README.md
June 29, 2026 · View on GitHub
A full AI development team — researcher, PM, architect, dev, QA, and reviewer — as native Codex agents, orchestrated through a configurable pipeline.
Quick Start
curl -fsSL https://raw.githubusercontent.com/yimwoo/codex-agenteam/main/install.sh | bash
Restart Codex, go to Plugins > Local Plugins, install AgenTeam, then:
@ATeam build my team
If the role picker does not show @Architect, @Dev, and the others right
away, confirm the project has .codex/agents/*.toml, then open a new thread
or restart Codex so it reloads workspace agents.
Your Team
| Role | Mention | Job |
|---|---|---|
| Researcher | @Researcher | Investigates docs, trends, prior art |
| PM | @Pm | Prioritizes work, writes specs |
| Architect | @Architect | Designs systems, identifies risks |
| Dev | @Dev | Writes production code |
| Qa | @Qa | Writes tests, catches regressions |
| Reviewer | @Reviewer | Reviews for correctness and security |
@Architect review this API design
@Dev fix the race condition in src/queue.py
@ATeam add user authentication
Documentation
- Setup & Installation — prerequisites, install, update
- Configuration — roles, profiles, model routing, two-layer config
- Pipeline & Profiles — stages, gates, verification, resume, CI repair
- CLI Reference — all skills and runtime commands
- HOTL Integration — structured execution with the HOTL plugin
- Benchmarking — reproducible strategy comparisons from run evidence
Current Capabilities
- Run control: configurable stages, role routing, isolation, verification, retries, rework, gates, and resume.
- Compatibility and observability:
doctor,status,trace, standups, reports, and portable evidence bundles. - Portable planning: deterministic workspace-agent export plus evidence-backed benchmark conversion and reporting.
- Structured handoffs: opt-in schema-validated role results with SHA-256 provenance, downstream prompt reuse, and trace/evidence metrics.
- Governed delivery: optional decision logs, tripwires, scope audits, and approval evidence for larger or longer-lived work.
These capabilities are additive. Quick fixes and standard pipelines remain lightweight unless you enable structured handoffs or governed delivery.
Contributing
Found a bug or have a role idea? Open an issue or submit a PR.
License
MIT © Yiming Wu