Sub-Agents — Delegate Large Work Without Losing Main Context

Why Sub-Agents Are Helpful

Sub-agents let you keep the main conversation focused while delegating a larger unit of work to a specialist.

This is most useful when:

• one task needs deep focus (migration, refactor, incident triage)
• you want independent reasoning threads in parallel
• you need the main chat to stay concise and decision-focused

In practice, sub-agents reduce context clutter and improve throughput on bigger tasks.

Claude Code Sub-Agents

Claude Code supports built-in and custom sub-agents. Custom definitions can be project-scoped or personal:

• project: .claude/agents/
• personal: ~/.claude/agents/

Helpful capabilities:

• route specialized tasks to purpose-built agent roles
• invoke by name for explicit delegation
• run multiple sub-agents in parallel for independent workstreams

OpenCode Sub-Agents

OpenCode also supports custom agents, with common locations:

• project: .opencode/agents/
• global: ~/.config/opencode/agents/

This gives a similar delegation model: keep the main chat as coordinator, dispatch substantial tasks to specialized agents, then merge results.

Practical Delegation Pattern

1. Keep the main chat for goals, constraints, and final decisions
2. Delegate one clear deliverable per sub-agent
3. Require each sub-agent to return three items: what changed, what was verified, and open risks/questions
4. Merge results in main chat and decide next step

Good First Use Cases

• “Refactor module X and keep API compatibility”
• “Add tests for changed files only”
• “Audit this diff for security regressions”
• “Draft migration plan plus rollback checks”

Common Mistakes

Vague delegation. “Look at this” is too broad; assign a concrete output.

No integration contract. Require output shape so results are mergeable.

Over-fragmentation. Too many sub-agents on tiny tasks adds overhead instead of speed.

References

• Claude Code: Sub-agents (official)
• OpenCode: Agents guide (official)
• External References