Automated Daily Standup with Claude Code and launchd

If you write standups manually every day, you are probably spending 10–20 minutes on repetitive context switching.

This setup automates the full loop:

trigger at a fixed time
gather updates from GitHub + Linear
generate a clean summary
publish to Notion
keep logs for traceability

System architecture

Automated standup architecture

At a high level:

launchd starts a script every weekday at 13:30.
The script calls Claude Code in non-interactive mode.
Claude reads a strict prompt and collects updates from tools.
Claude writes the standup to Notion and exits.
The script stores logs for debugging and auditability.

Why this is reliable in practice

deterministic schedule (launchd)
weekend skip logic in the script
timeout watchdog to avoid hung runs
timezone-aware reporting window
local run logs for fast troubleshooting

A key detail: for PRs, use merged_at in your prompt logic, not created_at. Otherwise, old PRs merged today can be missed.

Reporting window design (the part most people get wrong)

Reporting window diagram

Standup quality depends on the time window. The safest approach is:

report from previous cutoff to current run time
on Monday, include Friday→Monday gap
keep one explicit timezone constant in both script and prompt

This avoids duplicates and missing updates around day boundaries.

Prerequisites

macOS (this guide uses launchd)
Claude Code CLI installed
GitHub CLI (gh) installed and authenticated
MCP integrations configured for Linear + Notion
a Notion destination page for daily standups

Non-interactive runs cannot answer permission prompts. If your setup requires --dangerously-skip-permissions, keep your prompt scoped to trusted tools only.

Step 1: prompt file

Create ~/.claude/standup-prompt.txt.

Keep the prompt strict and structured:

define exact reporting window rules
require grouped output sections (PRs, tickets, blockers, next)
define fallback behavior when today's Notion section does not exist
forbid any actions outside standup generation

Example prompt skeleton:

TXT

You are generating a daily standup update.

Timezone: Europe/Bucharest
Current run time: provided by system clock.

Reporting window:
- Standard days: previous run cutoff -> current run time.
- Monday: previous Friday cutoff -> current run time.

Data sources:
1) GitHub via gh CLI (author = me, use merged_at for PRs)
2) Linear via MCP (my tickets updated/closed in window)
3) Notion via MCP (append under today's section)

Output format (strict):
## Yesterday
- ...

## Today
- ...

## Blockers
- ... or "None"

Fallback behavior:
- If today's section is missing, append to latest visible daily section and include a NOTE line.

Safety constraints:
- Do not create unrelated pages.
- Do not modify historical entries outside the current target section.

Step 2: runner script

Create ~/.claude/standup-runner.sh and make it executable.

BASH

#!/usr/bin/env bash
set -euo pipefail

BASE_DIR="$HOME/.claude"
PROMPT_FILE="$BASE_DIR/standup-prompt.txt"
LOG_DIR="$BASE_DIR/standup-logs"
mkdir -p "$LOG_DIR"

DAY="$(date +%F)"
LOG_FILE="$LOG_DIR/$DAY.log"

FORCE=false
SHOW_LOGS=false
CRON_MODE=false

for arg in "$@"; do
  case "$arg" in
    --force) FORCE=true ;;
    --logs) SHOW_LOGS=true ;;
    --cron) CRON_MODE=true ;;
  esac
done

if [ "$SHOW_LOGS" = true ]; then
  tail -n 200 "$LOG_FILE" 2>/dev/null || echo "No log yet for $DAY"
  exit 0
fi

# Skip weekends unless manually forced
DOW="$(date +%u)" # 1..7 (Mon..Sun)
if [ "$FORCE" = false ] && { [ "$DOW" -eq 6 ] || [ "$DOW" -eq 7 ]; }; then
  echo "[$(date)] Weekend detected, skipping." >> "$LOG_FILE"
  exit 0
fi

CMD=(
  claude
  --print
  --dangerously-skip-permissions
  -p "$(cat "$PROMPT_FILE")"
)

# 5-minute watchdog
if command -v gtimeout >/dev/null 2>&1; then
  TIMEOUT_BIN="gtimeout"
else
  TIMEOUT_BIN="timeout"
fi

{
  echo "===== $(date) ====="
  "$TIMEOUT_BIN" 300 "${CMD[@]}"
  echo ""
} >> "$LOG_FILE" 2>&1

if [ "$CRON_MODE" = false ]; then
  echo "Standup run complete. Log: $LOG_FILE"
fi

BASH

chmod +x ~/.claude/standup-runner.sh

Step 3: launchd job

Create ~/Library/LaunchAgents/com.claude.standup.plist:

XML

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
  <dict>
    <key>Label</key>
    <string>com.claude.standup</string>

    <key>ProgramArguments</key>
    <array>
      <string>/bin/bash</string>
      <string>-lc</string>
      <string>~/.claude/standup-runner.sh --cron</string>
    </array>

    <key>StartCalendarInterval</key>
    <dict>
      <key>Hour</key><integer>13</integer>
      <key>Minute</key><integer>30</integer>
    </dict>

    <key>RunAtLoad</key>
    <false/>

    <key>StandardOutPath</key>
    <string>~/.claude/standup-logs/launchd-stdout.log</string>

    <key>StandardErrorPath</key>
    <string>~/.claude/standup-logs/launchd-stderr.log</string>
  </dict>
</plist>

Load and verify:

BASH

launchctl unload ~/Library/LaunchAgents/com.claude.standup.plist 2>/dev/null || true
launchctl load ~/Library/LaunchAgents/com.claude.standup.plist
launchctl list | grep claude.standup

Operational commands

BASH

# Manual test
~/.claude/standup-runner.sh

# View today's run log
~/.claude/standup-runner.sh --logs

# Restart after plist edits
launchctl unload ~/Library/LaunchAgents/com.claude.standup.plist
launchctl load ~/Library/LaunchAgents/com.claude.standup.plist

Troubleshooting

Nothing runs
- verify agent is loaded with launchctl list | grep claude.standup
- check launchd stderr log for path/env issues
Notion update fails
- re-auth MCP integrations interactively and re-run manually
- validate fallback logic for missing daily section
Laptop sleeping during run
- this setup requires machine awake + internet at runtime
- if reliability is critical, move the same script to an always-on server
Wrong standup content window
- ensure timezone is explicitly set in both prompt and script assumptions
- check Monday override logic

Do not allow broad autonomous actions in this workflow. Keep scope narrow: read activity, write one standup destination, exit.

Final thoughts

This automation works best when you treat it like a small production system:

clear inputs
deterministic schedule
explicit safety boundaries
observability via logs

If you do that, your standup becomes reliable infrastructure instead of daily manual overhead.