Ready-to-install multi-agent teams for OpenCode. Pick a team, install it into a project, and start with @team_lead_agent.

中文 | Quick Start | PPT Demo | Teams | What You Get

Open Agent Team is a library of reusable multi-agent teams. Each team packages role prompts, shared quality protocols, delivery standards, and optional skills for one complete workflow.

The goal is simple: make advanced multi-agent collaboration usable in one command.

choose a team -> install it into your project -> open OpenCode -> ask @team_lead_agent

• One-command installation: install a complete team into any OpenCode project.
• Complete deliverables: teams are designed to produce finished output packages, not loose chat notes.
• Reliable workflows: shared protocols define quality gates, source tracing, review loops, and final delivery checks.
• Reusable team library: use built-in teams or register your own local team directory.
• OpenCode-native layout: installs agents, protocols, skills, AGENTS.md, and opencode.json into .opencode/.

From this repository root:

python -m pip install -e .
agent_team list

Install a team into your project:

agent_team install --name ppt_writer_team --path /path/to/your/project

Open that project with OpenCode and ask:

@team_lead_agent Create a 15-slide investor deck from the materials in ./materials.

The installer creates:

/path/to/your/project/.opencode/
  AGENTS.md
  opencode.json
  agents/
  protocols/
  skills/

ppt_writer_team turns source materials, templates, notes, spreadsheets, PDFs, and public research into a complete PowerPoint package:

• delivery/final_deck.pptx: editable .pptx, not a static image deck.
• delivery/speaker_notes.md: presenter notes and talk track.
• delivery/source_trace.md: source trace for claims and data.
• deck_spec/deck_spec.json: structured deck plan for review and regeneration.

The team can also extract visual style from an existing PPT template while avoiding unrelated template content reuse.

▶ Click the image to watch the ppt_writer_team demo

Two complete input/output examples are included under examples/ppt_writer_examples:

These examples show the intended end state: a polished editable PPT, a readable talk track, an executive summary, and source-backed trace files that make the deck easier to review and trust.

paper_survey_team helps with research direction surveys, daily paper radar, and single-paper deep reads:

• ranked paper lists with why each paper matters;
• clear metadata, authors, institutions, years, links, and reading priority;
• method lineage, benchmark evolution, key labs, representative people, and unresolved problems;
• single-paper reports covering background, contribution, method, experiments, limitations, and follow-up reading.

The output is built for scanning, filtering, and deciding what to read next.

Engineering teams produce practical artifacts:

• code review findings ordered by severity;
• test plans, unit tests, integration tests, and edge cases;
• bug triage reports with reproduction plans, root-cause hypotheses, fix options, and verification plans;
• architecture maps, ADR-ready design notes, API/data model plans, and implementation roadmaps.

List teams at any time:

agent_team list
agent_team list --verbose
agent_team list --json

Install into the current directory:

agent_team install --name code_review_team

Preview without writing files:

agent_team install --name code_review_team --path /path/to/project --dry-run

Clean generated OpenCode team files before reinstalling:

agent_team install --name code_review_team --path /path/to/project --clean

--clean removes generated .opencode/agents, .opencode/protocols, and .opencode/skills. It keeps other files in .opencode.

Most teams follow this layout:

teams/<team_name>/
  README.md
  agents/
    team_lead_agent.md
    *_agent.md
  protocols/
    quality_protocol.md
    delivery_protocol.md
    skill_registry.md
  skills/
    <skill_name>/
      SKILL.md

During installation:

agents/*.md      -> .opencode/agents/
protocols/*.md   -> .opencode/protocols/
skills/*         -> .opencode/skills/

The installer also writes .opencode/AGENTS.md and merges .opencode/opencode.json so OpenCode loads the team guidance automatically.

Create a team directory:

my_team/
  README.md
  agents/team_lead_agent.md
  protocols/quality_protocol.md
  protocols/delivery_protocol.md
  protocols/skill_registry.md
  skills/

Register it:

agent_team add --name my_team --path /path/to/my_team
agent_team install --name my_team --path /path/to/project

Update or remove it later:

agent_team update --name my_team --path /new/path/to/my_team
agent_team remove --name my_team

• Teams use a team_lead_agent as the default coordinator.
• Protocols define mandatory quality and delivery rules.
• Skills provide task-specific methods and scripts.
• Deliverables are named and structured so they can be reviewed, regenerated, or handed off.
• Source-heavy teams emphasize traceability and evidence cards instead of unsupported claims.

This project is released under the MIT License. See LICENSE.

Some bundled skills are adapted from third-party open-source repositories and remain subject to their original licenses. See NOTICE and THIRD_PARTY_LICENSES for attribution and bundled license texts.