WorkflowLearner

What is WorkflowLearner?

WorkflowLearner is a tool that uses PSR.exe (the Windows Problem Steps Recorder) to record user operations into MHT files, then parses those files and feeds the extracted workflow information to a large language model (LLM) so the LLM can learn and understand the user's workflow. It is designed for users who want to automate or analyze step‑by‑step procedures on Windows.

How to use WorkflowLearner?

After cloning the repository and installing dependencies (pip install -r requirements.txt), start PSR.exe on Windows, click “Start Record”, perform your actions, then “Stop Record” to obtain a zip containing an MHT file. Run python parse_mht.py path/to/your/file.mht to extract the workflow, then python learn_workflow.py path/to/parsed/data.json to let the LLM learn it.

Key features of WorkflowLearner

• Records user operations via PSR.exe (Windows built‑in).
• Parses MHT files into structured workflow data.
• Uses an LLM to learn and understand recorded workflows.
• Works entirely on Windows with Python 3.x.
• Outputs parsed data as JSON for further processing.

Use cases of WorkflowLearner

• Automating repetitive desktop tasks by recording then replaying via an LLM agent.
• Training an LLM to follow a specific software workflow from recorded examples.
• Analyzing and documenting step‑by‑step procedures for training or compliance.

FAQ from WorkflowLearner

What is PSR.exe and do I need to install it?

PSR.exe (Problem Steps Recorder) is a Windows utility included with Windows 7, 8, and 10. You do not need to download it separately; just search for it in the start menu.

What does PSR.exe record?

It records mouse actions (clicks, drags, scrolls) and keyboard actions (only the fact that a key was pressed, not the actual typed content). The output is a zip file containing an MHT file with step‑by‑step descriptions and screenshots.

What are the runtime requirements for WorkflowLearner?

Windows operating system, Python 3.x, and PSR.exe (already part of Windows). All Python dependencies are listed in requirements.txt.

How do I parse a recorded MHT file?

Run python parse_mht.py path/to/your/file.mht in the project directory. The parsed data will be saved in the data/ folder as JSON.

How do I use an LLM to learn the workflow?

After obtaining the parsed JSON file, run python learn_workflow.py path/to/parsed/data.json. The script will feed the workflow data to the LLM for learning.