create-agent-list

name	create-agent-list
description	Create AgentLists from web searches, descriptions, local files, or programmatic generation
allowed-tools	Read, Glob, Bash(python:*), WebSearch, WebFetch, AskUserQuestion
user-invocable	true
arguments	agent_list_description

Creating AgentLists

Generating Agents

Agents can be generated from descriptions or external sources:

# Example: Generate agents from web search results
# 1. Search for data (e.g., sports roster, company employees, historical figures)
# 2. Extract relevant traits
# 3. Build AgentList programmatically

from edsl import Agent, AgentList

# Generated from research/web data
agents = AgentList([
    Agent(name="Person A", traits={"role": "CEO", "age": 45, "company": "Acme"}),
    Agent(name="Person B", traits={"role": "CTO", "age": 38, "company": "Acme"}),
])

From a List of Agents

from edsl import Agent, AgentList

# Create agents individually
agent1 = Agent(traits={"age": 25, "occupation": "teacher"})
agent2 = Agent(traits={"age": 35, "occupation": "doctor"})

# Combine into AgentList
agents = AgentList([agent1, agent2])

Agents can take a separate name parameter e.g.,

a = Agent(name = 'John', traits={"age": 25, "occupation": "teacher"})

From External Data Sources

The from_source() method auto-detects the source type:

from edsl import AgentList

# From CSV file
agents = AgentList.from_source("people.csv")

# From Excel file
agents = AgentList.from_source("data.xlsx", sheet_name="Participants")

# From dictionary
agents = AgentList.from_source({
    "age": [25, 30, 35],
    "name": ["Alice", "Bob", "Charlie"],
    "occupation": ["teacher", "doctor", "engineer"]
})

# From pandas DataFrame
import pandas as pd
df = pd.DataFrame({"age": [25, 30], "city": ["NYC", "LA"]})
agents = AgentList.from_source(df)

With Instructions and Codebook

# Apply instructions to all agents at creation time
agents = AgentList.from_source(
    "people.csv",
    instructions="Answer as if you were this person",
    codebook={"age": "Age in years", "income": "Annual income in USD"},
    name_field="respondent_name"  # Use this column as agent names
)

# Or load codebook from a CSV file (2 columns: key, description)
agents = AgentList.from_source(
    "people.csv",
    codebook="codebook.csv"
)

Programmatically with Combinations

from edsl import Agent, AgentList
from itertools import product

# Create agents for all combinations
ages = [25, 35, 45]
occupations = ["teacher", "doctor", "engineer"]

agents = AgentList([
    Agent(traits={"age": age, "occupation": occ})
    for age, occ in product(ages, occupations)
])
# Creates 9 agents (3 ages × 3 occupations)

Quick Reference

Source	Example
List of Agents	AgentList([agent1, agent2])
CSV file	AgentList.from_source("file.csv")
Excel file	AgentList.from_source("file.xlsx", sheet_name="Sheet1")
Dictionary	AgentList.from_source({"col": [1, 2, 3]})
DataFrame	AgentList.from_source(df)

Saving / Persistence

You will create a Python file with a descriptive name e.g., 'occupation_agent_list.py' Whatever the name of your agent list, you will also save it as local JSON file:

agent.save('occupation_agent_list')

Sharing

Ask the user if they want to push that agent list to coop (Expected Parrot's servers).

Use AskUserQuestion to ask the user:

• "Should we push this agent list to Expected Parrot?"
• Options: Yes / No

If they answer 'Yes' ask them for the visibility setting with AskUserQuestion:

• "What visibility setting?
• Options: "public", "private", "unlisted"

Only proceed after receiving a response.

The description should be a short paragraph you write. The alias should be a valid URL slug e.g., 'exit-interview'

agent_list.push(
    visibility = "unlisted", 
    description = "<paragraph description of the survey>", 
    alias = "<valid url slug>"
)

After pushing, you should print the results so the user can see them. If there is any error in pushing from your parameters, update the names.