Executive Assistant

A reusable AI executive assistant built around one file. State lives outside the chat, updates append rather than overwrite, and “what’s next” always has a real answer.

DETAIL LEVEL

Architecture. Genericized skeleton published in skills/executive-assistant/; the live version reads a private task board, which stays personal.

STACK

Claude Code

The problem

A task that isn’t urgent can quietly slip when it only lives in a chat thread. And checking status the obvious way, my phone, risks derailing into everything else waiting there. I wanted “what’s next” to have a real answer without opening it.

The approach

One living board file gets read and updated every time the skill runs. State lives in a file, not a transcript that eventually scrolls out of reach.

How it works

  • Triggers on natural phrasing (“what’s next,” “where am I”), not only when named directly
  • Reads the board first to get real status before saying anything
  • Updates task markers as things move
  • Appends a dated log entry instead of editing over the old one
  • The published skeleton documents structure and rules only; my real task board, with my actual info, stays private

Design decisions

  • One source of truth, never overwritten: a change adds a new entry rather than erasing the old one, so the board doubles as a record of how a task actually moved, not just its current state
  • Because state lives in the file, not the chat, the assistant behaves like the same teammate across conversations that share no history

What this demonstrates

A reusable tool with encoded judgment, not a to-do list wrapper: a single source of truth and an audit trail that survives across sessions.

Related Posts

Healthcare Ops & Finance

Power BI dashboard built on 143,613 real 2024 hospital discharge records for one New York region, read the way a health insurer’s medical-economics team would read it, not a hospital’s own finance view.

Read More

Operator Brain Bootstrap

The public, benchmarked version of Operator Brain. A staged interview writes real files from real answers, and a hard scope boundary is tested and measured, not just claimed.

Read More

Let's Talk