GeneralBots/botbook

Fork 0

Rodrigo Rodriguez (Pragmatismo) 217a8a73c5 Rename Auto Tasks to Tasks, fix SVG with proper PROMPT.md guidelines

2025-12-12 23:34:30 -03:00

8.6 KiB

Raw Blame History

Tasks

The machine does the work

What is Tasks?

Tasks is where the machine does the work for you. Instead of manually tracking to-do items, you describe what you want to accomplish in natural language, and the system compiles your intent into an executable plan with automatic step-by-step execution.

Old Way	Tasks Way
Create a task	Describe your goal
Do it yourself	Machine plans it
Mark complete	Machine executes it
Repeat	You approve critical steps

How It Works

Step	What Happens
1. Describe	Write what you want in plain English
2. Compile	LLM analyzes intent, generates execution plan
3. Review	See steps, estimates, risks before execution
4. Execute	System runs the plan, pausing for approvals
5. Monitor	Watch progress, make decisions when needed

Creating a Task

Write Your Intent

In the intent box, describe what you want to accomplish:

Good Examples
"Make a financial CRM for Deloitte with client management and reporting"
"Create a website that collects leads and sends them to Salesforce"
"Build an automated email campaign for our product launch"
"Analyze Q4 sales data and generate a PDF report with charts"

Tips for better results:

Be specific about the outcome you want
Mention the client or project name
Include key features or requirements
Specify integrations if needed

Choose Execution Mode

Mode	Best For	How It Works
Semi-Automatic	Most tasks	Runs automatically, pauses for high-risk steps
Supervised	Learning/sensitive	Pauses before each step for your approval
Fully Automatic	Trusted workflows	Runs everything without stopping
Dry Run	Testing	Simulates execution without making changes

Set Priority

Priority	Meaning
Critical	Urgent, run immediately
High	Important, prioritize
Medium	Normal priority (default)
Low	Run when resources available
Background	Run during idle time

Click Compile & Plan

The LLM will:

Extract entities (action, target, domain, client)
Generate an execution plan with ordered steps
Assess risks and estimate resources
Generate the underlying BASIC program

Understanding the Execution Plan

After compilation, you see a detailed plan:

Plan Header

Field	Description
Plan Name	Auto-generated title for your task
Description	Summary of what will be accomplished
Confidence	How confident the LLM is in the plan (aim for 80%+)
Risk Level	Overall risk assessment (None/Low/Medium/High/Critical)
Estimated Duration	How long execution should take
Estimated Cost	API and compute costs

Execution Steps

Each step shows:

Field	Description
Step Number	Order of execution
Priority	CRITICAL, HIGH, MEDIUM, LOW
Step Name	What this step does
Keywords	BASIC keywords that will be used
Risk Level	Risk for this specific step

Plan Actions

Button	Action
Discard	Delete the plan, start over
Edit Plan	Modify steps before execution
Simulate	Preview impact without executing
Execute	Start running the plan

Monitoring Tasks

Task States

Status	Meaning
Draft	Created, not yet compiled
Compiling	LLM is generating the plan
Ready	Plan generated, waiting to start
Running	Currently executing
Paused	Execution paused by user
Pending Approval	Waiting for you to approve a step
Waiting Decision	Needs your input to continue
Completed	Successfully finished
Failed	Encountered an error
Cancelled	Stopped by user

Filter Tabs

Tab	Shows
All	Every task regardless of status
Running	Currently executing tasks
Need Approval	Tasks waiting for your approval
Decisions	Tasks needing your input

Progress Tracking

Each task shows:

Current step number and name
Progress bar with percentage
Time elapsed and estimated remaining

Approvals & Decisions

When Approvals Are Required

High-impact actions require your approval:

Sending emails to many recipients
Modifying production databases
Making external API calls with side effects
Deploying to live environments
Actions exceeding cost thresholds

Approval Dialog

When approval is needed, you see:

Field	Description
Action description	What will happen
Impact summary	What could be affected
Risk level	How risky this step is
Simulation result	Preview of the outcome

Options:

Action	Result
Approve	Continue execution
Defer	Decide later
Reject	Skip this step or stop execution

Making Decisions

Sometimes the system needs your input:

Choosing between alternative approaches
Selecting from multiple options
Providing missing information

Each option shows pros, cons, and impact estimates.

Task Actions

Action	When Available	What It Does
Details	Always	View full task information
Simulate	Before execution	Preview impact
Pause	While running	Temporarily stop execution
Resume	When paused	Continue execution
Cancel	Anytime	Stop and discard task

Creating Tasks from Chat

You can also create tasks by talking to your bot:

You: "I need to build a customer portal for Acme Corp"

Bot: "I'll create a task for that. Here's the plan:

5 steps, estimated 3 hours
Includes: database setup, authentication, dashboard, API integration
Risk: Low

Should I execute this plan?"

You: "Yes, go ahead"

Bot: "Task started. I'll notify you when approvals are needed."

Generated BASIC Code

Every task generates a BASIC program behind the scenes. You can view and copy this code:

' AUTO-GENERATED BASIC PROGRAM
' Plan: Financial CRM for Deloitte

PLAN_START "Financial CRM", "Client management system"
  STEP 1, "Create database schema", CRITICAL
  STEP 2, "Setup authentication", HIGH
  STEP 3, "Build client module", HIGH
PLAN_END

NEW_TABLE "clients"
  COLUMN "name", STRING
  COLUMN "email", STRING
  COLUMN "revenue", DECIMAL
SAVE_TABLE

This code can be:

Copied for manual execution
Modified and saved as a template
Reused for similar projects

Best Practices

Writing Good Intents

Do	Don't
"Create a sales dashboard for Q4 data with charts showing revenue by region"	"Make something"
"Build an email drip campaign: welcome email, 3-day follow-up, 7-day offer"	"Do the thing we discussed"
"Analyze customer feedback CSV and generate sentiment report"	"Fix it"

Choosing Execution Mode

Situation	Recommended Mode
New to Tasks	Supervised
Routine tasks	Semi-Automatic
Trusted, tested workflows	Fully Automatic
Experimenting	Dry Run

Managing Risk

Review the risk assessment before executing
Use Simulate for high-risk tasks
Set up approval thresholds in settings
Monitor running tasks actively

Troubleshooting

Compilation Failed

Check that your intent is clear and specific
Avoid ambiguous language
Include necessary context (client name, data sources)

Task Stuck on Running

Check if an approval is pending
Look for decision requests
Review the execution log for errors

Unexpected Results

Review the generated plan before executing
Use Dry Run to test first
Check the BASIC code for issues

Settings

Configure Tasks behavior in Settings:

Setting	Description
Default execution mode	Your preferred mode
Approval thresholds	When to require approval
Cost limits	Maximum spend per task
Notification preferences	How to alert you
Auto-cleanup	Remove completed tasks after X days

8.6 KiB Raw Blame History