utility-scripts
Comprehensive guide to teamwork script usage patterns for project, task, and wave management. Use this skill to understand available scripts, their parameters, and common usage patterns.
SKILL.md
| Name | utility-scripts |
| Description | Comprehensive guide to teamwork script usage patterns for project, task, and wave management. Use this skill to understand available scripts, their parameters, and common usage patterns. |
hardworker-marketplace
Claude Code plugin marketplace focused on "hardworker" productivity patterns: verification-first development, multi-session collaboration, and evidence-based completion.
TL;DR
- ultrawork - Strict verification-first development. Use for solo features with evidence requirements.
- teamwork - Multi-session collaboration. Use for team projects with parallel workers.
- knowledge-extraction - Capture and extract insights. Use for learning and documentation.
Quick Install:
claude plugin marketplace add mnthe/hardworker-marketplace
claude plugin install ultrawork@hardworker-marketplace
Detailed Comparison → | Architecture →
Prerequisites
- Claude Code CLI (latest version with plugin support) - https://claude.ai/download
- Bun 1.3+ (runtime for all plugins) - https://bun.sh/
- Git (version control)
- Platform: Windows, MacOS, or Linux
Quick Start
Installation
# Add marketplace
claude plugin marketplace add mnthe/hardworker-marketplace
# Install verification-first development
claude plugin install ultrawork@hardworker-marketplace
# Install multi-session collaboration
claude plugin install teamwork@hardworker-marketplace
# Install knowledge extraction
claude plugin install knowledge-extraction@hardworker-marketplace
Example: ultrawork Session
# Start verification-first development session
/ultrawork "implement user authentication with JWT"
# System creates plan with tasks and success criteria
# Worker agents execute tasks in parallel
# Verifier checks all criteria with concrete evidence
# Check status
/ultrawork-status
Example: teamwork Session
# Start coordination (orchestrator spawns native teammates automatically)
/teamwork "build REST API with tests and docs"
# Orchestrator creates tasks, spawns role-based workers as native teammates
# Workers execute tasks, coordinate via SendMessage
# Event-driven: TaskCompleted and TeammateIdle hooks drive progress
# Check status
/teamwork-status
Which Plugin Should I Use?
| Scenario | Recommendation |
|---|---|
| Solo developer, strict verification | ultrawork |
| Team collaboration, parallel work | teamwork |
| Capturing code patterns and insights | knowledge-extraction |
| CI/CD and automation | ultrawork --auto |
| Parallel agent collaboration | teamwork |
| TDD workflow enforcement | ultrawork |
See Detailed Scenarios below for more guidance.
Plugins
ultrawork - Verification-First Development
Strict verification-first development mode with session isolation and mandatory evidence collection.
Key Features: Session isolation, mandatory planning with success criteria, evidence-based completion, Execute→Verify loop with auto-retry, parallel task execution, cross-platform support.
teamwork - Native Teammate Collaboration
Role-based worker agents for parallel development using Claude Code's native teammate API.
Key Features: Role-based worker agents (frontend, backend, devops, test, docs, security, review), native teammate API, event-driven coordination (TaskCompleted, TeammateIdle hooks), dashboard status, project and team isolation.
knowledge-extraction - Extract Knowledge from Codebases
Extract and manage knowledge from codebases for AI agent context.
Key Features: Pattern-based knowledge extraction, markdown documentation generation, context management for AI agents, integration with ultrawork and teamwork.
