Skip to main content
Lesson 4 of 5 15 min

Install Beads

Agent-native task tracking via Claude Code.

Install Beads

Beads is agent-native task management. While Taskwarrior was built for humans, Beads was built for AI agents—including Claude Code.

Why Beads?

The problem: How does an AI agent remember what it was working on across sessions?

Context windows have limits. Conversations get summarized. But work continues.

Beads solves this by:

  1. Storing issues in Git: Your tasks travel with your code
  2. Providing machine-readable output: --robot-priority for agent consumption
  3. Tracking dependencies: Which tasks block which
  4. Persisting across sessions: Claude Code can resume work

The Core Workflow

# Session Start: What's highest impact?
bd ready

# During Work: Capture discovered tasks
bd create "Fix the authentication bug"

# Track Dependencies
bd dep add <id> blocks <other-id>

# Session End: Close and sync
bd close <id>
bd sync

That's it. Four commands cover 90% of use.

Installation via Claude Code

You'll use Claude Code to install Beads. This continues the bootstrap pattern—tools installing tools.

Start Claude Code in any project directory:

claude

Then use the praxis prompt to have Claude Code walk you through installation.

Key Commands

After installation, these are your essential commands:

Command Purpose
bd ready Show unblocked work (start here)
bd create "..." Create a new issue
bd close <id> Mark issue complete
bd sync Sync with git remote
bd list Show all open issues
bd show <id> Detailed issue view

The .beads Directory

Beads stores everything in .beads/:

.beads/
├── beads.db      # Local SQLite cache (gitignored)
├── issues.jsonl  # Source of truth (git-tracked)
└── config.yaml   # Repository settings

The issues.jsonl file commits with your code. This is intentional—issues and code travel together.

Robot Mode

For Claude Code consumption:

bv --robot-priority    # Ranked by PageRank + Critical Path
bv --robot-insights    # Bottleneck detection
bv --robot-plan        # Suggested execution order

Claude Code uses these flags to understand what matters most.

Label Conventions

We use consistent labels across CREATE SOMETHING:

Scope (where):

  • agency — Client work
  • io — Research, documentation
  • space — Practice, learning
  • ltd — Philosophy, canon

Type (what):

  • feature — New capability
  • bug — Something broken
  • research — Investigation
  • refactor — Structural improvement

Philosophy Note

Beads enables hermeneutic continuity. When Claude Code starts a session, it can run bd ready and immediately understand the state of work.

This isn't just convenience—it's a different model of human-agent collaboration. Work persists. Context transfers. The conversation continues.

Next Step

With Beads installed, you have the core trifecta: terminal (WezTerm), partner (Claude Code), memory (Beads).

One more tool remains: Neomutt for terminal email.


See Also

  • Beads Patterns — Complete command reference, molecules/chemistry system, and robot mode flags
  • Harness Patterns — Multi-session autonomous work orchestration using Beads

Complete the praxis exercise to install Beads and create your first task.

Praxis Exercise

This lesson includes a hands-on exercise: beads-setup

Start Exercise →