원클릭으로
aim-create-tasks
当用户已经明确批准创建新的 AIM Task,且需要把稳定的用户意图整理成候选 Task Spec 并提交到 POST /tasks 时使用。
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
当用户已经明确批准创建新的 AIM Task,且需要把稳定的用户意图整理成候选 Task Spec 并提交到 POST /tasks 时使用。
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Required entry skill when you are an AIM Developer working on an existing AIM Task and must read the task via AIM Server, validate it against the latest baseline, execute the worktree and PR lifecycle, report field facts, and complete the bound OpenCode session.
Strongly recommended when you need to clarify questions, converge directions and uncertainties, do design or orienting work, or answer open-ended questions before deciding how to proceed and producing a task spec, especially when key clarifications could change the route.
Coordinator decision entry for AIM Task Pool maintenance; form an approvable POST /tasks/batch operations plan from Manager output, latest baseline facts, current Tasks, and rejected Task feedback before applying atomic Task Pool writes.
Use when judging how README claims compare to the latest origin/main baseline and the coordinator needs direction signals without task creation or execution decisions.
AIM Manager guidance for evaluating README goals against the latest baseline, defining iteration direction and Dimensions, and preparing Coordinator-consumable evaluation signals without creating tasks or executing work.
Use when a GitHub repository needs its merge settings, default-branch ruleset required checks, or PR auto-merge behavior verified or standardized with gh.
| name | aim-create-tasks |
| description | 当用户已经明确批准创建新的 AIM Task,且需要把稳定的用户意图整理成候选 Task Spec 并提交到 POST /tasks 时使用。 |
这个 skill 只负责一件事:把已经获批的用户意图整理成候选 AIM Task Spec,并在得到明确批准后创建 Task。
它不负责替代调度器,不负责执行 aim-verify-task-spec 之外的校验,不负责 implementation plan、实现执行、生命周期推进或 PR 跟进。
project_id。POST /tasks 时。aim-verify-task-spec。project_id。project_id 对应 Project 信息,且 Project 中存在 global_provider_id 与 global_model_id。如果用户给的只是方向、主题或一句需求,先继续访谈。
至少要补齐:
project_id 是什么。global_provider_id 与 global_model_id 是什么。不要猜 project_id 或开发模型字段,也不要把背景故事直接抄进 Spec。
起草前先做只读了解,确认最新基线、相关 AIM Task、相邻 spec 或设计文档,目标是:
Assumptions 贴近当前可验证事实。这一步只用于起草候选 Task Spec,不得顺势扩展成 implementation plan 或执行清单。
候选 Task 不能只给标题、摘要或实现提示。每个候选都必须包含完整五段。
TitleAssumptionsGoal vs Non-GoalCore PathValue Alignment参考 Markdown 格式如下:
# Custom Task Title
Descriptions Paragraphs...
## Fact Assumptions
(To clarify the current baseline, please explicitly describe the key facts and assumptions that are relevant to this Task. This will help downstream to understand the starting point and avoid misunderstandings.)
### Fact 1 Title
- What is Fact 1 (Concepts only, no implementation details)
- How to verify Fact 1
### Fact 2 Title
- What is Fact 2 (Concepts only, no implementation details)
- How to verify Fact 2
## Goal vs Non-Goal
(To clarify the scope of this Task, please explicitly describe what is in scope (Goal) and what is out of scope (Non-Goal). This will help downstream to understand the intended outcome and avoid scope creep.)
### Goal 1 Title
- What is Goal 1 (Concepts only, no implementation details)
- How to verify Goal 1 is achieved
- Why Goal 1 is important
### Goal 2 Title
- What is Goal 2 (Concepts only, no implementation details)
- How to verify Goal 2 is achieved
- Why Goal 2 is important
### Non-Goal 1 Title
- What is Non-Goal 1 (Concepts only, no implementation details)
- How to verify Non-Goal 1 is not in scope
- Why Non-Goal 1 is not in scope
### Non-Goal 2 Title
- What is Non-Goal 2 (Concepts only, no implementation details)
- How to verify Non-Goal 2 is not in scope
- Why Non-Goal 2 is not in scope
## Core Path
(To construct a proof of this path can take actual baseline to goal. Extreme High Inference Effort is needed in this section, please be very careful when writing this part. Notion of "Core Path" is very different from "Implementation Plan". Core Path should be more about the key concept changes, the main path selection and the trade-offs with alternative paths. It should not be a file change list, a function change list or a mechanical step list.)
Summary paragraph of the main path to achieve the Goal from the current baseline.
Steps of the Core Path...
- Step 1: Description of Step 1
Related concepts, why this step is needed, and how to verify it is done.
