Early Discovery Phase — Claude Code plugin

Feature planning
built for agents

One command drives the entire feature lifecycle — discussion, research, planning, and execution — with parallel AI agents and full git tracking.

A Claude Code plugin. Works with any codebase, scales to monorepos.

$ npx specdacular

Built for how agents actually work

Every design decision optimizes for agent context, not human convenience.

One Command Lifecycle

/specd.continue reads task state and always knows what comes next. Discussion, research, planning, execution — fully orchestrated without remembering which command to run.

Parallel Agents

Specialized agents run simultaneously with fresh 200k context each. Codebase mapping, research, and execution agents never share context — no token pollution, faster results.

Plans as Prompts

Each PLAN.md is exactly what you would send to an implementing agent. Plans contain everything needed to implement without questions — requirements, context, and decisions.

Git-Tracked and Resumable

Every state change commits automatically. Stop mid-feature, switch branches, hand off to a teammate — /specd.continue picks up exactly where things left off.

Customizable Pipeline

The full pipeline is defined in pipeline.json. Swap any workflow, add pre/post hooks, or replace the entire pipeline. Nothing is hardcoded.

Monorepo Native

Auto-detects sub-projects, maps each one in parallel, and produces system-level topology docs. Cross-project dependency tracking and contract validation built in.

Six commands. Full control.

Everything runs through slash commands in Claude Code. Simple to start, powerful when you need it.

/specd.new <name>

Initialize a new task and start the AI-driven discussion

/specd.continue <name>

Continue the lifecycle — picks up exactly where you left off

/specd.status

Dashboard showing all tasks, stages, and progress

/specd.codebase.map

Spawn parallel agents to analyze and document your codebase

/specd.toolbox <name>

Direct access to any individual operation

/specd.config

Configure auto-commit, pipeline, and project settings

The lifecycle

One command drives every stage. You decide when to pause, when to continue, and when to hand off.

Claude Code — /specd.continue
$ /specd.continue my-feature
Reading task state...
Current stage: research
Spawning 3 research agents in parallel...
Agent 1: codebase integration patterns
Agent 2: external library approaches
Agent 3: known pitfalls and edge cases
Research complete. Writing RESEARCH.md...
Ready to plan. Continue? [Y/n]
1
/specd.new my-feature
Initialize

Creates the task folder and starts the first discussion. Claude probes requirements, follows threads, and captures technical decisions. Stop any time — state persists.

2
Discussion
Clarify Requirements

Claude probes gray areas, asks targeted questions, and records decisions with rationale in DECISIONS.md. Context accumulates across sessions.

3
Research
Parallel Investigation

Three agents spawn simultaneously — codebase integration, external patterns, and pitfalls. Each gets a fresh 200k context window. Output: RESEARCH.md.

4
Planning
Create Phases

Generates ROADMAP.md with phases derived from dependency analysis. Phase goals only — no detailed plans yet. Later phases adapt based on what earlier phases actually produce.

5
Execution
Just-in-Time Implementation

Each phase gets a detailed PLAN.md right before execution. Agents implement, commit per task, and track progress. A review agent compares output against plan intent after each phase.

6
Complete
Reviewed and Committed

All phases pass review. State is fully committed to git. Every decision, context file, and phase plan is preserved for future reference or handoff.

specdacular — bash
$ /specd.new user-auth
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
NEW TASK: user-auth
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Tell me about this feature...
> User authentication with OAuth...
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
RESEARCHING: user-auth
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Spawning 3 research agents...
- Codebase Integration: ✓
- Implementation Patterns: ✓
- Pitfalls: ✓
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
PLANNING COMPLETE
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
3 phases planned
Phase 1: executing...
Phase 1: ✓ complete
Phase 2: executing...
Phase 2: ✓ complete
Phase 3: executing...
Phase 3: ✓ complete
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
TASK COMPLETE: user-auth
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

The Brain

A config-driven orchestrator reads your task state and drives the entire lifecycle. Every step is automatic, resumable, and git-tracked.

DDiscussionDefine goalsRResearchParallel agentsPPlanningRoadmapEExecutionCode & buildloopVReviewValidateCCompleteShip itA1A2A3paralleliterates

Built for teams

specdacular works with the codebases and workflows you already have. No new conventions to adopt, no lock-in.

Monorepo native

Auto-detects sub-projects via package.json, go.mod, Cargo.toml, and others. Maps each in parallel, then produces system-level topology: PROJECTS.md, TOPOLOGY.md, CONTRACTS.md. Cross-project dependency graphs with cycle validation.

Seamless handoff

Git-tracked state means any team member can pick up where another left off. All decisions, research, context, and progress persist across sessions and people — no verbal handoff required.

Persistent decision log

DECISIONS.md records every technical decision with date, context, rationale, and implications. Once recorded, decisions are not re-litigated. The full reasoning trail is always accessible.

Existing codebase integration

Start with /specd.codebase.map to generate AI-optimized context docs — MAP.md, PATTERNS.md, STRUCTURE.md, CONCERNS.md. Documents what Claude cannot get from reading code: tribal knowledge, gotchas, conventions.

Work with us

We build with specdacular.

Our team uses specdacular on every project. If you want help integrating AI-driven development workflows — or want a team that already works this way — get in touch.

hello@specdacular.dev