Skip to content
B2FPublic

About

No description, website, or topics provided.

Resources

Stars

7 stars

Watchers

0 watching

Forks

Latest commit

 

History

33 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

OMAD Workbench

OMAD — OpenCode Method for Agentic Development

An independent, opinionated opencode configuration that can serve as a default starting point for agentic development. It includes a practical, native-opencode adaptation of selected BMAD workflows.

OMAD coordinates specialized agents, MCP servers, research, planning, implementation, review, and durable workflow artifacts without depending on BMAD's installer, resolver, or configuration runtime.

OMAD is not affiliated with or endorsed by BMad Code, LLC or the opencode team.

If you use opencode for AI-assisted development, this repo gives you a reproducible starting point for:

  • managing providers and permissions, and
  • using specialized subagents for writing, review, architecture, research, and product work.

Native workflow commands

The workspace exposes reusable workflows as opencode commands:

Full lifecycle entrypoints:

  • /omad-align — frame the problem, users, outcomes, and constraints
  • /omad-discover — research assumptions and generate options
  • /omad-shape — challenge the concept and define requirements
  • /omad-design — define UX, architecture, contracts, and NFRs
  • /omad-decompose — create traceable epics, stories, and acceptance criteria
  • /omad-plan — sequence work and refresh sprint tracking
  • /omad-build — implement one approved story with tests
  • /omad-verify — review, test, checkpoint, and capture evidence
  • /omad-learn — capture learning and choose the next decision

Lower-level workflows:

  • /checkpoint — walk through a change from intent to approval
  • /review — review code, diffs, specs, plans, and documents through focused lenses
  • /research — draft, run, process, refresh, or deepen decision-grade research
  • /prfaq — run the Working Backwards PRFAQ challenge
  • /create-epics-and-stories — decompose requirements into implementation-ready stories
  • /sprint-planning — generate sprint tracking from epics
  • /sprint-status — summarize progress, risks, and the next action
  • /distill — create a source-backed, token-efficient document distillate
  • /party-mode — run a multi-agent roundtable

Full inventories and workflow guidance are in docs/agents-skills-commands.md.

The recommended BMAD-inspired command sequence and artifact contract are in docs/bmad-inspired-workflow.md.

The most distinctive native workflows are:

  • Review lenses: combine adversarial, edge-case, code, editorial, and fact-checking perspectives without forcing every review through every lens.
  • Research runs: separate decision framing, evidence gathering, source digests, synthesis, confidence, and open questions. Research artifacts can be resumed from research/<slug>/.
  • BMAD-derived planning: retain useful PRFAQ, epics, sprint, checkpoint, and distillation procedures while using native opencode project context and agents.
  • Scoped artifacts: keep project-wide documentation in the active project and independent feature work under docs/features/<slug>/ with a manifest and portfolio index.

Agent-neutral identities

OMAD improves on the stock BMAD persona model by keeping skills role-focused instead of binding them to branded personal names. Architect, analyst, developer, designer, writer, and other skills are invoked by capability, not by remembering a persona roster. This makes prompts more portable, reduces unnecessary anthropomorphism, and keeps multi-agent party-mode examples clear without changing the underlying workflow responsibilities.

Features

  • 26 BMAD-derived custom agent prompt files in agents/, alongside native OMAD roles
  • Permission defaults configured for common tools (read, glob, bash, etc.)
  • Preconfigured Brave Search MCP + subagent
  • Preconfigured DuckDuckGo Search MCP server
  • Custom DuckDuckGo research subagent (agents/duckduckgo-search.md)
  • Six research workflow subagents adapted from web-forager skill patterns:
    • agents/web-research.md
    • agents/fact-check.md
    • agents/news-monitor.md
    • agents/competitive-intel.md
    • agents/tech-advisor.md
  • agents/scholarly-fact-check.md
  • Preconfigured LM Studio model sample

MCP capabilities, package dependencies, installation commands, environment variables, timeout policy, and rate-limiter behavior are documented in docs/mcp-capabilities.md.

YouTube research subagent

  • agents/youtube-research.md: combines official YouTube Data API metadata (youtube MCP) with transcript analysis (youtube-transcript MCP).

Research subagent routing policy

These research subagents share a common policy:

  • Brave is default only for fact-critical or quality-critical tasks.
  • DuckDuckGo is default for exploratory scans.
  • Fetch order is duckduckgo-search_fetch_content first, webfetch fallback.
  • Every non-trivial factual claim should include at least one source link.

Skill-specific intent:

  • web-research: balanced synthesis
  • fact-check: direct claim verification
  • news-monitor: fast brief news updates
  • competitive-intel: comparison and market deltas
  • tech-advisor: technical option recommendations
  • scholarly-fact-check: stronger scientific/medical evidence checks

Project structure

.
├── agents/             # Custom subagent definitions (Markdown + frontmatter)
├── commands/           # Native opencode workflow commands
├── skills/             # Native opencode workflows and supporting assets
├── docs/               # Agent, workflow, and MCP capability documentation
├── opencode.json       # Workspace config (permissions/providers/models)
├── tools/               # Local MCP reliability helpers

OMAD commands resolve project artifacts from the active opencode project or worktree, never from this configuration repository. For independent work, invoke a phase with a stable feature slug, for example /omad-align feature=checkout-redesign. Feature artifacts live under docs/features/checkout-redesign/, while lifecycle workflows maintain the project feature index and report evidence-based progress.

License

This project is released under the MIT License.

Some workflow and agent material was adapted from BMad Method, which is also MIT licensed. The BMad copyright and attribution requirements apply to copied or adapted portions. BMad names, logos, and branding are trademarks of BMad Code, LLC and are not licensed by MIT.

OMAD is an independent project. “BMad” and “opencode” are used only to describe lineage, compatibility, or implementation context; neither organization endorses this project.

About

No description, website, or topics provided.

Resources

Stars

7 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages