← Back to Skills Marketplace
82
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install deeptask
Description
AI 自动拆解需求与任务管理工具。实现 AI 自动拆解需求 → 人工审核 → AI 执行任务 → UT 验证 → **Git Commit** 的完整闭环。使用 SQLite 数据库管理项目、会话、子任务、MUF(最小功能单元)、单元测试。**核心特性:每完成一个 MUF 并通过 UT 后自动执行 git com...
README (SKILL.md)
DeepTask Skill
AI 自动拆解需求 → 人工审核 → AI 执行 → UT 验证 → Git Commit 的闭环流程管理工具。
核心目标
- 任务分解粒度可控:项目 → 会话 → 子任务 → MUF(最小功能单元)
- 明确状态机:每个环节有清晰的状态流转
- 人工审核精准介入:仅在关键节点(会话级)触发审核
- 失败自动熔断:7 次失败或 5 小时超时自动暂停
- 环境预检机制:执行前先检查工具链/环境是否可用
- Hello World 验证:新语言先写测试代码验证语法
- ⭐ Git Commit 追踪:每完成一个 MUF 并通过 UT 后自动执行 git commit
⭐ Git Commit 机制(核心特性)
Commit 触发条件
MUF 完成 → UT 验证通过 → 执行 git commit
Commit 信息格式
git commit -m "SE_ID:SE-1, ST_ID:ST-1, MUF_ID:MUF-1, UT_ID:UT-1"
完整流程
# ai_worker.py 中的 execute_muf 函数
def execute_muf(muf_id, project_dir, tool_name, se_id, st_id, code_content):
# 1. 更新 MUF 状态为 in_progress
db.update_status("mufs", muf_id, "in_progress")
# 2. 实现 MUF 代码(写入文件,包含追踪注释)
# 代码文件头部:# SE_ID:SE-1, ST_ID:ST-1, MUF_ID:MUF-1, UT_ID:UT-1
# 3. 运行 UT 验证
ut_passed = run_unit_test(ut_id, project_dir, tool_name)
if not ut_passed:
# UT 失败,标记为 failed,不执行 commit
db.update_status("mufs", muf_id, "failed")
return False
# 4. ⭐ UT 通过,执行 git commit
git_commit(project_dir, se_id, st_id, muf_id, ut_id)
# commit 信息:"SE_ID:SE-1, ST_ID:ST-1, MUF_ID:MUF-1, UT_ID:UT-1"
# 5. 验证 git 历史
verify_git_history(project_dir, se_id, st_id, muf_id, ut_id)
# 6. 更新 MUF 状态为 completed
db.update_status("mufs", muf_id, "completed")
return True
验证 Commit
# 查看项目 git 历史
git log --oneline
# 查找特定 MUF 的 commit
git log --grep "SE_ID:SE-1" --grep "MUF_ID:MUF-1"
# 示例输出:
# a1b2c3d SE_ID:SE-1, ST_ID:ST-1, MUF_ID:MUF-1, UT_ID:UT-1
快速开始
1. 初始化数据库
python ~/.openclaw/skills/deeptask/scripts/cli.py init
2. 创建项目
python ~/.openclaw/skills/deeptask/scripts/cli.py create_project \
--name "用户管理系统" \
--desc "支持注册/登录/权限管理" \
--summary "核心功能:RBAC 权限模型"
3. 执行完整周期(含 Git Commit)
python ~/.openclaw/skills/deeptask/scripts/ai_worker.py \
--project DT-1 \
--tool python3 \
--workspace ~/.openclaw/workspace \
--cycle
输出示例:
=== 执行 MUF: MUF-1 ===
SE: SE-1, ST: ST-1, UT: UT-1
📝 实现 MUF 代码...
✅ 代码已写入:project_DT-1/muf_1.py
🧪 运行 UT 验证...
✅ UT 验证通过
📦 执行 git commit...
