Concept

Agent Architecture

EvolC agents are autonomous AI workers that read context from the filesystem and execute real business tasks.

How Agents Work

Unlike simple chatbots, EvolC agents are full autonomous workers. The execution flow is:

Agent reads AGENTS.md for business context
Agent receives a task (from UI or schedule)
Agent uses tools (file read/write, web search, APIs) to execute
Agent writes output to the filesystem
Status updates are reflected in the UI grid

Agent Definition Files

Agents are defined as markdown files in .claude/agents/ or .agents/ directories. Each file contains:

# Content Writer Agent

## Role
Create blog posts and articles for marketing.

## Tasks
- Write 2 blog posts per week
- Follow brand voice guidelines
- Output articles to ../content/blog/

## Success Criteria
- 1500-2500 words per post
- SEO score 80+
- Published within 24 hours

Context Hierarchy

Agents inherit context from multiple levels:

/business/
├── AGENTS.md              ← Company-wide context
├── marketing/
│   ├── AGENTS.md          ← Marketing context
│   └── .agents/
│       ├── content-writer.md  ← Agent-specific
│       └── seo-optimizer.md   ← Agent-specific
└── sales/
    ├── AGENTS.md          ← Sales context
    └── .agents/
        └── outreach.md    ← Agent-specific

The content-writer agent reads context from all three levels: company-wide, marketing department, and its own agent file.

Agent Coordination

Agents coordinate through the filesystem:

Shared files — Agents read and write to shared directories
Status tracking — Agent status stored in frontmatter
Task queues — Markdown checklists as task lists
Reports — Agents write outputs to reports/ directories

Plugins as Packaged Agents

The EvolC plugin system packages agents into installable modules. A plugin contains agent definitions, slash commands, hooks, and MCP server integrations — everything needed for a complete business function.