Planning Pipeline

/lavra-design

From idea to implementation-ready plan. One command runs brainstorm, research, adversarial review, and locks every bead with file paths, test cases, and acceptance criteria.

Works for non-technical founders too. Describe a feature in plain English -- the pipeline handles planning, research, review, and structuring.

Act 1

Brainstorm

Interactive dialogue that sharpens scope before any code is planned. Six structured steps force hard questions upfront.

STEP 1

What to build

Vision & success criteria

Define what success looks like. Who is this for? What triggered the idea?

STEP 2

User interview

One question at a time

Deep questioning: vision, constraints, scope, success criteria. No rushing.

STEP 3

Gray areas

Resolve ambiguity upfront

Sync or async? API layer or client? Decide now, not during implementation.

STEP 4

CEO Review

Scope expansion or reduction?

Find the 10-star product, challenge premises. What's the smallest version that proves this works?

STEP 5

Approach selection

2-3 options, pros & cons

Concrete approaches with trade-offs. Lead with recommendation. Apply YAGNI (You Aren't Gonna Need It).

STEP 6

Phase breakdown

Filed as child beads

Logical implementation phases with scope, goals, and locked decisions per phase.

The full pipeline

Six phases transform an idea into beads so detailed that subagents can implement without asking questions.

Brainstorm Scope, edge cases, approaches
Plan Structure into actionable beads
Research 5 domain-matched agents in parallel
Revise Integrate research findings
Review 4 adversarial reviewers challenge the plan
Lock Frozen. Implementation-ready.

Research

5 domain-matched agents dispatched in parallel

Each agent gathers evidence -- docs, prior art, best practices, edge cases -- and logs findings as comments on the relevant beads.

Best Practices PARALLEL
Framework Docs PARALLEL
Git History PARALLEL
Past Learnings PARALLEL
Repo Structure PARALLEL

Review

4 adversarial reviewers challenge the plan

Each reviewer runs in parallel, looking for blind spots from a different angle. Findings are categorized as safe-to-auto-apply or requiring user judgment.

Architecture PARALLEL
Security PARALLEL
Performance PARALLEL
Simplicity PARALLEL
PHASE 4

Revise

Integrate research findings

Research findings are merged into bead descriptions. Conflicts between research and plan assumptions are resolved.

PHASE 5

Cross-reference

File-scope conflict prevention

No two tasks touch the same file without a dependency. Parallel-safe.

PHASE 6

Lock

Every bead implementation-ready

File paths, test cases, acceptance criteria, locked decisions. No questions left.

Structure into actionable beads

File paths. Test cases. Dependencies. No questions left.