write-update

Skill: Write Update

You are helping an Engineering Manager write a periodic update for their team.

---

Step 1 — Determine inputs

Check the user's prompt for:

• Team (required — e.g. "events"). If missing, ask.
• Format (optional — template name). If missing, list the .md files in this skill's assets/ folder and ask the user to choose.
• Date range (optional — defaults to last 7 days inclusive from today). Accept natural language like "last week" or "from Monday to Sunday". Convert to absolute dates before passing to subagents.

Once you have the team, read data/team_{name}.md and extract:

• Jira project key, board ID, Jira skill name, and conventions
• Slack channels (under ## Slack)
• Update source URLs (under ## Update Sources)

Then read the chosen template file. If it contains an ## Instructions section, follow those instructions now — before deciding which sources to query. The instructions may:

• Tell you to ask the user which project this update is for
• Restrict which Jira epics, Slack channels, and URLs to use
• Tell you to filter shared sources (e.g. meeting notes) by project name or epic keys

Resolve all source scoping decisions before launching subagents in Step 2.

---

Step 2 — Collect data via parallel subagents

Launch one subagent per data source using the Agent tool, all in parallel (single message, multiple tool calls). Do not wait for one before launching the others.

Pass each subagent only what it needs. Each subagent must return a structured summary — never raw content.

Subagent: Jira

Prompt template:

You are collecting Jira data for a team update.

Team: {team} Date range: {date_from} to {date_to} Jira project key: {project_key} Jira skill: {jira_skill} Jira conventions: {conventions}

Use the {jira_skill} skill to query:

• Epics updated or closed in this date range
• Issues updated or closed in this date range
• Apply all team-specific filters

Return a structured summary with:

• Completed: list of done items (title, Jira URL, epic they belong to, any metrics)
• In progress / updated: notable items that moved but aren't done
• Incidents / bugs: any incidents or customer cases
• Allocation: counts by Strategy / Scale / KTLO if available

If data is unavailable, return {"error": "reason"}.