Goal Decomposer

将高层自然语言目标拆解为可执行的多层级任务列表。 触发场景：用户给出模糊目标需要具体执行步骤、复杂任务需要拆解、需要生成任务树。

Security Scan

This skill is internally coherent: it declares no secrets or installs, its instructions and included script match the stated goal-decomposition purpose, and there are no unexpected requirements.

Purpose & Capability

Name/description (task decomposition) align with the provided code and SKILL.md. The included script implements simple template-based decomposition and the examples/README show expected outputs. There are no unrelated binaries, credentials, or config paths requested.

Instruction Scope

SKILL.md stays on-topic (decompose goals into tasks). It references integrations (sessions_spawn, RESOLVER, AUTO-FIX, web-access) and uses spawn_hint examples that instruct the agent to route sub-tasks to other capabilities (e.g., web-access for research). That is appropriate for a decomposer but means the agent may autonomously request networked sub-tasks when these platform capabilities are allowed.

Install Mechanism

No install spec is provided (instruction-only skill). A small local script (scripts/decompose.py) is included; there are no downloads, package installs, or archive extraction. No install-time execution from external URLs is present.

Credentials

The skill requests no environment variables, no credentials, and no config paths. There are no secret-like environment names or unrelated credential requests.

Persistence & Privilege

always:false and no special persistence is requested. The skill documents spawning sub-tasks (sessions_spawn) and integration points; autonomous invocation is allowed by platform default but the skill itself does not request elevated or persistent privileges.

Assessment

This skill appears safe and coherent for its stated purpose. Before installing: (1) be aware SKILL.md suggests routing research subtasks to web-access and sessions_spawn — if you don't want the agent to perform network calls or spawn other skills, restrict those platform capabilities or test in a sandbox; (2) the bundled Python script is simple and local (no hidden endpoints), but review it if you require code-auditability; (3) check the pricing/usage terms in README and the GitHub issues link if you need support; and (4) test with non-sensitive goals first to confirm behavior in your environment.

Goal Decomposer Skill

核心能力

将用户自然语言描述的目标，自动拆解为结构化的任务树：

用户目标："我想做个产品调研"
    ↓
任务树：
├─ Task 1: 确定调研范围
│  ├─ 1.1 明确产品类别
│  └─ 1.2 确定目标市场
├─ Task 2: 收集数据
│  ├─ 2.1 搜索竞品信息
│  └─ 2.2 整理市场数据
└─ Task 3: 生成报告
   ├─ 3.1 分析优劣势
   └─ 3.2 输出结论

设计原则

1. MECE原则：任务互斥完全穷尽
2. 可执行性：每个子任务可独立执行
3. 层级清晰：不超过3层深度
4. 优先级排序：标注P0/P1/P2

输入格式

{
  "goal": "用户自然语言目标",
  "context": "背景信息（可选）",
  "constraints": ["约束条件（可选）"]
}

输出格式

{
  "root_goal": "原目标",
  "tasks": [
    {
      "id": "T1",
      "title": "任务标题",
      "priority": "P0",
      "children": [
        {"id": "T1.1", "title": "子任务", "priority": "P0"}
      ],
      "spawn_hint": "建议spawn方式（可选）"
    }
  ],
  "execution_order": ["T1", "T2", "T3"]
}

使用方式

# 方式1：直接调用
import json
result = decompose_goal("做个产品调研")

# 方式2：通过sessions_spawn
sessions_spawn(agentId="goal-decomposer", task="拆解目标：做个产品调研")

与现有能力集成

能力	集成方式
RESOLVER	任务生成后路由到对应skill
sessions_spawn	子任务可独立spawn执行
AUTO-FIX	拆解失败触发修复
web-access	调研类任务联网获取数据

实现脚本

见 scripts/decompose.py

---

⚠️ 常见坑

症状	原因	解决办法
拆解层级不清	目标表述模糊	明确"做什么+为什么做"
任务过于碎片	缺少MECE检查	启用MECE验证开关
依赖关系缺失	未标注前提任务	检查是否需要前置条件
执行顺序混乱	依赖标注遗漏	补充"depends_on"字段