Specboard — Summary

Specboard is a web-based dashboard and CLI tool that monitors OpenSpec progress across multiple workspaces in real time. It renders all active features as a kanban-style board — swimlanes per feature, columns for Todo/In Progress/Done — and watches the filesystem for live updates. The tool is distributed as an npm package (@sflueckiger/specboard) with a dedicated specboard CLI binary that launches an HTTP server on port 3456. Users can browse proposal, specification, design, and plan documents; toggle Manual QA subtask completion; and open worktrees directly in Finder, VS Code, or Terminal. It integrates with Conductor to discover workspaces and supports URL-based routing for navigation persistence. Unlike seed frameworks that embed workflow logic or prompts, Specboard is purely an observability dashboard — it adds zero prompts or skills and performs no agent coordination.

differs_from_seeds: Closest in spirit to the UI surface of openspec (which ships a CLI and web dashboard), but Specboard is single-purpose: it does not orchestrate any agent actions and ships no prompts, skills, or hooks. Where openspec's openspecui is a full-stack companion with OPSX compose panels, PTY terminals, and workflow execution, Specboard is a read-mostly kanban monitor for multi-worktree Conductor deployments.

Specboard — Overview

Origin

Specboard (v1.1.2) is authored by Stefan Flueckiger (sflueckiger) and first published in early 2026. It emerged as a companion dashboard for teams using OpenSpec with Conductor-managed workspaces.

Philosophy

The README describes it as: "A web-based monitoring tool for tracking OpenSpec progress across Conductor workspaces." The core belief is that parallel AI-driven worktrees need a unified status surface — agents working in separate git worktrees on separate features need one place where a human can see the whole picture at a glance without reading raw Markdown files.

Manifesto-style statements (from README)

• "Live Kanban View: See all active features as swimlanes with tasks organized in Todo/In Progress/Done columns"
• "Real-time Updates: File changes are detected automatically via filesystem watching"
• "Multi-Repository Support: Switch between different repositories and worktrees"
• "Quick Actions: Open worktrees in Finder, VS Code, or Terminal with one click"

Target user

Teams using OpenSpec + Conductor who run multiple parallel AI agents/worktrees and want a single-pane-of-glass status dashboard. No AI or LLM interaction is involved at runtime.

Specboard — Architecture

Distribution

• npm package: @sflueckiger/specboard (v1.1.2)
• Published to npmjs.com

Install methods

bunx @sflueckiger/specboard          # run without installing (recommended)
npx @sflueckiger/specboard           # same via npx
npm install -g @sflueckiger/specboard  # global install
bun install -g @sflueckiger/specboard

CLI binary

Binary: specboard (defined in package.json#bin.specboard → ./cli.ts)

Usage:

specboard [path] [options]
  -p, --port    Port (default: 3456)
  -o, --open    Open browser after starting
  -h, --help
  -v, --version

Tech stack

• Runtime: Bun (v1.0+)
• Server: Custom Bun HTTP server (server.ts) with REST API and SSE for live updates
• Frontend: Static HTML/JS in public/ directory
• Filesystem watcher: Node.js fs.watch API
• Communication: Server-Sent Events (SSE) for push to browser

Required runtime

• Bun v1.0+
• Node.js ≥18.0.0 (fallback)
• OpenSpec-compatible directory structure

Directory structure

specboard/
├── cli.ts          # CLI entry point
├── server.ts       # Bun HTTP server + REST API + SSE
├── public/         # Static frontend assets
├── package.json
└── assets/

Target AI tools

None — Specboard is a passive observer that reads OpenSpec files; it does not invoke any AI model.

Expected workspace layout (from README)

workspaces/
  repo-name/
    feature-branch/
      openspec/
        changes/
          feature-name/
            tasks.md
            proposal.md
            design.md

Specboard — Components

Specboard ships no commands, skills, agents, hooks, or MCP servers. It is a single-purpose web dashboard with two runtime components:

CLI binary: specboard

• Entry point: cli.ts
• Parses CLI flags (--port, --open, --help, --version)
• Starts the Bun HTTP server

Server: server.ts

• REST API: Provides JSON endpoints for reading OpenSpec workspace data
• SSE endpoint: /events — pushes file-change events to connected browsers
• Static file server: Serves public/ directory
• Filesystem watcher: Uses fs.watch to detect changes in the watched root path
• Toggle endpoint: Allows toggling Manual QA subtask completion (writes back to tasks.md)

Key data types (from server.ts)

• Feature — name, path, worktree, tasks, hasProposal, hasDesign, specs[], hasPlan, isArchived
• TaskCard — id, title, subtasks[], status (todo|in_progress|done)
• Subtask — id, title, completed
• Repository — name, path, worktrees[]

Frontend (public/)

• Kanban board with swimlanes per feature
• Artifact viewer (collapsible sections for proposal/design/spec/plan docs)
• URL-based routing for navigation persistence
• Multi-repository switcher

No prompts, skills, hooks, or MCP servers

Specboard ships zero AI-facing primitives. It is purely a file-system reading and display tool.

Specboard — Uniqueness & Positioning

differs_from_seeds

Specboard has no analog among the 11 seed frameworks. The closest seed is openspec, which includes a web-based UI component (openspecui), but even that comparison breaks down: openspecui is a full workflow execution surface with OPSX compose panels, PTY terminals, and artifact generation, while Specboard is strictly a read-mostly kanban observer. No seed framework ships a dedicated observability dashboard for multi-worktree parallel AI development. Specboard occupies a unique niche: it assumes agents are already running in parallel worktrees and provides the human operator a single-pane-of-glass view without interfering with agent execution.

Positioning

• Category: OpenSpec ecosystem tooling — observability layer
• User: Human operator overseeing multiple parallel AI coding agents
• Integration: Layered on top of OpenSpec + Conductor, not a replacement for either
• Value: Makes the parallel-worktree development pattern legible to humans in real time

Observable failure modes

1. Stale reads: If Bun's fs.watch misses events (platform-dependent), the dashboard may lag behind actual file state
2. Path configuration: Specboard assumes a specific Conductor workspace layout; non-standard structures may not parse correctly
3. No auth: The dashboard is unauthenticated — any local user with network access to port 3456 can see and toggle tasks
4. tasks.md format sensitivity: The task parser supports multiple formats; malformed tasks.md may render incorrectly
5. Bun dependency: Requires Bun runtime, not standard Node.js-only environments

Cross-references

• Requires: OpenSpec (for the file structure it reads)
• Compatible with: Conductor (for workspace management)
• Related: spek (another OpenSpec viewer, but VS Code + read-only, no kanban)