用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/tomevault-io/skills-registry --skill misatay-needs-help命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
基于 SOC 职业分类
| name | misatay-needs-help |
| description | Resolve tasks where the AI got stuck and needs user guidance Use when this capability is needed. |
| metadata | {"author":"dshearer"} |
This skill guides you through resolving tasks that are marked as needs_help. These are tasks where the AI got stuck during execution and needs the user's input to move forward.
Use this skill when:
needs_help tasksneeds_help taskneeds_help task in the task status view and asks about itneeds_help and user is ready to helpdshearer.misatay/listTasks to find tasks with status needs_helpRead the task's description — it should explain what was attempted and what help is needed (the execution skill updates the description before setting needs_help).
Present this to the user clearly:
"Task <task-id> needs your help: <task title>
Here's what happened:
<summary of what was attempted, from the task description>
What's needed:
<what specific help is needed, from the task description>"
If there are partial commits for this task, mention them so the user has full context:
git log --grep="(task-id)" --format="%H %s"
Work with the user to resolve the issue. This varies by situation:
If the task description was unclear:
If the implementation was difficult:
If the task conflicts with other tasks:
Listen to the user. They know their codebase and requirements better than you do.
Once the issue is resolved, update the task based on the outcome:
If the task needs to be re-implemented:
Use dshearer.misatay/updateTask:
- taskId: the task ID
- updates: { status: "ready", description: "<updated description with clarified requirements>" }
Then suggest switching to the Execution Skill to implement it.
If the user resolves the issue during this conversation (e.g., provides the answer and you can implement immediately):
Use dshearer.misatay/updateTask:
- taskId: the task ID
- updates: { status: "in_progress" }
Then switch to the Execution Skill workflow — implement the changes, mark as committed, and commit.
If the task should be removed or restructured:
Work with the user to decide what to do — perhaps the task should be split, merged with another, or deleted entirely.
After updating the task, confirm with the user:
"Task <task-id> is now <new status>. <Brief summary of what was decided.>
Would you like to:
- Continue with execution (implement this or other ready tasks)
- Address another needs_help task
- Do something else?"
When the user asks "what do I need to do?", "what needs attention?", or similar, always mention both:
needs_help status) - These need user input before the AI can continuecommitted status) - These are done and waiting for the user to reviewPresent needs_help tasks first since they're blocking progress:
"Here's what needs your attention:
Needs help:
- <task-id>: <title> — <brief reason from description>
Ready for review:
- <task-id>: <title>
Which would you like to address?"
If the user's guidance is still unclear after discussion, ask for more clarification rather than guessing. The whole point of needs_help is that the AI couldn't figure it out alone.
Always update the task description to reflect the resolution. Future sessions may need this context.
If there are partial commits from the failed attempt, don't try to undo them. The Execution Skill will create new commits on top.
Resolve one needs_help task before moving to the next. Don't try to batch-resolve multiple stuck tasks.
Converted and distributed by TomeVault — claim your Tome and manage your conversions.