- Step 2: Description of Step 2
Related concepts, why this step is needed, and how to verify it is done.
- Step Final: Description of the final step
How to verify the Goal is achieved after this step.
## Value Alignment
(When there are conflicts, please describe the conflict scenario and the value priority in this section. This will help downstream to make informed trade-offs when they encounter similar conflicts.)
### Value 1 is better than Value 2 when...
The conflict scenario between Value 1 and Value 2, and why Value 1 is preferred in this scenario.
### Value 3 is better than Value 4 when...
The conflict scenario between Value 3 and Value 4, and why Value 3 is preferred in this scenario.
并且还要附带:
titleproject_iddependenciestitle 是 Task Schema 的独立字段,必须从 Task Spec 的标题提炼为同一个基线增量目标;不要只把标题埋在 task_spec Markdown 里。
创建前先读取 Project 信息(例如通过 GET ${SERVER_BASE_URL:-http://localhost:8192}/projects 后按 project_id 匹配),确认目标 Project 已配置 global_provider_id 与 global_model_id;创建请求只提交 project_id,不得在 Task 上复制 provider / model。
五段写法必须遵循 docs/task-spec.md 的语义:
示例:
为 Task 创建阶段补齐独立的 Spec 校验门Task 创建示例:
当前 Task 创建 skill 允许起草者直接自我判断 Spec 是否可创建。之前大家在创建 Task 时经常觉得流程不够严格,所以需要加强。Goal 和 Non-Goal 必须一起把范围夹紧。Goal 说明这次增量要推进到哪里。Non-Goal 要具体到足以阻止范围蔓延,不能只是礼貌性地说“暂不处理其他问题”。示例:
要求每个候选 Task Spec 在创建前都经过独立校验。不改动调度器排序逻辑,不在创建阶段引入 implementation plan 或执行状态字段。其他优化后续再说。示例:
把“候选 Spec 起草”和“Spec 可创建性校验”拆成两个职责:起草侧只负责表达用户批准的增量,校验侧独立判断该增量在最新基线上是否仍然成立;不采用起草者自检路径,因为那会把起草偏见带入创建门。修改 SKILL.md,并在创建前多调用一个 skill。示例:
当“更快创建 Task”和“避免把失效 Spec 写入系统”冲突时,优先后者;当“覆盖更多候选方向”和“保持每个候选都有清晰边界”冲突时,优先后者。优先保持简单。在任何创建动作之前,每一个候选 Task Spec 都必须先调用 aim-verify-task-spec 做校验。
这道校验必须通过 SubAgent 派发完成,不能由同一个起草 agent 一边写候选、一边自己判定“已经没问题”。
要求如下:
aim-verify-task-spec。pass 时,候选才能进入用户审批;waiting_assumptions 或 failed 都不得进入创建。如果一次生成多个候选 Task Spec,且它们之间没有相互依赖,应并行派发多个 SubAgent 分别校验,避免串行拖慢确认过程。
不要在这里重复 aim-verify-task-spec 的完整判定细节;创建前校验的职责边界以该 skill 为准。
候选经过独立校验后,先把候选列表返回给用户确认。
只有用户明确表示批准创建,才能调用 POST /tasks。像“看起来差不多”“先这样吧”这类模糊反馈,不算批准。
如果用户要求修改候选,回到访谈或起草步骤,重新形成候选并再次经过独立校验。
创建目标是 POST ${SERVER_BASE_URL:-http://localhost:8192}/tasks。
示例:
curl -X POST "${SERVER_BASE_URL:-http://localhost:8192}/tasks" \
-H "Content-Type: application/json" \
--data '{
"title": "Custom Task Title",
"task_spec": "# Title\n\n## Assumptions\n...",
"project_id": "00000000-0000-4000-8000-000000000001",
"dependencies": ["task-id-a"]
}'
title、task_spec 和 project_id 是创建必需字段。global_provider_id 与 global_model_id 读取,不写入 Task 创建请求。dependencies 只是软提示,不是调度门禁。project_id:停止并补齐,不要猜测。global_provider_id / global_model_id:停止并暴露阻塞,不要猜测开发模型字段。POST /tasks。POST /tasks 失败:区分哪些候选已创建、哪些失败,以及失败原因。报告失败时,区分内容问题和传输问题:
title、task_spec、project_id 或开发模型字段缺失,请求体格式错误。aim-verify-task-spec 本身的校验逻辑。dependencies 升级为硬性顺序约束。aim-verify-task-spec,禁止自我批准。Core Path 写成文件修改清单。
修正:回到概念变化、主路径选择和备选路径取舍。dependencies 当成必须先完成的硬门。
修正:只把它当成调度提示,最终顺序仍由调度器和最新基线决定。