daily-meeting-update | SkillBundle

Daily Meeting Update

Generate a daily standup/meeting update through an interactive interview. Never assume tools are configured—ask first.

Workflow

START
  │
  ▼
┌─────────────────────────────────────────────────────┐
│ Phase 1: DETECT & OFFER INTEGRATIONS                │
│ • Check: Claude Code history? gh CLI? jira CLI?     │
│ • Claude Code → Pull yesterday's session digest     │
│   → User selects relevant items via multiSelect     │
│ • GitHub/Jira → Ask user, pull if approved          │
│ • Pull data NOW (before interview)                  │
├─────────────────────────────────────────────────────┤
│ Phase 2: INTERVIEW (with insights)                  │
│ • Show pulled data as context                       │
│ • Yesterday: "I see you merged PR #123, what else?" │
│ • Today: What will you work on?                     │
│ • Blockers: Anything blocking you?                  │

Phase 1: Detect & Offer Integrations

Step 1: Silent Detection

Check for available integrations silently (suppress errors, don't show to user):

Integration	Detection
Claude Code History	`~/.claude/projects` directory exists with `.jsonl` files
GitHub CLI	`gh auth status` succeeds
Jira CLI	`jira` command exists
Atlassian MCP	`mcp__atlassian__*` tools available
Git	Inside a git repository

Step 2: Offer GitHub/Jira Integrations (if available)

Claude Code users: Use AskUserQuestionTool tool for all questions in this phase.

GitHub/Git:

If HAS_GH or HAS_GIT:

If yes:

Jira:

If HAS_JIRA_CLI or HAS_ATLASSIAN_MCP:

Step 3: Pull GitHub/Jira Data (if approved)

GitHub/Git — For each approved repo:

Commits by user since yesterday
PRs opened/merged by user
Reviews done by user

Jira — Tickets assigned to user, updated in last 24h

Key insight: Store results to use as context in Phase 2 interview.

Step 4: Offer Claude Code History

This integration captures everything you worked on with Claude Code — useful for recalling work that isn't in git or Jira.

Detection:

If Claude Code history exists, ask:

If yes, run the digest script:

Then present sessions with multiSelect:

Use AskUserQuestionTool with multiSelect: true to let user pick relevant items:

Key insight: User selects which sessions are work-related. Personal projects or experiments can be excluded.

Do NOT run digest script when:

User explicitly says "No" to Claude Code history
User says they'll provide everything manually
~/.claude/projects directory doesn't exist

If digest script fails:

Fallback: Skip Claude Code integration silently, proceed with interview
Common issues: Python not installed, no sessions from yesterday, permission errors
Do NOT block the standup flow — the script is supplemental, not required

Phase 2: Interview (with insights)

Claude Code users: Use AskUserQuestionTool tool to conduct the interview. This provides a better UX with structured options.

Use pulled data as context to make questions smarter.

Question 1: Yesterday

If data was pulled, show it first:

If no data pulled:

If user response is vague, ask follow-up:

"Can you give more details about X?"
"Did you complete anything specific?"

Question 2: Today

If Jira data was pulled, you can suggest:

Question 3: Blockers

Question 4: Topics for Discussion

Phase 3: Generate Update

Combine all information into clean Markdown:

Core Principles

Interview is primary — Tools supplement, they don't replace human context
Consent before access — Always ask before pulling from any integration
Context-aware questions — Show pulled data during interview to trigger memory ("I see you merged PR #123...")

Quick Reference

Phase	Action	Tool
1. Detect & Offer	Check gh/jira/claude history, ask user, pull data	Bash (silent), AskUserQuestionTool*
2. Interview	Ask 4 questions with insights	AskUserQuestionTool*
3. Generate	Format Markdown	Output text

*Claude Code only: Use AskUserQuestionTool tool for structured questions.

Claude Code Digest Script

Example Session

Output Example

Anti-Patterns

Avoid	Why (Expert Knowledge)	Instead
Run gh/jira without asking	Users may have personal repos visible, or be in a sensitive project context they don't want exposed	Always ask first, let user choose repos
Assume current directory is the only project	Developers often work on 2-5 repos simultaneously (frontend, backend, infra)	Ask "Which projects are you working on?"
Skip interview even with tool data	Tools capture WHAT happened but miss WHY and context (research, meetings, planning)	Interview is primary, tools supplement
Generate update before all 4 questions	User might have critical blocker or discussion topic that changes the narrative	Complete interview, then generate
Include raw commit messages	Commit messages are often cryptic ("fix", "wip") and don't tell the story	Summarize into human-readable outcomes
Ask for data after interview	Showing insights during interview makes questions smarter ("I see you merged PR #123, anything else?")	Pull data first, then interview with context

NEVER

NEVER assume tools are configured — Many devs have gh installed but not authenticated, or jira CLI pointing to wrong instance
NEVER skip the "Topics for Discussion" question — This is often the most valuable part of standup that tools can't capture
NEVER generate more than 15 bullets — Standup should be <2 minutes to read; long updates lose the audience
NEVER include ticket/PR numbers without context — "PROJ-123" means nothing; always include title or summary
NEVER pull data from repos user didn't explicitly approve — Even if you can see other repos, respect boundaries