SuperAPI AI Directory

task-polling

Office & productivity

View repository

Updated 10-15 21:52

About this skill

Task Polling

Purpose

The user only maintains the docs/task.mdin the project root directory, and no additional task JSON is needed. The document always contains five sections:

# 任务计划
## 未领取
## 进行中
## 已完成
## 远期规划
## 确认不做

The user places executable tasks into 未领取, and tasks must use a unique ID; titles support half-width or full-width colons, and the body is optional:

- [ ] TASK-001: 分析这个仓库,都有哪些核心模块
  - 验收&产物:生成一套wiki文档,每个模块一个md
  - 限制&约束:必须包含流程图和文件结构图

- [ ] TASK-002:修复按钮纵向不居中问题

In the section, - [ ] empty placeholder lines are ignored and will not become tasks

远期规划 and 确认不做 are maintained by the user, and the Agent will not automatically claim their contents.

Scheduling Rules

When the user asks to start, continue, or poll tasks, work according to the following protocol:

  • Prepare: first execute status, and read the current task status

    python3 <skill>/scripts/task_polling.py --json status
    
  • Poll: use /loop to check every 2 minutes, and each round execute claim

    python3 <skill>/scripts/task_polling.py --json claim
    
  • Claim: when it returns claimed , read through task.description and then execute

  • Continue: when it returns error , continue the current task according to task.description , and do not claim it again

  • Idle: when it returns idle , wait for the next round of /loop

  • Interrupt: when a new user message is received, stop polling and respond to it first

task field is id、title、description、checked. For completion or rejection, use complete、reopenrespectively, only for the current session, and do not create a Cron.

Do not repeatedly claim tasks during task execution.

Completion Rules

A task may be completed only when the code, documentation, or script is already finished, and the acceptance criteria in the task body have actually been executed:

python3 <skill>/scripts/task_polling.py --json complete --id TASK-001 --summary "测试命令及结果"

The completion command will move the task entirely into 已完成, mark it as [x] , and append the completion result. When acceptance fails, use reopen to reject 进行中, and keep the completion record:

python3 <skill>/scripts/task_polling.py --json reopen --id TASK-001 --reason "功能验收失败:具体问题"

This command only handles 已完成 tasks.

Boundaries

  • docs/task.md is the only source of task status; do not read from or write to external JSON
  • At most one 进行中 task at a time
  • Do not publish, push, delete data, or call external services without authorization
  • Do not execute long-term planning tasks early
  • Preserve the user's edits to the docs/task.md extension
  • Do not clean up 已完成 tasks without authorization; they are used as a pre-release CHANGELOG review