Plugin Comparison
| Feature | ultrawork | teamwork | knowledge-extraction |
|---|---|---|---|
| Primary Use Case | Solo developer, strict verification | Agent team collaboration, parallel work | Pattern capture, documentation |
| Session Model | Single session, isolated worktree | Single session, native teammate API | Session-based insights |
| Verification | Mandatory, multi-tier (task + final) | Final verifier agent | Not applicable |
| Evidence Collection | Automatic via hooks | Structured via workers | Insight extraction |
| Concurrency | Parallel workers, single session | Native teammates, event-driven coordination | Single session |
| Role Specialization | No | Yes (8 worker roles) | No |
| Auto Mode | Yes (--auto flag) | No (interactive only) | No (hook-based) |
| Best For | Feature implementation, bug fixes | Large features, parallel agent work | Learning, documentation |
Architecture Overview
System Architecture
flowchart TD
CLI[Claude Code CLI] --> Commands
CLI --> Hooks[Lifecycle Hooks]
subgraph ultrawork
UW_CMD["/ultrawork"] --> UW_AGENTS[6 Agents]
UW_AGENTS --> UW_STATE["~/.claude/ultrawork/"]
Hooks --> UW_HOOKS["Evidence/Gates"]
UW_HOOKS --> UW_STATE
end
subgraph teamwork
TW_CMD["/teamwork"] --> TW_AGENTS[10 Agents]
TW_AGENTS --> TW_STATE["~/.claude/teamwork/"]
Hooks --> TW_HOOKS[Event Hooks]
TW_HOOKS --> TW_STATE
end
subgraph knowledge-extraction
KE_CMD["/insights"] --> KE_AGENT[Extractor]
Hooks --> KE_HOOKS[Auto Extract]
KE_HOOKS --> KE_STATE["~/.claude/knowledge-extraction/"]
KE_AGENT --> KE_STATE
end
Commands --> UW_CMD
Commands --> TW_CMD
Commands --> KE_CMD
UW_STATE -.-> KE_HOOKS
TW_STATE -.-> KE_HOOKS
Data Flow
- ultrawork: Goal → Explorer → Planner → Workers (parallel) → Verifier → Complete
- teamwork: Goal → Orchestrator → Tasks ← Workers (native teammates) → Verification → Complete
- knowledge-extraction: Session → Hooks → Storage → Extract → Components
Decision Guide
Scenario 1: Solo Developer, Single Feature Implementation
Recommendation: ultrawork
Use ultrawork when:
- Working alone on a well-defined feature
- Need strict verification and evidence
- Want automatic retry on failures
- Require TDD workflow enforcement
- Need session isolation (worktree support)
/ultrawork "implement user authentication with JWT"
Scenario 2: Team Collaboration, Large Feature
Recommendation: teamwork
Use teamwork when:
- Need parallel agent collaboration on a project
- Tasks can be parallelized across roles
- Need specialization (frontend, backend, test, etc.)
- Want event-driven coordination with native teammate API
- Require final verification by dedicated verifier agent
# Single session: Orchestrator spawns and coordinates native teammates
/teamwork "build REST API with authentication"
# Orchestrator automatically:
# - Creates tasks via TaskCreate
# - Spawns workers: Task(teamwork:backend), Task(teamwork:frontend), etc.
# - Assigns tasks via TaskUpdate(owner)
# - Monitors via TaskCompleted and TeammateIdle hooks
Scenario 3: Strict Verification Required
Recommendation: ultrawork
Use ultrawork when:
- Zero tolerance for incomplete work
- Need blocked pattern detection ("TODO", "FIXME")
- Require concrete evidence for every success criterion
- Want automatic execute→verify retry loop
- Need final verifier to audit all evidence
Scenario 4: Parallel Agent Development
Recommendation: teamwork
Use teamwork when:
- Need multiple agents working in parallel
- Workers should receive tasks from orchestrator
- Need role-based task assignment
- Require event-driven coordination
- Want native teammate API for agent lifecycle management
Scenario 5: Capturing Code Patterns and Insights
Recommendation: knowledge-extraction
Use knowledge-extraction when:
- Want to learn from past sessions
- Need to document discovered patterns
- Building reusable skills and commands
- Creating project-specific rules files
- Extracting architectural decisions
# During any session, Claude generates insights
★ Insight ─────────────────────────────────────
JWT tokens should include minimal claims.
─────────────────────────────────────────────────
# Later, extract to components
/insights extract
Scenario 6: CI/CD and Automation
Recommendation: ultrawork --auto
Use ultrawork auto mode when running in CI/CD pipelines with no user interaction possible, well-defined tasks with clear criteria, or need fully autonomous operation.
/ultrawork --auto "add unit tests for payment module"
Skills System Overview
Plugins provide reusable skills that can be referenced by agents and commands. Skills are markdown files containing documented patterns, workflows, and best practices.
ultrawork Skills (15 skills)
Core workflow skills: planning (task decomposition), overview-exploration (project discovery), ultrawork (session lifecycle), tdd-workflow (test-first enforcement).
