The bot basically stalks my git commits from the last 24 hours and formats them into a "What I Did / What's Blocked / What's Next" list. It runs automatically every weekday at 9 AM and dumps the history into a custom dashboard page. It's a legitimate AI workflow that saves me from staring at a blank Slack channel for ten minutes.
If you want to try this from scratch, here is the actual blueprint.
The Architecture #
An app here isn't just a script; it's a bundle. You can mix and match agents, MCP servers, and UI pages. For this standup bot, the file structure looks like this:
standup-bot/
├── app.json ← manifest (identity + resources)
├── agents/
│ └── standup-agent.json ← agent definition
├── skills/
│ └── standup-format/
│ └── SKILL.md ← formatting rules
└── ui/
└── src/App.tsx ← dashboard page
Step-by-Step Deployment #
- The Manifest: You need an
app.json
. This is where you tell the system what the hell this app is and what files it needs to load.
{
"name": "standup-bot",
"version": "1.0.0",
"displayName": "Daily Standup Bot",
"description": "Auto-generates standup notes from git commits.",
"author": "sarvar_04",
"agents": ["agents/standup-agent.json"],
"skills": ["skills/standup-format"],
"ui": {
"entry": "dist/index.mjs",
"pages": [{
"route": "/apps/standup-bot",
"label": "Standups",
"icon": "ClipboardList"
}]
},
"crons": [{
"name": "morning-standup",
"cron_expr": "0 9 * * 1-5",
"message": "Generate today's standup summary from my recent git logs."
}]
}
- The Agent: Create
agents/standup-agent.json
. This defines the LLM's persona. I keep mine focused so it doesn't start waxing poetic about my commit messages.
- The Skill: This is just a markdown file in
skills/standup-format/SKILL.md
. It teaches the agent exactly how to format the output so it doesn't look like a wall of text.
- The Dashboard: You can actually write a React component in
ui/src/App.tsx
to create a dedicated page in the sidebar. This is where the bot posts the summaries.
- The Automation: The
crons
section in the manifest handles the scheduling. No need to mess with system-level crontabs and pray they work; it's all handled by the orchestrator.
The beauty of this setup is that it's isolated. You aren't just adding a prompt to a global list; you're building a versioned tool. Once it's done, you can package it and other people can install it with a single command. It's basically an App Store for LLM agents. If you're tired of writing the same five prompts every morning, this is the way to actually automate the boredom.
Next GitHub Copilot Autofix can introduce security holes if you trust →
a practical ChatGPT prompt guide, with plenty of directly applicable cases.