✅ Git commit 成功:SE_ID:SE-1, ST_ID:ST-1, MUF_ID:MUF-1, UT_ID:UT-1
✅ Git commit 已验证
✅ MUF 完成
4. 查看 Git 历史
cd ~/.openclaw/workspace/project_DT-1
git log --oneline
5. 人工审核
# 批准会话
python ~/.openclaw/skills/deeptask/scripts/cli.py review \
--session SE-1 \
--status approved \
--reviewer "张三" \
--comments "Git commit 验证通过,代码质量良好"
执行流程(增强版)
0. 环境预检
# 检查工具链
env_ok, env_msg = check_environment("python3")
if not env_ok:
trigger_fuse("环境不可用")
1. Hello World 验证
# 验证语法
hw_ok, hw_msg = hello_world_test("python3")
if not hw_ok:
trigger_fuse("语法验证失败")
2. 初始化 Git
init_git(project_dir)
3. AI 拆解项目
项目 → 会话 → 子任务 → MUF
4. 执行 MUF(含 UT 验证和 Git Commit)⭐
MUF 实现 → UT 验证 → ✅ 通过 → Git Commit → 状态更新
❌ 失败 → 标记 failed → 不 commit
5. 更新进度
检查子任务/会话完成情况,自动更新状态
6. 检查熔断
失败 MUF 自动记录审查请求
命令行工具
可用命令
| 命令 | 说明 |
|---|---|
init |
初始化数据库 |
create_project |
创建新项目 |
list_projects |
列出所有项目 |
show_project |
显示项目详情 |
review |
人工审核 |
status |
显示系统状态 |
check_env |
检查环境 |
reset_project |
重置项目状态 |
git_log |
查看 Git Commit 历史 ⭐ NEW |
Git Log 命令示例
# 查看项目 DT-1 的 git 历史
python cli.py git_log --project DT-1
# 查看特定 MUF 的 commit
python cli.py git_log --muf MUF-1
# 验证 commit 是否包含正确的追踪信息
python cli.py git_log --project DT-1 --verify
数据库结构
核心表
- projects - 项目表
- sessions - 会话表
- subtasks - 子任务表
- mufs - 最小功能单元表
- unit_tests - 单元测试表
- review_records - 人工审查记录表
- git_commits - Git Commit 记录 ⭐ NEW
Git Commits 表结构
| 字段 | 类型 | 说明 |
|---|---|---|
| id | INTEGER | 自增 ID |
| se_id | TEXT | 会话 ID |
| st_id | TEXT | 子任务 ID |
| muf_id | TEXT | MUF ID |
| ut_id | TEXT | UT ID |
| commit_hash | TEXT | Git commit hash |
| commit_msg | TEXT | Commit 信息 |
| committed_at | DATETIME | Commit 时间 |
检查清单(每次任务必做)
执行前检查
- 工具链是否安装? (
python3 --version) - Hello World 是否通过?
- Git 是否初始化?
执行后验证
- MUF 代码已写入?
- UT 验证通过?
- Git commit 已执行? ⭐
- Commit 信息包含 SE_ID/ST_ID/MUF_ID/UT_ID? ⭐
- Git 历史可查询? ⭐
使用场景
场景 1: 完整项目周期
# 1. 创建项目
python cli.py create_project --name "电商系统" --desc "在线购物平台"
# 2. 执行完整周期(含 Git Commit)
python ai_worker.py --project DT-1 --tool python3 --cycle
# 3. 查看 Git 历史
cd workspace/project_DT-1 && git log --oneline
场景 2: 验证 Git Commit
# 1. 执行 MUF
python ai_worker.py --project DT-1 --tool python3 --cycle
# 2. 验证 commit
git log --grep "MUF_ID:MUF-1"
# 3. 输出应包含:
# abc123 SE_ID:SE-1, ST_ID:ST-1, MUF_ID:MUF-1, UT_ID:UT-1
场景 3: 审计追踪
# 查找特定会话的所有 commit
git log --grep "SE_ID:SE-1" --oneline
# 统计每个 MUF 的 commit
git log --grep "MUF_ID" --oneline | wc -l
注意事项
- Git Commit 仅在 UT 通过后执行 - UT 失败不 commit
- Commit 信息格式固定 - 必须包含 SE_ID, ST_ID, MUF_ID, UT_ID
- 代码文件包含追踪注释 - 文件头部应有
# SE_ID:xxx, ST_ID:xxx... - Git 仓库自动初始化 - 首次执行时自动
git init - Commit 失败不阻断流程 - Git 失败记录警告,继续执行
故障排查
Git Commit 未执行
# 1. 检查 UT 是否通过
python cli.py status # 查看 UT 状态
# 2. 检查是否有代码更改
cd project_dir && git status
# 3. 手动执行 commit
git add -A && git commit -m "SE_ID:SE-1, ST_ID:ST-1, MUF_ID:MUF-1, UT_ID:UT-1"
Commit 信息格式错误
# 正确格式
git commit -m "SE_ID:SE-1, ST_ID:ST-1, MUF_ID:MUF-1, UT_ID:UT-1"
# 错误格式(缺少字段)
git commit -m "完成 MUF-1" # ❌
Git 历史验证失败
# 验证 commit 是否存在
git log --grep "SE_ID:SE-1" --grep "MUF_ID:MUF-1"
# 如果找不到,检查 ai_worker.py 日志
版本历史
- v1.0.0 - 初始版本
- v1.1.0 - 增加环境预检机制
- v1.1.0 - 增加 Hello World 验证
- v1.2.0 - 增加 Git Commit 追踪机制 ⭐
- v1.2.0 - 增加 git_log 命令
- v1.2.0 - 增加 commit 验证功能
Usage Guidance
What to consider before installing/using DeepTask:
- The tool will create and write code files and run unit tests (it executes code on your machine). That means arbitrary code (AI-generated or stored tests) can run commands on your host — only use it where you trust the inputs (e.g., an isolated VM/container or an empty, disposable workspace).
