AGENT SKILL

answer-workspace-questions

Use when the stakeholder asks a factual or status question about the workspace - call the matching read tool and answer conversationally, without creating tasks or asking for approval

License 许可
Apache-2.0
最近更新
2026-11-15
Stars
★ 122
Forks
18

01 前言 Front Matter

name
answer-workspace-questions
category
pm
description
Use when the stakeholder asks a factual or status question about the workspace - call the matching read tool and answer conversationally, without creating tasks or asking for approval

03 SKILL.md 正文

Answer Workspace Questions

Overview

A factual question deserves a factual answer, not a plan. The failure mode is treating "how many projects do we have?" as work to orchestrate — creating tasks, asking approval, or claiming there's "no active context." Just call the read tool and answer.

Core principle: Read tool first, then answer in one or two sentences with the real data.

The read tools

Question Tool
How many projects / what are they list_projects
What repositories exist / which project each belongs to list_repositories
Who is on the team / which role does what list_team
What's on the board / a task's status list_board_tasks
Overall status / "how are we doing" get_board_summary (aggregated by column/type/priority)
What's next / what's ready to start list_ready_tasks

These read tools are always available and never need an active repository — never answer "no active project/repo context." Call the tool and use the real data.

Rules

  1. Call the matching read tool FIRST.
  2. Prefer get_board_summary for "status/how are we doing" over listing every task — it returns the total and counts grouped by column, type and priority in one call. Only drill into list_board_tasks when the stakeholder asks about a specific task or you need titles/assignees.
  3. Answer conversationally with actual numbers/names — e.g. "We have 2 projects right now: Alpha and Beta."
  4. Translate counts into an outcome update, don't just recite them: what's in progress, what's waiting (backlog/todo), what's blocked or in review (need_revision/human_uat), and the biggest bucket (see stakeholder-communication).
  5. Do NOT create tasks, ask for approval, open analiz, or describe your internal plan for a plain question — just answer like a human PM.

Worked Example

Stakeholder: "How many repos do we have?" → call list_repositories → "We have 3 repos: backend-api, web, and mobile." Stakeholder: "Who is on the team?" → call list_team → answer from what it returns, e.g. "system-architect and security-agent review, backend, frontend, mobile, data and game developers build, qa tests — each one owns its own area." Stakeholder: "How are we doing?" → get_board_summary → {total 24; in_progress 3, code_review 2, need_revision 4, backlog 10, done 5} → "24 tasks in total. 3 are in active development, 2 are in architecture review. 4 went back for revision — that's where the biggest risk is. 10 are queued and 5 are done." Not a wall of 24 task titles.

Common Mistakes

  • Answering from memory instead of calling the tool.
  • "No active project context" — the read tools never need one.
  • Turning a plain question into an orchestration plan.
  • Paging list_board_tasks for a status question get_board_summary answers in one call.
  • Reporting raw counts with no interpretation or risk callout.

Red Flags

  • You created a task in response to a factual question.
  • You answered a count without calling a read tool.
  • A status update with no mention of the blocked/revision bucket.

04 同仓库更多 Skills