beadboard/.agents/skills/linus-beads-discipline/workflows/plan.md

2.6 KiB

Plan Workflow

"Let's figure out what needs to be done."

Trigger

  • "Plan the project" / "Create a roadmap"
  • "Break this down" / "What are the steps?"
  • "How should we approach..."
  • Starting a new epic/initiative

The Flow

1. Check Skills

Any relevant skills for planning?
- Domain decomposition skills
- Dependency analysis skills
- Use if helpful, skip if not

2. Read Context

# Check for existing related work
bd query "labels~=<area>"

# Check for constraints/decisions
bd query "labels~=design OR labels~=decision"

3. First-Principles Decomposition

Start with the goal, work backward:

1. What's the end state?
2. What needs to be true for that?
3. What's the minimal path there?
4. What are the dependencies?
5. What can be parallelized?

4. Create Bead Structure

# Create epic
bd create "<Epic Name>" --type epic --priority P1

# Create children with dependencies
bd create "<Task 1>" --type task --priority P1
bd create "<Task 2>" --type task --priority P1
bd dep add <task2-id> <task1-id>  # Task 2 depends on Task 1

# Document the plan
bd update <epic-id> --notes "## Goal
<what we're achieving>

## Approach
<high-level strategy>

## Phases
1. <phase 1>: <tasks>
2. <phase 2>: <tasks>

## Parallelization
<what can run in parallel>

## Risks
<what could go wrong>

## Success Criteria
<how we know we're done>"

5. Validate Structure

# Check for cycles
bd dep cycles

# Check dependencies are correct
bd dep tree <epic-id>

# Validate parallel paths
bd ready  # Should show unblocked beads

Planning Principles

Dependency Direction

Dependencies = execution order
NOT visual order or "nice to have"

A → B means: A must complete before B can start

Minimal Dependencies

Add dependency only if:
- Blocked task CANNOT proceed without blocker
- Real hard dependency, not soft preference

Over-chaining kills parallelization.

Priorities

P0: Critical, blocking everything
P1: Important, on the critical path
P2: Valuable, can wait briefly
P3: Nice to have, defer

Good Plans

  • Clear goal - Anyone can see what success looks like
  • Decomposed tasks - Each task is doable in one session
  • Correct dependencies - Execution order is explicit
  • Parallelization - Independent work is visible
  • Risks identified - Known unknowns are documented
  • Acceptance criteria - Each task has clear completion

For Other Agents

Your plan should let any agent:

  • Understand the overall goal
  • Pick up any ready task and go
  • Know what depends on their work
  • See where they fit in the bigger picture