- It will initialize git and automatically commit changes to whatever project directory you configure (default ~/.openclaw/workspace/project_<ID>). Ensure the workspace is not pointed at important repositories or directories containing secrets to avoid accidental commits of sensitive data.
- Review the bundled scripts (ai_worker.py, cli.py, db_manager.py) before running. Pay attention to how test_code or code_content from the DB is executed. If you need to run it, prefer a sandbox (container, VM) and disable network or restrict permissions where possible.
- Recommended mitigations: set workspace to an isolated directory, back up important repos first, inspect/approve generated code before letting the skill run cycles that auto-commit, and consider running under a user account with limited access. If you need higher assurance, request the full untruncated code paths for review of the unit-test execution functions (run_unit_test/_run_python_test/_run_moon_test) to audit how tests are invoked and whether they can spawn unbounded shell commands.
Capability Analysis
Type: OpenClaw Skill
Name: deeptask
Version: 1.2.0
The skill bundle implements an AI-driven development workflow involving task decomposition, code generation, and automated testing. It is classified as suspicious due to high-risk capabilities, specifically the execution of arbitrary code via `subprocess.run` and `python3 -c` in `scripts/ai_worker.py`, and automated Git repository manipulation (init/commit). While these actions are aligned with the stated purpose and include safety features like manual review steps and failure-based 'fuses,' the inherent risk of shell-based execution of AI-generated content presents a significant attack surface.
Capability Assessment
Purpose & Capability
Name/description promise (AI task decomposition, running unit tests, and auto git commits) matches the included scripts. The CLI, ai_worker, and db manager implement project/session/MUF lifecycle, run tooling (python/node/moon), run unit tests, write code files, initialize git, commit and verify commits — all coherent with the declared purpose.
Instruction Scope
Runtime instructions and code explicitly direct the agent to write code files into a workspace (default ~/.openclaw/workspace), run unit tests, execute subprocesses (python/node/moon/git), initialize git, and commit changes with structured messages. Executing AI-generated code and stored test_code from the DB means the skill will execute arbitrary code on the host. The SKILL.md/CLI do not require any explicit prior user confirmation for commits beyond the normal flow, and they operate on whatever project_dir/workspace is provided — raising risk of unintended commits or execution in sensitive folders.
Install Mechanism
No install spec; this is instruction+code only. No downloads or external installers are used. The code is bundled in the skill and will run via Python when invoked by the user/agent.
Credentials
The skill requests no environment variables or external credentials, which is proportionate. However it requires filesystem access to the user's workspace and will call system binaries (git, python3, node, moon) via subprocess. Because it writes files and executes them, filesystem and execution privileges are effectively required; those are normal for this tool but may be surprising to users who expect a safer/sandboxed operation.
Persistence & Privilege
The skill is not marked always:true and does not request elevated platform privileges. It stores state in a local SQLite DB under ~/.openclaw/deeptask.db and uses a workspace under ~/.openclaw/workspace — both are self-contained and expected for this application.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install deeptask - After installation, invoke the skill by name or use
/deeptask - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.2.0
AI 自动拆解需求与任务管理工具,支持 Git Commit 追踪
Metadata
Frequently Asked Questions
What is DeepTask?
AI 自动拆解需求与任务管理工具。实现 AI 自动拆解需求 → 人工审核 → AI 执行任务 → UT 验证 → **Git Commit** 的完整闭环。使用 SQLite 数据库管理项目、会话、子任务、MUF(最小功能单元)、单元测试。**核心特性:每完成一个 MUF 并通过 UT 后自动执行 git com... It is an AI Agent Skill for Claude Code / OpenClaw, with 82 downloads so far.
How do I install DeepTask?
Run "/install deeptask" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is DeepTask free?
Yes, DeepTask is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does DeepTask support?
DeepTask is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created DeepTask?
It is built and maintained by Spaceack (@spaceack); the current version is v1.2.0.
More Skills