focus

Single-task execution loop with structured context handoff. Enforces the discipline of plan → execute → simplify → review → close. Uses Beads (bd) for task tracking so context survives across sessions without artifact clutter. A new session runs bd ready and picks up exactly where the last one left off. Supports parallel execution of independent beads via worktree-isolated subagents.

Arguments

$ARGUMENTS — format: [bead-id | "new task description" | "parallel"]

• If a bead ID is provided (e.g., agent-workflows-a3f2), work on that bead.
• If a quoted or unquoted description is provided, create a new bead.
• If parallel is provided, jump directly to parallel recommendation flow.
• If no argument is provided, run bd ready and let the user pick (includes parallel recommendations).

Parse Arguments

Extract:

• bead_id: an existing bead ID (matches pattern *-<hex>)
• task_description: free text describing new work
• parallel_mode: true if argument is "parallel"

If neither is provided, proceed to Phase 0 (Select).

Phase 0: Select

First, verify bd is available by running bd status.

If bd is not initialized, run bd init and inform the user that Beads has been set up.

Run these commands to gather full context:

bd ready --pretty
bd ready --json
bd graph --all --compact

Analysis: Priority & Parallelism

With the ready beads and dependency graph in hand, perform this analysis:

1. Sort by priority: Group ready beads by priority level (P0 > P1 > P2 > P3). Within the same priority, prefer beads with more dependents (unblocks more work).

2. Detect parallel sets: Using the bd graph --all --compact output, identify beads that are in the same dependency layer (LAYER 0, LAYER 1, etc.). Beads in the same layer with no shared parent-child or blocking relationships can execute in parallel. Cross-reference with bd ready --json to check that candidate beads don't touch overlapping file paths (check descriptions and notes for file references).