Principled agentic coding: plan → build → review → learn workflow with builder and reviewer subagents
Use this agent when the plan is confirmed and work needs to be done. The builder implements the next unfinished task from .trio/plan.md, then writes .trio/handoff.md so the reviewer can evaluate the work in a separate context.
Use this agent when .trio/handoff.md exists and the builder's work needs evaluation. The reviewer checks the work against the plan and verification criteria in a separate context, then writes .trio/review.md with status, findings, and follow-up.
A minimal, all natural language bootstrap for brand new or existing repos of any kind that want to follow principled agentic coding. Zero code.
Once upon a time, humans hand-wired computers to program them. Thankfully, someone invented silicon chips coding agents.
agent trio is not a swarm framework. It does not optimize for agent count or lines of code output. It is for introducing just the right amount of friction. You should spend tokens to maximize judgement and quality, not output or wasteful rework. agent trio helps you do that. See the intro blog post for more.
How is this different from /review? /review is a one-shot critique of git state; agent-trio holds out .trio/criteria.md before building, so the review phase checks the implementation against criteria set before it existed.
How is this different from openai/codex-plugin-cc? That plugin ships cross-provider adversarial review as a skill; agent-trio is the plan → build → review → learn loop around it, and the reviewer can invoke /codex:adversarial-review as one of its inputs.
Recommendations. I recommend you select a more efficient (but not deliberately cheap) builder model, and a similar-or-stronger reviewer model. Keep your expensive tokens for reasoning and judgement. You could for example use Codex + GPT-5.4 (xhigh) as the planner and reviewer, and GPT-5.4 (med) for the builder. Claude Code can spawn Opus reviewer and Sonnet builder subagents. Or go rainbow providers — you just need to configure your agents accordingly.
Agent-trio installs as a plugin that injects the workflow contract and agents into your coding agent session.
Claude Code:
Register the plugin marketplace, then install:
/plugin marketplace add haplesshero13/agent-trio
/plugin install agent-trio@trio-marketplace
Or add to your project's .claude/settings.json:
{
"plugins": ["agent-trio@git+https://github.com/haplesshero13/agent-trio.git"]
}
Codex:
codex plugin marketplace add haplesshero13/agent-trio
Then open Codex, run /plugins, select Agent Trio, and install it. For local
development from this checkout, run codex plugin marketplace add ..
The skill operates in consult, plan, or execute mode depending on what you ask — it will not create planning files for read-only questions or simple guidance. Builder and reviewer agents are available when Codex runs inside a checkout; without them, the head instance acts locally and still honors the full artifact contract.
See .codex/INSTALL.md for full details and examples.
OpenCode:
If agent-trio is available from npm:
Add agent-trio to the plugin array in your opencode.json:
{
"plugin": ["agent-trio"]
}
For local development or git-checkout installation, clone the repo and add a
global shim under ~/.config/opencode/plugins/; see .opencode/INSTALL.md.
Restart OpenCode. The plugin registers the repo-root skills/ directory and
the .opencode/agents/ wrappers, so using-agent-trio, @builder, and
@reviewer become available.
See .opencode/INSTALL.md for details.
Once agent-trio is installed, run the setup flow to pick builder/reviewer models and peer-review preferences without hand-editing config:
/agent-trio:setupsetup/agent-trio-setup.md and walk you through the same choices.The setup flow only writes to files it tells you about first; see setup/agent-trio-setup.md for the full list of files it may create or modify per platform.
AGENTS.md: developer guide — repo structure, how to make changes, validation checklistskills/using-agent-trio/SKILL.md: the workflow contract (imperative, with first-session bootstrap and delegation)agents/builder.md, agents/reviewer.md: the builder/reviewer agent definitions.trio/learnings.md: the checked-in lessons that still change how the next loop runs.trio/plan.md, .trio/handoff.md, .trio/review.md: the gitignored agent coordination files.claude/agents/, .codex/agents/: provider wrappers around the rolesplugins/agent-trio/.codex-plugin/, .agents/plugins/: Codex plugin manifest and marketplace metadata.opencode/: OpenCode plugin — plugins/agent-trio.js, package.json, INSTALL.md.claude-plugin/: plugin manifest for marketplace distributioncommands/setup.md: /agent-trio:setup slash command for Claude Codesetup/agent-trio-setup.md: platform-neutral interactive setup instructionsUses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
npx claudepluginhub haplesshero13/agent-trio --plugin agent-trioHarness-native ECC operator layer - 67 agents, 271 skills, 92 legacy command shims, reusable hooks, rules, selective install profiles, and production-ready workflows for Claude Code, Codex, OpenCode, Cursor, and related agent harnesses
Comprehensive skill pack with 66 specialized skills for full-stack developers: 12 language experts (Python, TypeScript, Go, Rust, C++, Swift, Kotlin, C#, PHP, Java, SQL, JavaScript), 10 backend frameworks, 6 frontend/mobile, plus infrastructure, DevOps, security, and testing. Features progressive disclosure architecture for 50% faster loading.
Access thousands of AI prompts and skills directly in your AI coding assistant. Search prompts, discover skills, save your own, and improve prompts with AI.
Upstash Context7 MCP server for up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your LLM context.
Complete developer toolkit for Claude Code
Develop, test, build, and deploy Godot 4.x games with Claude Code. Includes GdUnit4 testing, web/desktop exports, CI/CD pipelines, and deployment to Vercel/GitHub Pages/itch.io.