ai-chief-of-staff-starter

KO | EN

gitlite — search

#python #hacktoberfest #react #ai #typescript #llm #deep-learning #nodejs #awesome #cli #kubernetes #vue #openai #open-source

ai-chief-of-staff-starter

(★ 12)

A Starter Project to create your Chief of Staff

ai-chief-of-staff-starter 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# AI Chief of Staff — Starter

*This starter was written by my Agentic Chief of Staff.*

A tiny, opinionated template for knowledge workers and leaders who want to build
their own **agentic chief of staff**: a local folder of notes that doubles as
your AI agent's long-term memory, plus a handful of simple **skills** the agent
runs on your behalf.

It's deliberately small. The point isn't to hand you a finished product — it's
to show you the shape of one, and prove that you build it just by *talking to
your agent in plain English*. No coding required.

---

## The idea (without the showmanship)

You may have seen demos where someone gives their agents names — "Penny" the
planning coach, "Lila" the data analyst, and so on. The names are a fun framing,
nothing more. **Under the hood each one is just a skill**: a markdown file with
instructions, connected to the tools you already use. You do **not** need to
name your agents, and you don't need a new subscription — use whatever assistant
you already have at work, ideally the one with the best access to your company's
tools and data. Tool access is what gives these models their superpowers.

A few principles this template is built around:

### Your vault is the agent's long-term memory
This whole folder — your daily notes, concepts, goals, meeting notes — is the
memory your agent reads from and writes to. **Daily notes are the superpower:**
write down what you do, decide, and notice, and you give your agent enormous
context to work with.

### Coaching vs. doing — the most important question
For every skill, decide: should the agent **do** the work, or **coach** you so
you stay sharp at it? Some things you happily delegate (searching chat history,
summarising long threads — you don't need to get better at those). Other things
are core to your role and you want to *keep improving* — so the agent coaches
instead of replacing you. In this starter, `plan-my-day` and `weekly-review`
**coach**; a future "search my chat" skill would **do**. Be deliberate about
which is which.

### The training loop: v1 → v2 → v3
Skills start dumb and get smarter as you use them. The `plan-my-day` here is a
v1: it just reads yesterday's notes and your goals. A v2 might add weekly goals;
a v3 might pull in your calendar. Each version comes from actually living with
the previous one. **Onboard one skill at a time** — they need maintenance, and
a pile of half-working skills helps no one.

### A word on security
Your prompts and your vault are **sensitive surface area**. What you connect,
who can see what, and where your data goes are architecture decisions, not
afterthoughts. Be deliberate, and involve your security team early — if they're
not in the loop, you may be leaking. **Keep credentials in a proper secrets
manager / vault** (your OS keychain, a password manager, or your cloud
provider's secrets manager) rather than in plain files; only fall back to a
git-ignored `.env` if you have no vault available. And think twice before
connecting anything to a public repo.

### The golden rule of automation
**Don't automate human connection.** Summaries, analysis, planning — great. But
when you're talking to a person, use *your own voice*. People don't want to be
managed by a machine, and AI-written personal messages rarely land. Be explicit
when you do use AI to communicate.

---

## Which tools does this work in?

This is a **local folder / repo you open in an agent that can read and write
files.** It shines in file-aware, coding-style agents that auto-discover skills.
In a plain chat window you can still use everything — you just feed the skills in
yourself. Here's how triggering works per tool:

- **Cursor** — skills auto-load from `.cursor/skills/`. Open this folder as a
  project and just say the trigger phrase (e.g. "plan my day"). Cursor matches
  it to the skill's `description` and runs the steps.
- **Claude Code** — skills auto-load from `.claude/skills/`. Same idea: open the
  project and say the trigger phrase.
- **Other agents (ChatGPT, Gemini, Notion AI, the Claude.ai web app, etc.)** —
  there's **no automatic skill discovery**. The skills still work because
  they're plain English: open or attach this folder (or paste the relevant
  `SKILL.md`) into the chat and tell the agent to follow it. You trigger them
  manually rather than by a magic phrase.

