Visual task board (Backlog.md)

See open work on a Kanban board while specs stay the source of truth.

Requires packs: spec and backlog. (spec is pulled in automatically.) See Packs.

npm

npx create-lean-agent-kit@latest . --enable-pack backlog

pnpm

pnpm dlx create-lean-agent-kit@latest . --enable-pack backlog

yarn

yarn dlx create-lean-agent-kit@latest . --enable-pack backlog

bun

bunx create-lean-agent-kit@latest . --enable-pack backlog

What it is

Lean Agent Kit tracks work in Markdown — docs/specs/, ACTIVE_CONTEXT, PROGRESS. That keeps the agent's context lean, but there is no built-in Kanban or web UI.

Backlog.md fills that gap as an optional integration: a Markdown-native task manager with a terminal board and browser UI, synced to your kit specs through the leanagentkit-backlog skill.

The scaffolder (npm create lean-agent-kit) does not install Backlog.md. You install it yourself (or let the agent guide you during bootstrap).

The spec remains the source of truth for problem, scope, and acceptance criteria. The Backlog card is the status layer — column, checkboxes, notes.

Do I need this pack?

• Enable if you want backlog board / browser UI synced to specs, and you (or your team) will keep the Backlog CLI installed.
• Skip if Markdown specs + PROGRESS.md are enough visualization — zero impact without the pack and without a Backlog project.

Use cases

• Team standup — open backlog board or the browser UI; cards mirror To Do / In Progress / Done without leaving Markdown.
• Spec ↔ card link — new-spec creates a card with --ref to the spec; implement/finalize move status only when the work is real.
• Session priming — ambient start / optional start-session lists open Backlog cards so you resume the right feature.
• Trevor UX — optional Trevor wrapper for board commands without changing Done rules.

Why integrate?

Without Backlog.md	With Backlog.md
Specs + memory files only	Same specs + visual Kanban
Progress in PROGRESS.md	Cards move on the board as work advances
Agent reads ACTIVE_CONTEXT	You see open tasks at a glance (backlog board)

Install

Install the global CLI with your package manager of choice:

npm

npm i -g backlog.md

pnpm

pnpm add -g backlog.md

yarn

yarn global add backlog.md

bun

bun add -g backlog.md

brew

brew install backlog-md

Then initialize in your project root:

backlog init "My Project"

# No Git repo? Filesystem-only mode
backlog init "My Project" --no-git

During backlog init, when asked about AI integration, choose Skip — Lean Agent Kit owns AGENTS.md. Do not run backlog agents --update-instructions.

Enable in Lean Agent Kit

1. Scaffold or upgrade the kit in your project.
2. Run bootstrap and answer Yes to the Backlog.md offer (Step 3b), or say:

   Read .agent/skills/leanagentkit-backlog.md and follow it.

3. leanagentkit-match-stack detects Backlog and advertises the skill in AGENTS.md §7.

Integration is active only when both are true:

• backlog is on your PATH
• A Backlog project exists (backlog/, .backlog/, or backlog.config.yml)

Otherwise lifecycle skills skip Backlog steps silently — zero impact.

How it maps to the daily loop

leanagentkit-start-session / ambient start  →  lists open Backlog cards (To Do / In Progress)
leanagentkit-new-spec       →  creates card linked to spec (--ref docs/specs/…)
leanagentkit-implement-spec →  moves card to In Progress; checks ACs as you go
leanagentkit-check          →  reports linked Backlog card id (non-blocking)
leanagentkit-end-session / finalize    →  Done only when spec Status: done

Linking spec and card

When new-spec creates a Backlog card, the returned task id is stored in the spec frontmatter:

> Backlog: BACK-12 · Status: active · Updated: 2026-07-02

The card references the spec via --ref docs/specs/NNN-feature.md.

No false "Done"

A card moves to Done only when:

• All acceptance criteria in the spec are checked, and
• leanagentkit-check passes, and
• The spec is Status: done

Ending a session alone does not complete a card. Incomplete work gets --append-notes only.

Human visualization

backlog board      # terminal Kanban
backlog browser    # web UI (default http://localhost:6420)
backlog overview   # project statistics

Optional MCP (Cursor / Claude Code)

CLI is the portable baseline for all agents. For Cursor or Claude Code you may add the Backlog MCP server — see .agent/skills/leanagentkit-backlog.md for setup. Same status rules apply regardless of CLI vs MCP.

Further reading

• Backlog.md repository
• Kit skill: template/.agent/skills/leanagentkit-backlog.md (in your project: .agent/skills/leanagentkit-backlog.md)
• Trevor assistant — optional Backlog UX wrapper via leanagentkit-ask-trevor
• Full guide — daily loop section includes Backlog.md overview