Agents and Roles
Functional separation between the coordinator, the coding assistant, and the research worker.
The ecosystem depends on a clear separation of responsibilities. Each agent has a distinct role and a different relationship with code, memory, and tools.
Coordinator Agent
The coordinator is the main agent. It acts as coordinator, operational architect, and continuity point.
Responsibilities:
- interpret user requests;
- decide which tool or agent to use;
- maintain memory and working rules;
- review important results;
- coordinate code changes;
- document decisions;
- protect security boundaries.
The coordinator represents the direction layer: it does not only execute tasks, it keeps the map of the system.
Coding Assistant
The coding assistant is the builder and reviewer role. Its job is to work on repositories, read code, implement changes, run checks, and explain results.
Responsibilities:
- modify code;
- create or improve technical documentation;
- run tests and linters;
- analyze errors;
- review diffs;
- produce implementations from scoped specifications.
In this project, the coding assistant works best when it receives structured context: AGENTS.md, TOOL.md, README files, installed skills, and small tasks.
Research Worker
The research worker is a specialized agent separated from the main process. It has two domains:
- structured research;
- entertainment and media workflows.
Responsibilities:
- process research missions;
- execute phase-based pipelines;
- use media tools when appropriate;
- generate reports;
- work in isolation so it does not block the main agent.
Relationship Between Agents
Learning Rule
The architecture does not try to make one agent smarter through prompting alone. It makes each agent more reliable by placing it inside a clear role with tools and contracts.