Apollo Writing Plans

把大任务拆成小步骤，每步几分钟，写成清晰计划再动手。适合在动手写代码前先做规划。

Security Scan

The skill's instructions and requested footprint match its stated purpose (breaking tasks into small, verifiable steps and saving plans); nothing in the package asks for unrelated credentials or installs arbitrary code.

Purpose & Capability

Name/description (task decomposition and plan generation) align with the instructions (creating step-by-step plans, saving markdown files, suggesting test/verification commands). The skill does reference running test commands (pytest) and creating files under docs/plans/, which is consistent with its goal. It does not declare required binaries (e.g., pytest) — this is not incoherent but is a small documentation gap.

Instruction Scope

SKILL.md stays within the stated scope: understand the task, break into small steps, write a plan file, and offer execution options. It instructs writing files to repo paths (docs/plans/...) and running verification commands (e.g., pytest). It also suggests spawning 'subAgent' instances to execute steps; that implies the agent may execute code or commands in the workspace. This is within the skill's purpose but increases runtime risk because executing tests or code can run arbitrary user project code — the SKILL.md does not include explicit sandboxing or safety constraints for that execution.

Install Mechanism

Instruction-only skill with no install spec and no code files — minimal risk from installation. Nothing will be downloaded or written by an installer.

Credentials

The skill requests no environment variables, credentials, or config paths. It only refers to repository file paths and test commands, which are proportional to its purpose.

Persistence & Privilege

always:false and disable-model-invocation:false (normal). The only notable privilege is that the SKILL.md promotes autonomous/subAgent execution of steps; combined with executing tests, this means an agent could run commands against the user's workspace without extra credentials. This is a functional behavior, not a hidden privilege, but users should be aware and limit autonomous execution or run it in a sandbox.

Scan Findings in Context

[NO_FINDINGS] expected: Regex scanner had nothing to analyze because this is an instruction-only skill with no code files. That absence is expected but not a safety guarantee.

Assessment

This skill appears coherent for planning and incremental execution. Before enabling it or allowing autonomous/subAgent runs: (1) confirm you want the agent to write files into your repository (docs/plans/...), (2) be cautious about letting it run tests/commands (pytest) against your code — those commands execute your code and could have side effects, so run in an isolated environment or sandbox if possible, (3) require manual approval for any step that executes code if you want tighter control, and (4) if you expect the agent to run tests, ensure required tooling (pytest, language runtimes) exists in the environment or update the skill documentation to declare them.

Apollo 写作计划

触发场景

当用户提到"写计划"、"任务分解"、"实现步骤"、"先做计划"、"执行计划"时激活。

核心原则

1. 先计划，后行动：在写代码前先写完整计划
2. 小步执行：每步控制在2-5分钟
3. 可验证：每个步骤有明确的验证方式

工作流程

Step 1：理解任务

明确：

• 用户想要实现什么？
• 技术栈是什么？
• 有什么限制条件？

Step 2：任务分解

将任务分解为独立步骤，每步：

• 一个具体操作（2-5分钟）
• 有明确的输入输出
• 有验证方式

Step 3：生成计划文档

保存到：docs/plans/YYYY-MM-DD-<功能名>.md

格式：

# [功能名称] 实现计划

目标：[一句话描述要实现什么]

## 技术方案
[2-3句话描述技术路径]

## 步骤

### 步骤1：[具体操作]
**文件：**
- 创建：`src/path/file.py`

**Step 1：写失败的测试**
[测试代码]

**Step 2：验证测试失败**
运行：`pytest tests/path/test.py -v`
预期：FAIL

**Step 3：写实现代码**
[实现代码]

**Step 4：验证测试通过**
运行：`pytest tests/path/test.py -v`
预期：PASS

### 步骤2：[...]

执行选择

计划完成后询问用户：

"计划已保存。两种执行方式：

1. 子Agent驱动（当前会话）

• 每个步骤派一个新的子Agent执行
• 每步完成后人工审核
• 快速迭代

2. 独立会话执行

• 在工作目录开新会话
• 批量执行所有步骤
• 有检查点存档

选哪种？"

注意事项

• 精确的文件路径
• 完整的代码（不是"添加验证"）
• 带预期输出的验证命令
• 小步快跑，每步都验证