-
Couldn't load subscription status.
- Fork 15
Open
0 / 40 of 4 issues completedOpen
0 / 40 of 4 issues completed
Copy link
Labels
Feature 🌱Research 🧐wf-backendmain rpeo: https://github.com/zk1tty/wf-backendmain rpeo: https://github.com/zk1tty/wf-backend
Description
Goal
Compare the key factors of automation of browser experiences.
| indicator | our assumption | reality |
|---|---|---|
| Speed | 20x faster | ? |
| Accuracy | 10%+ success rate of workflow | ? |
| Workflow optimization | ?? | ?? |
| Visibility | ?? | ?? |
| development time | ?? | ?? |
Preparation
- Step 1: Browser action history file with Intent annotation
- Add time stamp to transcript
- merge transcript to a user-input history file based on timestamp
transcript
{
entries: [
{ t: 15000, text: "Click login", final: true },
{ t: 23000, text: "Enter my email", final: true }
],
startedAtMs: 1000000, // Absolute starting timestamp
partial: ""
}rrweb recorder data
{
steps: [
{ type: "click", timestamp: 1016000, ... }, // Timestamps are absolute in transcript
{ type: "input", timestamp: 1023500, ... }
]
}- Step 2: Compiler to convert history file to optimized workflows
Compiler:
-
input: user-input history + intent(transcript) for each steps
-
Process:
-
- Correlate Voice with Actions
-
-
-
output: optimized workflows run by browser-use agent[actions + tools] + direct playwright(patchwright)
- is an optimized workflow a subset of [actions + tools]?
- how to design data structure:
- the list of steps with user-intent
- what is the list of user-intent?
-
Step 3: Measure the indicator's performance
Components

Sub-issues
Metadata
Metadata
Assignees
Labels
Feature 🌱Research 🧐wf-backendmain rpeo: https://github.com/zk1tty/wf-backendmain rpeo: https://github.com/zk1tty/wf-backend