AgentSkillsCN

team-templates

生成标准化的团队模板（项目启动会、数据契约、事件事后复盘/RCA、每周状态更新、Unity Catalog访问申请）。当有人提出需要模板、PRD/一页纸概要、设计文档、事后复盘、每周更新、数据契约，或访问申请时使用。

SKILL.md
--- frontmatter
name: team-templates
description: Generate standardized team templates (project kickoff, data contract, incident postmortem/RCA, weekly status update, Unity Catalog access request). Use when someone asks for a template, PRD/1-pager, design doc, postmortem, weekly update, data contract, or access request.
license: Proprietary. For internal team use.
compatibility: Designed for Databricks Assistant agent mode in Databricks workspaces.
metadata:
  author: your-team
  version: "1.0"

Team templates library

Use this skill to generate consistent, copy/paste-ready templates for common data/analytics team workflows.

Template catalog

Pick the best match based on the user request:

  • •Project kickoff / 1‑pager → assets/kickoff.md
  • •Weekly status update (Slack + longer-form) → assets/weekly-status-update.md
  • •Incident postmortem / RCA (Sev/Outage) → assets/incident-postmortem.md
  • •Data contract (dataset/table contract in YAML) → assets/data-contract.yml
  • •Unity Catalog access request (permissions + optional GRANT SQL) → assets/unity-catalog-access-request.md

How to use this skill

1) Identify the template type

If the user explicitly names one (kickoff, postmortem, weekly update, data contract, access request), use it.

If not explicit, infer from keywords:

  • •“RCA”, “postmortem”, “outage”, “incident”, “Sev” → incident postmortem
  • •“1‑pager”, “kickoff”, “proposal”, “PRD”, “new project” → kickoff doc
  • •“data contract”, “schema contract”, “dataset contract” → data contract YAML
  • •“grant”, “permission”, “Unity Catalog”, “access to schema/table” → UC access request
  • •“weekly update”, “status”, “what did we do this week” → weekly status update

2) Ask for only missing inputs (max 3 questions)

If required details are missing, ask up to three targeted questions. Prefer defaults over interrogations.

Common inputs:

  • •Names: project/dataset/table, team, owner
  • •Dates: start date, incident date/time range
  • •Audience: internal team vs stakeholders vs exec summary
  • •Access scope: catalog.schema.table, permission level (SELECT/MODIFY/OWN), duration

If the user provided none, assume:

  • •Audience: “internal + stakeholders”
  • •Timezone: user local (unknown) → label as “local time”
  • •Dates: today’s date is fine only if user asked “today”; otherwise leave placeholders.

3) Generate a filled template (deliverable-first)

Output the chosen template in Markdown (or YAML for the data contract), with placeholders replaced when you have information.

Rules:

  • •Keep headings intact so the doc is scannable.
  • •Keep checklists as - [ ] so teams can track work.
  • •Use neutral language. Avoid blaming individuals.

4) Add a short “Next steps” section

After the template, add 3–8 concrete next steps tailored to the template type.

Examples

Example A — kickoff

User: “Create a kickoff doc for a new customer churn dashboard in Databricks.”

Assistant output: A filled version of assets/kickoff.md including goals, stakeholders, datasets, milestones, and a definition of done.

Example B — postmortem

User: “We had a 2‑hour outage because the DLT pipeline failed. Draft the postmortem.”

Assistant output: A filled version of assets/incident-postmortem.md with impact, timeline, root cause hypotheses, and action items.

Edge cases

  • •User wants Confluence/Google Doc formatting: Output Markdown, but keep headings and checklists; it will paste cleanly.
  • •User wants multiple templates: Output one at a time; ask which is highest priority.
  • •User provides sensitive data: Keep it high-level; replace secrets with [REDACTED].
  • •Unknown incident details: Keep placeholders and add a section “Information needed” listing what’s missing.