Agents Overview¶

polymarket-go uses multi-agent-spec format for defining AI agents. This allows the same agent definitions to be deployed across different platforms.

Agent Team¶

The default trading team consists of three specialized agents:

graph LR
    A[market-analyst] -->|market_candidates| B[superforecaster]
    B -->|forecasts| C[trader]

Agent	Role	Model	Tools
market-analyst	Discover opportunities	sonnet	WebSearch, WebFetch, Read, Write
superforecaster	Generate probabilities	sonnet	WebSearch, WebFetch, Read
trader	Execute trades	haiku	Read, Write

Agent Definition Format¶

Agents are defined in Markdown files with YAML frontmatter:

---
name: agent-name
namespace: optional-namespace
description: What the agent does
model: sonnet|opus|haiku
tools: [Tool1, Tool2]
role: Agent's role
goal: What the agent tries to achieve
backstory: Background context
dependencies: [other-agent]
---

# Instructions

Detailed instructions for the agent in Markdown format.

Required Fields¶

Field	Type	Description
`name`	string	Unique agent identifier
`model`	string	LLM model to use
`tools`	array	Available tools

Optional Fields¶

Field	Type	Description
`namespace`	string	Agent namespace (derived from directory if not set)
`description`	string	Brief description
`role`	string	Agent's role in the team
`goal`	string	Primary objective
`backstory`	string	Context and background
`dependencies`	array	Other agents this one depends on

Workflow Types¶

The team workflow supports multiple execution patterns:

Deterministic (Schema-Controlled)¶

Type	Description
`chain`	Sequential execution A → B → C
`scatter`	Parallel execution, results merged
`graph`	DAG-based with dependencies

Self-Directed (Agent-Controlled)¶

Type	Description
`crew`	CrewAI-style delegation
`swarm`	OpenAI Swarm-style handoffs
`council`	Multi-agent deliberation

Deployment Targets¶

The same agent specs can deploy to:

Platform	Config File	Use Case
Go Server	`deployment-go-server.json`	Production
Claude Code	`deployment-claude-code.json`	Development

See Team Configuration for workflow and deployment details.