Quick start

Get super-backlog installed and running in a few commands.

Install

Use the official installer for your platform. Both check Node.js and npm, install super-backlog, and run sbl init without relying on npx.

Windows (PowerShell) — works even when npx is blocked by Execution Policy:

irm https://raw.githubusercontent.com/adam-s-k-i/super-backlog/master/scripts/install.ps1 | iex

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/adam-s-k-i/super-backlog/master/scripts/install.sh | bash

If you already have Node and npm and prefer a manual install, run sbl init directly without npx:

# Global install
npm install -g super-backlog
sbl init
sbl init --models

# Local install
npm install super-backlog
node ./node_modules/super-backlog/dist/bin.js init

init is idempotent — run it again any time to upgrade injected files to the latest kit version.

Add --models to also enable the optional model router.

Daily commands

Command	What it does
sbl init	Wire Backlog.md, Superpowers skills, npm scripts, and hooks into the current project.
sbl init --models	Same as above, plus the optional model router.
npm run board	Open the Backlog.md kanban board.
npm run tasks	List Backlog tasks in the terminal.
npm run browser	Open the Backlog browser UI.
npm run dashboard	Start the live Project Dashboard server on http://localhost:6428.
sbl dashboard --port 8080	Start the dashboard server on a custom port.
sbl phase TASK-1	Show where the task stands in the pipeline (phase/spec → plan → impl → verify).
sbl phase TASK-1 plan	Advance the phase after the gate is passed (done clears the label).
sbl update	First self-updates a globally installed CLI to the latest npm version and re-runs itself (opt out with --no-self or SBL_SKIP_UPDATE_CHECK), then refreshes all injected files and prints harness/plugin versions.
sbl doctor	Check Node, PowerShell policy, the backlog CLI, and phase-label hygiene.
sbl uninstall	Remove everything super-backlog owns; keep your backlog/ data.
sbl uninstall --with-backlog	Remove everything, including task data.

Model router (opt-in)

The model router routes cheap models to simple agents and keeps your main model for complex work.

Command	What it does
sbl init --models	Install the router and harness adapters.
sbl models enable	Turn routing on.
sbl models disable	Turn routing off.
sbl models show	Show the current router config.
sbl models discover	Discover available OpenCode models and rank them into tiers.

When enabled:

• OpenCode uses a chat.params plugin to rewrite the model for sbl-worker (workhorse) and sbl-worker-cheap / explore (budget) agents.
• Claude Code uses a SessionStart hook to update agent model: placeholders based on your current main model.
• The dashboard server exposes /api/models (with an installed flag) plus /api/models/enable, /api/models/disable, and /api/models/discover.
• The Board & Quick Actions section has a Model Router button that opens a modal: it shows the installed/enabled state, toggles the router, runs discovery with the resolved workhorse/budget tiers, and offers a copyable sbl init --models when the router is not installed yet.

The router is removed cleanly by sbl uninstall.

Project dashboard

sbl dashboard

sbl dashboard (alias: db) starts a local server on port 6428 that serves a self-contained Project Dashboard rendered from your backlog data: board status, milestones, tasks, dependencies, activity, and decisions. The server watches backlog/ and reloads connected browser tabs automatically. If a hub from an older or otherwise mismatched version of super-backlog is already running, the dashboard automatically stops it and restarts a fresh one.

Use --port to choose a different port and --no-open to prevent the browser from opening automatically:

sbl dashboard --port 8080 --no-open

There is no static dashboard.html written to your project.

Backlog button

The Board & Quick Actions section has a single Backlog button. Clicking it makes the dashboard hub start a backlog browser process for the current project on a free local port (reused while it is alive, stopped when the hub stops) and opens the full Backlog.md UI — board, tasks, documents, decisions — in a near-fullscreen overlay. Use the overlay's "open in new tab" link for a standalone browser tab. If the backlog CLI is not installed, the button reports the failure and nothing is spawned.

Task details

Clicking a task row opens a centered detail dialog: title and status up top (the header stays visible while scrolling), metadata as a labelled grid with priority colouring and the task's pipeline phase, the description in readable paragraphs, acceptance criteria with a progress bar, dependency chips with done-markers, and copyable backlog task edit <id> and sbl phase <id> <next> commands to jump into editing or advance the phase. Descriptions and acceptance criteria are read from the task markdown files, so the dialog stays complete even where backlog task list --json omits them.

Feature Cycle steps

The Feature Cycle section shows the nine workflow phases as a compact stepper — step number and name only, with the human gates highlighted. Steps 5–8 carry a live count of the tasks currently sitting in the matching pipeline phase (phase/spec … phase/verify). Click a step to open its detail panel: the gate description plus, for tool-driven phases, the command that drives the phase (for example /superpowers:brainstorming for Brainstorming) with a one-click copy button. Purely human steps such as Idea and the Design gate carry no command.

Version and update badge

The sidebar shows the installed super-backlog version. When the daily version check has cached a newer release, an update badge appears next to it; clicking the badge copies npm i -g super-backlog to your clipboard. The dashboard never contacts the npm registry itself — the badge is fed entirely by the cache under ~/.super-backlog/version-check.json.

Pipeline phases

Every task carries its pipeline phase as a label, so you and your agents always know where work stands — no more reconstructing context at the start of a session. Tasks created through the spec-to-backlog flow start at phase/spec automatically; you advance phases only after the matching human gate:

sbl phase TASK-1          # where does this task stand?
sbl phase TASK-1 plan     # review gate passed -> advance to plan
sbl phase TASK-1 done     # archived -> clear the label
sbl doctor                # flags missing, duplicate, or unknown phase labels

The dashboard shows the live phase everywhere: counts on the Feature Cycle stepper, chips in the tasks table, and a copyable advance command in the task modal.

Full lifecycle and details: Pipeline phases.

Harness support

• OpenCode — plugin entry + file-based skills under .opencode/skill/.
• Claude Code — file-based skills under .claude/skills/ work immediately. init prints the marketplace plugin command to paste into Claude Code.

See Harness support for the full matrix.

Windows PowerShell

If you prefer npx or a globally installed sbl command, PowerShell's Execution Policy may block .ps1 shims. Use the installer above (irm ... | iex) to avoid this entirely. If you still want to use npx, run Set-ExecutionPolicy -Scope CurrentUser RemoteSigned once, or use sbl doctor for the exact policy and fix.

Next steps

• Architecture deep-dive: Architecture
• Operations and release chain: Operations
• Troubleshooting and exit codes: Troubleshooting