opencode-pair

OpenCode harness with a five-agent setup: one primary, one general subagent, one ideation subagent, one frontend design subagent, and one validation-focused subagent.

What it does

• MrRobot is the primary agent. He routes work and answers plainly.
• Eliot is the general subagent. He handles implementation, refactors, repo exploration, and other scoped task work.
• Tyrell is the ideation subagent. It handles brainstorming, naming, UX direction, product ideas, and open-ended exploratory packets.
• Claude is the frontend design subagent. He is the default implementation lane for pages, components, styling, layout, and visual polish unless the user explicitly asks for review-only output or no file edits, on openai/gpt-5.5-fast low with bundled Impeccable plus stack-aware taste/redesign skills.
• Implementation packets should be edited directly in the repo by the assigned subagent; research, review, and ideation packets should return findings without edits unless edits are explicitly requested.
• Ongoing subagent work should continue with the same task_id by default when the lane and workstream still match.
• Turing is the validation-focused subagent.
• No plan/execute mode or harness slash-command flow.
• No cross-session memory, pattern learning, or observation logs. Workflow-local task tracking only keeps recent task_id hints for continuation inside related sessions.
• Comment guard blocks suspicious AI-style comments before file writes and surfaces anything that still slips through.

Agents

| Agent | Character | Role | Model | Variant | | ----- | --------- | ---- | ----- | ------- | | mrrobot | MrRobot | Primary agent — routes, synthesizes, answers | openai/gpt-5.5-fast | medium | | eliot | Eliot | General-purpose subagent | openai/gpt-5.5-fast | low | | tyrell | Tyrell | Ideation-focused subagent | openai/gpt-5.5-fast | low | | claude | Claude | Frontend design subagent | openai/gpt-5.5-fast | low | | turing | Turing | Validation-focused review and verification | openai/gpt-5.5-fast | xhigh |

All harness agents use fast model IDs. MrRobot uses openai/gpt-5.5-fast medium, implementation/ideation/frontend subagents use openai/gpt-5.5-fast low, and Turing uses openai/gpt-5.5-fast xhigh.

MCP Servers

| MCP | What | API Key | | --- | ---- | ------- | | context7 | Library and framework documentation | No | | grep_app | GitHub code search across public repos | No | | searxng | Web search via self-hosted SearXNG | No | | web-agent-mcp | Browser testing and automation | No | | pg-mcp | PostgreSQL read-only client | No | | ssh-mcp | Remote command execution on configured SSH hosts | No | | openai-image-gen-mcp | Image generation via Codex auth store | No | | mariadb | MariaDB client | No |

Shared managed MCP roots stay under ~/.config/{mcp_name}.

All five agents receive the same enabled MCP set and the same default full tool access. The harness does not add per-agent MCP or tool restrictions.

Prerequisites

• Docker — required for SearXNG (auto-provisioned by installer)

Quick start

bunx opencode-pair install

The installer will:

1. Wire agents and MCPs into OpenCode config
2. Install shell strategy instructions
3. Vendor pg-mcp, ssh-mcp, web-agent-mcp, openai-image-gen-mcp, and bundled skills (including Impeccable plus taste/redesign skills for Claude)
4. Install dependencies inside each shared managed MCP root
5. Auto-provision SearXNG Docker container (--restart unless-stopped, 127.0.0.1:8099:8080)
6. Enable JSON format in SearXNG settings

From source:

git clone https://github.com/cemalturkcan/opencode-pair.git
cd opencode-pair
bun install && bun run build && bun link
opencode-pair install

Commands

opencode-pair install        # wire into OpenCode config
opencode-pair fresh-install  # rebuild harness files, keep user config
opencode-pair uninstall      # remove harness wiring
opencode-pair init           # create project-local config
opencode-pair print-config   # inspect generated config

Config

Merges from two layers (project wins):

• ~/.config/opencode/opencode-pair.jsonc — user-level
• <project>/.opencode/opencode-pair.jsonc — project-level

Create project config:

opencode-pair init

workflow.compact_subagent_context defaults to true. It shortens the project-fact line injected into subagent sessions; set it to false to keep the longer human-readable format.

Hooks

| Hook | What it does | | ---- | ------------ | | chat.message | Inject project docs, WSL notes, and active subagent task IDs for MrRobot; inject compact project facts for Eliot, Tyrell, Claude, and Turing | | tool.execute.before | Block suspicious AI-style comments before writes, enforce git-push build gate, auto-transform Node commands on WSL | | tool.execute.after | Surface suspicious comments that still remain after a write; capture subagent task IDs for continuation hints | | session.deleted | Clear ephemeral runtime state |

hooks.task_tracking defaults to true and controls task-id capture plus primary-session continuation hints.

Architecture

src/
├── prompts/
│   ├── mcp-access.ts    # Enabled MCP list and prompt guidance
│   ├── shared.ts        # Shared prompt rules and response style
│   ├── workers.ts       # Eliot, Tyrell, Claude, and Turing prompt builders
│   └── coordinator.ts   # MrRobot prompt builder plus routing rules
├── agents.ts            # Agent definitions (models and prompts)
├── mcp.ts               # MCP server registration
├── hooks/               # Runtime hooks (comment guard, WSL, cleanup)
├── config.ts            # Config schema + loading
├── installer.ts         # CLI installer
└── index.ts             # Plugin entry point