timeoutMs` (number, optional)
+
+`thinking` accepts the standard OpenClaw reasoning presets, such as `low` or `medium`.
+
+## Output
+
+Returns `details.json` containing the parsed JSON (and validates against
+`schema` when provided).
+
+## Example: Lobster workflow step
+
+```lobster
+openclaw.invoke --tool llm-task --action json --args-json '{
+  "prompt": "Given the input email, return intent and draft.",
+  "thinking": "low",
+  "input": {
+    "subject": "Hello",
+    "body": "Can you help?"
+  },
+  "schema": {
+    "type": "object",
+    "properties": {
+      "intent": { "type": "string" },
+      "draft": { "type": "string" }
+    },
+    "required": ["intent", "draft"],
+    "additionalProperties": false
+  }
+}'
+```
+
+## Safety notes
+
+- The tool is **JSON-only** and instructs the model to output only JSON (no
+  code fences, no commentary).
+- No tools are exposed to the model for this run.
+- Treat output as untrusted unless you validate with `schema`.
+- Put approvals before any side-effecting step (send, post, exec).