> **Why are skills in two folders?** They live in **both** `.cursor/skills/` and
> `.claude/skills/` (identical copies) so the project auto-loads whether you
> open it in Cursor or Claude Code. If you only use one of those tools, feel
> free to keep that folder and delete the other for a single source of truth.

---

## Quickstart

### 1. You create skills by *just asking your agent*

That's the core move. Open this folder in your AI tool and paste something like:

```
Create a skill called `plan-my-day` in .cursor/skills/plan-my-day/SKILL.md
(and .claude/skills/plan-my-day/SKILL.md), using the same YAML-frontmatter +
markdown format as the other skills here.

It should trigger when I say "plan my day". When I do:
1. Read yesterday's daily note in daily/.
2. List unfinished tasks AND anything that looks like a follow-up even if I
   didn't write it as a task.
3. Check meta/goals.md and flag any goal that's slipping.
4. Then coach me to pick today's top 1–3 priorities and put them in order.

Keep it simple — this is a v1.
```

The agent writes the file. You read it, tweak it, and you're done. (We've
included a ready-made version of this exact skill so you can see the result.)
The `.cursor/skills/_template/SKILL.md` (and `.claude/skills/_template/SKILL.md`)
file has another copy-pasteable prompt to riff on.

### 2. Trigger an existing skill

Just say the trigger phrase to your agent while this folder is open, e.g.:

> "Plan my day."
>
> "Let's do a weekly review."
>
> "Review this article against my notes: <link>."
>
> "Sync my notes." *(after you've connected a source — see the `sync` skill)*

The agent reads the matching `SKILL.md` and follows the steps.

---

## What's in here

| Folder | What it's for |
| --- | --- |
| `daily/` | Daily notes, one per day (`YYYY-MM-DD.md`). Tasks, notes, and a focus. Your agent's richest context. |
| `meetings/` | 1:1 and meeting notes — often synced from a shared docs tool into the vault. |
| `concepts/` | Your evolving views on topics. Each note starts with a `## Summary`. |
| `meta/` | Context about you: `goals.md` and `topics.md` (what's signal vs noise). |
| `working/` | Scratch space for analysis "notebook" projects (git-ignored). |
| `.cursor/skills/` & `.claude/skills/` | Your skills — one `SKILL.md` per folder. The agent's playbook. (Two identical copies — see below.) |

### The starter skills

- **`plan-my-day`** — your planning *coach*. Reads yesterday's note, surfaces
  unfinished tasks and implied follow-ups, flags slipping goals, then helps you
  decide today's focus. (You decide — it nudges.)
- **`weekly-review`** — step back, review progress against your goals, capture
  wins, and set next week's focus.
- **`review-against-concept`** — your *librarian*. Compares a new article /
  podcast / video against your existing view in `concepts/` and tells you what's
  genuinely new and what challenges your assumptions — so you skip the
  time-sinks.
- **`sync`** — pulls meeting/1:1 notes from your external notes provider into
  `meetings/` so the vault stays your agent's memory. Append-only and
  edit-safe. **You wire in your own provider** — see its "Connect your source"
  section.
- **`_template`** — a blank skill skeleton, plus the plain-English prompt you'd
  paste to have the agent generate a brand-new skill.

---

## A note on tools

Every skill describes *intent* in plain English, so it works with any agent.
Anything that needs an external service — your calendar, chat search, a notes
provider — is written as **"connect your own tool here"** with a placeholder,
never a hard dependency on a specific vendor. Wire in whatever you already use
(an MCP server, an API token from your secrets manager / vault, or even a manual
export).

---

## Where to start

Don't try to build all of this at once. **Pick one skill, onboard it tomorrow,
then add the next.**

ai-chief-of-staff-starter

ai-chief-of-staff-starter 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

ai-chief-of-staff-starter

ai-chief-of-staff-starter 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README