AGENTS.md

AGENTS.md

What This Repo Is

A collection of OpenCode agent configuration files. No code, no build, no tests — only .md files that define agent behavior via YAML frontmatter + prompt body, plus an opencode.jsonc for global settings.

Structure

opencode-agents/
├── opencode.jsonc          # Global config (permissions, default agent)
├── agents/
│   ├── Zero.md             # Primary agent (mimo-v2.5, multimodal)
│   ├── Erribaba.md         # Primary agent (mimo-v2.5-pro)
│   └── *.md                # 25 subagent definitions
├── AGENTS.md               # This file
├── CHANGELOG.md            # Version history
├── README.md               # Chinese
├── README.en.md            # English
└── LICENSE                 # MIT

Agent File Format

Every file in agents/ uses this structure:

---
description: <Chinese description — this is what OpenCode uses to decide when to invoke the agent>
mode: primary | subagent
model: <provider/model-name>
temperature: <0.0-1.0>
tools:            # optional — omit keys to allow all
  write: false
  edit: false
  bash: false
---

<System prompt in English or Chinese>

Key Conventions

• description is always in Chinese — this is the trigger text OpenCode matches against. Keep it consistent with existing files.
• mode: primary means the agent can be a top-level session. Only Zero.md (mimo-v2.5) and Erribaba.md (mimo-v2.5-pro) are primary agents.
• mode: subagent means the agent is only invoked via delegation from a primary agent.
• Tool restrictions in frontmatter (write: false, edit: false, bash: false) make an agent read-only. Most review/audit subagents are read-only; implementation agents have write access.
• Model naming: provider prefix matters — use opencode-go/mimo-v2.5, opencode-go/deepseek-v4-pro, etc.
• Filename = agent name used in delegation (e.g., reviewer.md → delegate as reviewer).
• Primary agents must list all subagents in their body — when adding a new subagent, update both Zero.md and Erribaba.md.

Model Assignments

Model	Used By	Reasoning
opencode-go/mimo-v2.5	Zero, frontend-reviewer, git-assistant, software-engineer, vision-dev	Multimodal, fast, general-purpose
opencode-go/mimo-v2.5-pro	Erribaba, api-designer, code-generator, devops, e2e-tester, refactorer, test-writer	Best coding capability
opencode-go/deepseek-v4-pro	db-engineer, debugger, migration, perf-optimizer, reviewer, security-auditor	Strong at analysis and reasoning
opencode-go/kimi-k2.6	frontend-dev, ui-designer	Good at frontend/creative tasks
opencode-go/qwen3.6-plus	doc-writer, explore, project-manager, research	Good at writing, planning, and exploration
opencode-go/glm-5.1	architect	Architecture-level reasoning
opencode-go/minimax-m2.7	executor, validator	Execution and validation focus

Temperature Guide

Temperature controls the randomness/creativity of the model's output. Choose based on the task:

Temperature	Use Case	Agents
0.1	Precise, deterministic output — code generation, database queries, command execution	code-generator, db-engineer, executor
0.2	Mostly deterministic with slight variation — debugging, review, security audit, validation	debugger, explore, frontend-reviewer, migration, perf-optimizer, reviewer, security-auditor, validator
0.3	Balanced — general implementation, refactoring, testing, DevOps, E2E, UI, vision	Erribaba, Zero, api-designer, devops, e2e-tester, frontend-dev, refactorer, software-engineer, test-writer, vision-dev
0.4	Slightly more creative — architecture design, Git workflow	architect, git-assistant
0.5	More creative — documentation, research, project management, UI design	doc-writer, project-manager, research, ui-designer

Guidelines:

• Code generation and execution tasks → lower temperature (0.1-0.2)
• Review and analysis tasks → lower temperature (0.2)
• Implementation tasks → balanced temperature (0.3)
• Design and planning tasks → slightly higher temperature (0.4)
• Writing and research tasks → higher temperature (0.5)

Prompt Writing Guidelines

DO

• Use ## sections for clear structure
• Include "What You Do NOT Do" section for delegation directions
• Include "Limitations" section for capability boundaries (only when different from delegation)
• Add "Quality Checklist" with actionable items
• Provide concrete output format examples
• Keep prompts focused — one role per agent

DON'T

• Hardcode model names in prompts (they're in the frontmatter)
• Duplicate the same info in multiple sections
• Mix Chinese and English in the same sentence
• Make prompts longer than necessary — concise is better
• Add build/CI/tooling — this repo has none and needs none

Section Order (subagents)

1. Role statement (one line)
2. "What You Do" or "Core Expertise"
3. "What You Do NOT Do" (delegation directions)
4. Domain-specific content (methodologies, patterns, etc.)
5. "Output Format"
6. "Limitations" (capability boundaries, if different from delegation)
7. "Interaction Style"
8. "Quality Checklist"

opencode.jsonc

The global config file controls:

{
  "$schema": "https://opencode.ai/config.json",
  "default_agent": "Zero",  // or "Erribaba"
  "permission": {
    "read": "allow",
    "edit": "allow",
    "glob": "allow",
    "grep": "allow",
    "bash": "allow",
    "webfetch": "allow",
    "websearch": "allow",
    "task": "allow",       // subagent delegation
    "skill": "allow",
    "todowrite": "allow",
    "question": "allow",
    "external_directory": "allow"
  }
}

When Adding or Editing Agents

1. Match the frontmatter field order: description, mode, model, temperature, then optional tools.
2. Write description in Chinese, matching the style of existing files (role + trigger scenarios).
3. Keep system prompts concise and structured with ## sections.
4. Primary agents must list all available subagents in their body — update both Zero.md and Erribaba.md if adding a new subagent.
5. Do not add build/CI/tooling — this repo has none and needs none.
6. Update CHANGELOG.md with a summary of changes.

Language

• Frontmatter description: Chinese
• System prompt body: English (most files) or Chinese (acceptable)
• README: Chinese primary, English secondary