AI Workbench
0%

AI Workbench

Automated office workspace powered by Claude Code

v v v

Overview

AI Workbench is an open-source automated office workspace by AI Lingnan, powered by Claude Code. Drop documents into inbox/, and the system automatically identifies the file type, routes it to the appropriate Skill for processing, archives the output, and updates all indexes.

Architecture

Architecture: inbox drop-off → tasks processing → archive storage, with three-tier index sync

Quick Start

1

Clone the Project

git clone https://github.com/ai-lingnan/ai-workbench.git
cd ai-workbench
2

Initialize

Send this prompt to Claude Code:

Read the README.md of this project, then initialize the workspace.
3

Explore

Tell me what this project can do and how it works.
That's it: two prompts and Claude handles all the configuration. After that, just drop task folders into inbox/.

Workflow

A task goes through 5 steps from start to finish: drop files, write a task note, route to a Skill, produce output, update index & git commit. You only participate in the first two.

Workflow

Drop → Note → Route → Output → Commit

Skill Routing

When writing the task note, you and Claude decide which Skill to use. Here's the mapping between file types and Skills.

Skill Routing Table

File type → Skill mapping

Automation Loops

Two background scheduled tasks keep watch for you, running automatically around the clock.

Automation Loops

inbox-monitor handles tasks, daily-routine generates reports

Directory Structure

DirectoryPurposeGit
inbox/Task drop-off boxignored
tasks/Task workspaceignored
reports/Daily & weekly reportsignored
daily/Daily file indexignored
demo/Example taskstracked
templates/Document templatestracked
.claude/Config, rules, Skillstracked
changelog.mdGlobal task logignored

Index System

IndexGranularityContent
changelog.mdTask-levelTime, task name, source files, output
daily/*.mdDay-levelAll files processed that day
README.mdDirectory-levelSummary index of entries in each directory

Workflow Details

Drop a Task

Place task folders into inbox/. Each folder = one task, and can contain multiple source files.

inbox/
├── Q1-Analysis/
│   ├── data.xlsx
│   └── template.docx
└── Meeting-Notes/
    └── recording.pdf

Generate a Task Note

Tell Claude what you want done. It generates a task.md note documenting the requirements and processing plan.

## Requirements
Generate Q1 data analysis report using the template
## Plan
- Skill: minimax-xlsx
- Steps: Read → Analyze → Output

PDF: Three Options

ScenarioSkillDescription
PDF → MarkdownmineruExtract text content
Merge/Split/ExtractpdfManipulate existing PDFs
Generate high-quality PDFminimax-pdfDesign-grade output

Loop Details

inbox-monitor Every 20 min

  • Scans inbox/ for subfolders
  • Folders with task.md enter processing
  • Moved into tasks/ and renamed
  • Routed to the appropriate Skill
  • Indexes updated + git commit

daily-routine Every 24 hours

  • Reads the day's changelog and git log
  • Generates a daily report (with retrospective)
  • On Fridays, also generates a weekly report
  • Saved to reports/
  • Updates README index
GitHub · AI Lingnan