Agent Management

clnode installs two default agents and provides the /clnode-agents skill for discovery and custom agent creation.

Default Agents

Agent	Model	Role
clnode-reviewer	opus	Code review — quality, security, pattern consistency
clnode-curator	opus	Knowledge curation — audits memories, cross-pollinates learnings

/clnode-agents Skill

Run /clnode-agents in Claude Code to:

Phase 1: Discovery

Scans your .claude/ directory and shows a summary table of:

Installed agents (name, model, description, skills)
Installed skills
Installed rules

Phase 2: Create New Agent

Interactive wizard that asks for:

Name (kebab-case, e.g., api-tester)
Role (Implementer / Reviewer / Architect / Specialist)
Model (sonnet / opus / haiku)

Then generates .claude/agents/{name}.md with:

Proper YAML frontmatter
Appropriate tools based on role
compress-output skill (or compress-review for reviewers)
“Before Returning” compression section

Phase 3: Update team.md

If .claude/rules/team.md exists, automatically rebuilds the Team Structure section based on all installed agents.

Model Recommendations

Role	Model	Reason
Reviewer / Architect / Curator	opus	Deep reasoning needed
Implementer (backend/frontend/CLI)	sonnet	Fast, capable code generation
Simple / mechanical tasks	haiku	Cost-efficient

Agent File Structure

Agents are markdown files in .claude/agents/ with YAML frontmatter:

---
name: my-agent
description: What this agent does
tools: Read, Edit, Write, Bash, Grep, Glob
model: sonnet
memory: project
skills:
  - compress-output
---

# My Agent

Role description and responsibilities...

Key frontmatter fields:

tools — Which Claude Code tools the agent can use
model — Which Claude model to use (opus/sonnet/haiku)
memory: project — Enables persistent memory in .claude/agent-memory/
skills — Skills preloaded into the agent’s context