Utility skills: scripts-path-usage (script access), data-access-patterns (state reading), utility-scripts (common patterns).
Review skills: code-quality-review, architecture-review, security-review, consistency-review.
Pattern skills: backend-patterns, frontend-patterns, security-patterns, testing-patterns.
Key Pattern: All agents use scripts-path-usage for correct script invocation.
teamwork Skills (4 skills)
Core workflow skills: worker-workflow (task execution with native API), event-coordination (hook-based orchestration patterns), task-decomposition (parallel planning with role assignment), teamwork-clean (project reset).
Key Pattern: The worker-workflow skill is shared by all 8 role-specific workers. The event-coordination skill drives the orchestrator's hook-based coordination via TaskCompleted and TeammateIdle events.
knowledge-extraction Skills (1 skill)
insight-awareness: Documents ★ Insight format and automatic capture workflow. Operates primarily via hooks.
Skills Invocation
Skills are referenced in agent frontmatter:
---
name: worker
skills: [scripts-path-usage, data-access-patterns, worker-workflow]
---
Skills are also available as standalone commands in Claude Code CLI (when skill has frontmatter with command: true).
Troubleshooting
Issue 1: Version Mismatch Between plugin.json and marketplace.json
Symptom: Plugin updates not reflected after installation from marketplace.
Cause: plugins/{plugin}/.claude-plugin/plugin.json version doesn't match .claude-plugin/marketplace.json version.
Solution:
# Verify version sync
bun -e "
const pluginVersion = require('./plugins/ultrawork/.claude-plugin/plugin.json').version;
const marketplaceVersion = require('./.claude-plugin/marketplace.json').plugins
.find(p => p.name === 'ultrawork').version;
if (pluginVersion !== marketplaceVersion) {
console.error('Version mismatch:', pluginVersion, '!=', marketplaceVersion);
process.exit(1);
}
console.log('Versions match:', pluginVersion);
"
Prevention: Always update both files when bumping plugin version.
Issue 2: Session Not Found (ultrawork)
Symptom: Command fails with "Session not found" error.
Solution: List sessions with /ultrawork-status --all. Verify $CLAUDE_SESSION_ID is set. Sessions in terminal states are auto-deleted after 7 days.
Issue 3: Workers Not Receiving Tasks (teamwork)
Symptom: Spawned workers are idle and not executing tasks.
Solution: Check /teamwork-status. Verify the orchestrator has created tasks via TaskCreate and assigned them via TaskUpdate(owner). Check that TaskCompleted and TeammateIdle hooks are registered in hooks/hooks.json. Workers receive tasks through orchestrator assignment, not self-claiming.
Issue 4: Evidence Collection Failing (ultrawork)
Symptom: Verifier reports "missing evidence" despite task completion.
Solution: View /ultrawork-evidence. Verify evidence includes command, output, and exit code. Check for blocked patterns ("should work", "TODO", "FIXME"). Ensure hooks are registered in hooks.json.
Issue 5: Hook Not Triggering (all plugins)
Symptom: Expected hook behavior doesn't occur (evidence not collected, insights not captured, gates not enforced).
Solution: Verify Bun is installed (bun --version). Check hook registration in hooks/hooks.json. Test manually: echo '{"session_id": "test"}' | bun hook-script.js. Use explicit bun prefix in hooks.json for cross-platform compatibility.
Contributing
See CLAUDE.md for full development guidelines.
Quick Guidelines
Version Sync: When updating plugin versions, MUST sync both plugins/{plugin}/.claude-plugin/plugin.json and .claude-plugin/marketplace.json.
Script Requirements:
- Flag-based parameters (no positional args)
- JSON output for structured data
- Exit codes: 0 (success), 1 (error)
Version Bumps:
- Patch (0.0.x): Bug fixes, typo fixes
- Minor (0.x.0): New commands/agents
- Major (x.0.0): Breaking changes
Commit Convention:
feat(plugin): Add new feature
fix(plugin): Fix bug
docs(plugin): Update documentation
refactor(plugin): Refactor code
License
MIT