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/marblo-app/marblo --skill tf-feedback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。
Use before merging agent-generated code — review a diff for correctness, security, and simplicity, and report findings ranked by severity.
Analyze requirements and map out components, roles, and dependencies.
Bulk-create tasks in Marblo MCP from the analysis results.
基于 SOC 职业分类
正在显示 SKILL.md
| name | tf-feedback |
| description | Check and respond to PM feedback left on the dashboard — a two-way communication channel. |
| disable-model-invocation | true |
| allowed-tools | Bash, Read, Glob, Grep, Write, Edit |
Check and reply to the feedback (comments) the PM left on tasks in the dashboard. A two-way communication channel that catches feedback an agent missed during work.
Never use the Claude Code built-in tools (TaskCreate, TaskList, TaskUpdate, TaskGet). Feedback check/reply MUST use the Marblo MCP tools:
check_feedback,acknowledge_feedback,add_activity,get_task_activities,get_all_tasks
Use check_feedback to fetch the unread feedback the PM left.
💬 Unread PM feedback: {n}
━━━━━━━━━━━━━━━━━━━━━━━
1. TASK-003: AI summary API (IN_PROGRESS)
💬 "Can you explain the parser implementation?"
— 11 minutes ago
2. TASK-005: Main screen (REVIEW)
💬 "Please check responsiveness. Something gets cut off on mobile."
— 2 hours ago
Which feedback should we handle first? (number or 'all')
✅ No unread feedback
💡 When the PM leaves feedback on the dashboard, you can check it here.
For each piece of feedback:
get_task_activities to review the full activity log (get context).| Feedback type | Response |
|---|---|
| Question ("explain this", "why this way?") | Analyze the code and write an answer |
| Change request ("change this", "add this") | Modify the code and report the result |
| Confirmation request ("is this right?", "is it tested?") | Verify, then answer with the result |
| Direction ("do it this way") | Confirm the approach + answer with a plan to apply it |
Record a reply to the PM feedback with add_activity:
add_activity:
task_id: {task_id}
message: "💬 PM feedback reply: {detailed answer}"
add_activity:
"💬 PM feedback applied: {file modified} — {what changed}"
Use acknowledge_feedback to mark the feedback as read.
💬 Feedback handling complete
━━━━━━━━━━━━━━━━━
Checked/replied: {n}
Code changes: {n}
Still unread: {n}
Reply log:
• TASK-003: explained the parser implementation structure
• TASK-005: fixed mobile responsive CSS
This skill is used during other skills as well:
/tf-work: calls check_feedback during work → notifies if there is feedback/tf-resume: calls check_feedback on resume → checks feedback that arrived in the meantime/tf-hold: shows whether there is unread feedback when pausingBut when the PM sends urgent feedback → use /tf-feedback to check and reply immediately.