Agents Overview

opencode-team-lead installs a complete multi-agent system into OpenCode. At the top sits team-lead, a pure orchestrator that never touches code directly. Everything flows through delegation: the team-lead plans, sub-agents execute, the review cluster validates.

Agent hierarchy — team-lead orchestrates everything; review-manager sub-orchestrates the review cluster.

Agents are split into two modes:

• mode: all — visible in the OpenCode agent list, accessible directly by the user or by the team-lead
• mode: subagent — invisible to the user, only reachable via task delegation from another agent

Agent Roster

Agent	Mode	Role
team-lead	all	Pure orchestrator. Plans, delegates, reviews, reports. Never touches code.
review-manager	subagent	Review orchestrator. Spawns reviewers in parallel, arbitrates verdicts.
requirements-reviewer	subagent	Verifies implementation matches original requirements.
code-reviewer	subagent	Evaluates logic, error handling, API design, maintainability.
security-reviewer	subagent	Identifies vulnerabilities across 7 threat categories.
bug-finder	subagent	Structured investigation. Forces root-cause before any fix.
brainstorm	all	Phase 0 discovery. Transforms vague ideas into structured product briefs.
harness	all	Encodes recurring patterns as permanent enforcement artifacts.
planning	all	Writes complex requests as exec-plans to disk.
gardener	all	Periodic maintenance. Fixes stale docs, detects code drift.
researcher	subagent	External knowledge retrieval. Searches docs, RFCs, APIs.

Agent Pages

• team-lead — the orchestrator at the center of everything
• Review Cluster — review-manager + requirements, code, and security reviewers
• Brainstorm — Phase 0 thinking partner for vague ideas
• Bug-Finder — structured investigation before any fix
• Harness — pattern encoder that makes recurring mistakes impossible
• Planning — turns complex requests into reviewable exec-plans
• Gardener — periodic hygiene and drift detection
• Researcher — external knowledge retrieval

How delegation flows

User
  └─► team-lead
        ├─► explore / general sub-agents  (implementation)
        ├─► review-manager               (reviews)
        │     ├─► requirements-reviewer
        │     ├─► code-reviewer
        │     └─► security-reviewer
        ├─► bug-finder                   (when debugging)
        ├─► brainstorm                   (when intent is unclear at vision level)
        ├─► planning                     (when request is ambiguous on structure)
        └─► researcher                   (when external knowledge needed)

Brainstorm — direct or automatic

You can invoke brainstorm directly when you have a vague idea, or let the team-lead invoke it automatically when your request is unclear at the vision level. Either path produces a structured product brief at docs/briefs/{project-name}.md that feeds directly into planning.

Harness runs after patterns emerge

The harness agent is triggered after recurring patterns are detected — either by the user, by the team-lead post-feature, or automatically by the Gardener. It encodes the pattern as a mechanical check so it never needs to be enforced manually again.