Claude Code Plugin

Your AI development
team, one command away.

pokayokay turns Claude Code into a structured team of 13 specialized agents that plan, implement, and review your features across sessions — with parallel execution, worktree isolation, and two-stage review built in.

npx pokayokay Star on GitHub
13 Agents
21 Skills
4 Modes
∞ Sessions
[01] The problem

Claude Code forgets.
pokayokay remembers.

Every session starts from scratch. Context is lost. Tests are skipped. pokayokay chains sessions, tracks progress, and verifies every commit.

Without pokayokay
⟳ Re-explaining codebase every session ~20 min
✗ Tests claimed but not written
✗ No spec verification
✗ Context lost at 45 minutes
? Manual progress tracking
With pokayokay
Webhook Integration 5/5 tasks
✓ Design webhook API a3f2b1c
✓ Implement signature verification 7d4e9f2
✓ Add retry logic b8c3a6d
✓ Write integration tests f1a7e3c
✓ Add monitoring dashboard 9e2d8b1
Spec: 12/12 Coverage: 94% Security: Clear
[02] How it works

Three commands.
From PRD to production.

01

Plan

Feed a PRD. Get structured epics, stories, and tasks with dependency chains and skill routing.

Terminal
$ /pokayokay:plan docs/auth-prd.md
Analyzing PRD... routing skills...
Epic: User Authentication
Story: Login flow → 3 tasks [api-design, testing-strategy]
Story: Password reset → 2 tasks [api-design, security-audit]
Story: OAuth integration → 4 tasks [api-integration]
Created 3 stories, 9 tasks with dependency chains
02

Work

Dispatch agents in parallel. Each task gets a fresh context, worktree isolation, and two-stage review.

Terminal
$ /pokayokay:work semi-auto -n 3
Dispatching 3 parallel agents...
yokay-implementer → design auth api a3f2b1c
yokay-implementer → implement jwt tokens 7d4e9f2
yokay-implementer → add login form b8c3a6d
yokay-spec-reviewer → PASS
yokay-quality-reviewer → PASS
Story complete: Login flow (3/3 tasks)
03

Audit

Verify completeness across 5 named dimensions. Know it's done before you merge.

Terminal
$ /pokayokay:audit --full
Accessibility L5 Feature user-accessible
Testing T4 94% coverage
Documentation D3 API docs complete
Security S4 OWASP scan clear
Observability O3 Metrics + tracing
Overall health: 94% — ready to merge
[03] Autonomy

You set the dial.

Start hands-on. Dial up autonomy as trust builds. Four levels, from full control to fully autonomous.

Supervised

Pauses after every task. You review everything before it continues.

First-time setup

Semi-Auto

Auto-continues within stories. Pauses at story boundaries for your review.

Recommended

Auto

Chains sessions automatically. Only pauses when an entire epic is complete.

Large migrations

Unattended

Fully autonomous. Chains, reviews, continues until all work is done.

CI/CD pipelines
[04] Architecture

Built for how Claude Code actually works.

Each task gets a fresh context window. Skills route domain knowledge. Sessions chain automatically when context fills.

Parallel Dispatch

Run up to 5 agents simultaneously. Dependency graph prevents unsafe parallelization.

Coordinator
implementer
implementer
implementer

Worktree Isolation

Features and bugs run in isolated git worktrees. Same-story tasks share a worktree.

main
feature/auth worktree
fix/nav-bug worktree

Two-Stage Review

Every task passes adversarial spec review, then code quality review. Both must PASS.

1
Spec Review Requirements met? No scope creep?
PASS
2
Quality Review Well-written? Tested?
PASS

Session Chaining

When context fills, sessions chain automatically. Handoff preserves progress and WIP state.

Session 1
4 tasks · 93% context
Session 2
1 task · 45% context
Done
[05] The team

13 agents. 21 skills.
One orchestrated pipeline.

Each agent runs in its own context window with enforced constraints. Read-only agents can't modify files. Haiku agents cut costs 5-10x for exploration.

Agent Model Purpose
yokay-planner Sonnet PRD analysis and structured plan generation
yokay-brainstormer Sonnet Refines ambiguous tasks into clear requirements
yokay-explorer Haiku Fast codebase exploration (5-10x cheaper)
yokay-implementer Sonnet TDD implementation with fresh context
yokay-fixer Sonnet Auto-retry on test failures with targeted fixes
yokay-spec-reviewer Sonnet Adversarial spec compliance review
yokay-quality-reviewer Sonnet Code quality review (after spec passes)
yokay-reviewer Sonnet General code review and analysis
yokay-auditor Sonnet L0-L5 completeness scanning
yokay-security-scanner Sonnet OWASP vulnerability scanning
yokay-browser-verifier Sonnet Browser verification for UI changes
yokay-spike-runner Sonnet Time-boxed investigations
yokay-test-runner Haiku Test execution with concise output

Skills loaded on demand

Orchestration
work-sessionplanningplan-revisionspikedeep-researchsession-reviewfeature-auditworktreesbrowser-verification
Domain
api-designapi-integrationdatabase-designarchitecture-reviewci-cdcloud-infrastructureobservabilitytesting-strategysecurity-auditerror-handlingsdk-developmentdocumentation
[06] Commands

One plugin, every workflow.

20 commands organized by intent. From quick fixes to multi-day research — each command routes to the right skills and agents.

Core Workflow
/plan Analyze PRD, create task breakdown
/work Start work session with mode selection
/audit Verify across 5 dimensions
/review Analyze session patterns
/handoff Prepare context for next session
Ad-Hoc
/quick Inline task, no agents needed
/fix Bug diagnosis + fix pipeline
/spike Time-boxed investigation
/hotfix Production incident response
Development
/api REST/GraphQL design
/arch Architecture review
/db Schema & migrations
/test Testing strategy
/integrate Third-party API integration
/sdk SDK creation & extraction
Ops & Quality
/cicd CI/CD pipelines
/security OWASP vulnerability scan
/observe Logging, metrics, tracing
/docs Technical documentation
/research Extended investigation

Before you start

  • Requires Claude Code v1.0+ — pokayokay is a plugin, not standalone
  • Uses ohno MCP for task tracking — npx pokayokay handles all setup
  • Best for multi-step features — for quick one-offs, use /pokayokay:quick
  • Open-source and evolving — MIT licensed, contributions welcome