agentleFS
ulpi-ioulpi-io/browse/.claude/skills/run-parallel-agents-feature-build/SKILL.md

run-parallel-agents-feature-build

Skill15 starsChanged 8 months agoApache-2.0View on GitHub

Automatically orchestrate multiple specialized agents working in parallel when building independent features, modules, or performing separate investigations. Use when the task list contains 3+ unrelated features/tasks that don't share state, don't have execution dependencies, and can be understood independently. Match each feature to the right expert agent (Laravel, Next.js, React, Node, NestJS, Remix, Express, Expo, Flutter, Magento) and run them concurrently to maximize development speed.

The file

Show its rules and text

We split this skill into the rules it gives an agent

26 rules. About 0 are common, 25 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈25

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before launching parallel agents, you ABSOLUTELY MUST:

    line 7
  2. Distinctive:

    Verify tasks are truly independent (no shared files, no data flow between them)

    line 9
  3. Distinctive:

    Create complete briefs with scope, files, output, and success criteria

    line 11
  4. Distinctive:

    Before launching ANY parallel agents, you MUST complete this checklist:

    line 23
  5. Distinctive:

    Use this skill automatically when:

    line 40
  6. Distinctive:

    No shared state or cross-dependencies between tasks

    line 48
  7. Distinctive:

    Do NOT use parallel agents when:

    line 68
  8. Distinctive:

    Shared state: - Do tasks modify the same files or data structures?

    line 98
  9. Distinctive:

    If all conditions for parallelization are met, proceed to Step 2.

    line 101
  10. Distinctive:

    Do NOT wait for agents to complete before launching others

    line 181
  11. Distinctive:

    No placeholder values - all parameters must be complete

    line 183
  12. Distinctive:

    After aggregating results, verify the parallel execution was successful:

    line 215
  13. Distinctive:

    Gate: Do NOT mark task complete until all 5 checks pass.

    line 237
  14. Distinctive:

    Launch: all three in a single message with Task tool

    line 253
  15. Distinctive:

    If agents modify overlapping files:

    line 340
  16. Distinctive:

    Always inform the user BEFORE launching parallel agents:

    line 349
  17. Distinctive:

    If an agent fails or gets blocked:

    line 362
  18. Distinctive:

    Don't retry automatically without user input

    line 367
  19. Distinctive:

    0 points: No briefs or generic briefs

    line 386
  20. Distinctive:

    0 points: No summary provided

    line 396
  21. Distinctive:

    Fix: Run dependency checks. If Output(A) ∈ Input(B), execute sequentially.

    line 424
  22. Very specific:

    Fix: Use technology detection patterns.

    line 429
  23. Distinctive:

    Fix: ALL Task tool calls in a SINGLE message.

    line 434
  24. Distinctive:

    Fix: Every brief must have Scope, Files, Output, Success criteria.

    line 439
  25. Distinctive:

    Fix: After aggregation, check file overlap.

    line 444
  26. Distinctive:

    When parallel execution is complete, announce:

    line 487

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md
<EXTREMELY-IMPORTANT> Before launching parallel agents, you **ABSOLUTELY MUST**:
  1. Verify tasks are truly independent (no shared files, no data flow between them)
  2. Match each task to the correct specialized agent
  3. Create complete briefs with scope, files, output, and success criteria
  4. Launch ALL agents in a SINGLE message with multiple Task tool calls

Launching agents without verification = conflicts, wasted work, merge hell

This is not optional. Parallel execution requires independence verification. </EXTREMELY-IMPORTANT>

Run Parallel Agents Feature Build

MANDATORY FIRST RESPONSE PROTOCOL

Before launching ANY parallel agents, you MUST complete this checklist:

  1. ☐ Count tasks — are there 3+ independent tasks?
  2. ☐ Check dependencies — do any tasks depend on others?
  3. ☐ Check file overlap — do tasks modify the same files?
  4. ☐ Match agents — assign correct agent type per task
  5. ☐ Prepare briefs — scope, files, output, success criteria for each
  6. ☐ Announce to user — "Launching N agents in parallel for [tasks]"

Launching agents WITHOUT completing this checklist = conflicts and failures.

Overview

Automatically detect opportunities for parallel execution and orchestrate multiple specialized agents working concurrently on independent features, modules, or investigations. Match each task to the appropriate domain expert and coordinate their work to deliver faster results without sacrificing quality.

When to Use This Skill

Use this skill automatically when:

Task List Indicators:

  • The task list contains 3 or more independent tasks or phases
  • Receiving a large plan with multiple unrelated features
  • Tasks are clearly scoped and don't overlap
  • Each task can be understood without context from others
  • No shared state or cross-dependencies between tasks

User Triggers:

  • "Build in parallel"
  • "Split this work across agents"
  • "Use multiple agents for this"
  • "Speed this up with parallel execution"

Common Scenarios:

  • Building independent features or endpoints (e.g., wishlist API, checkout flow, user dashboard)
  • Implementing multiple UI components or microservices simultaneously
  • Generating documentation or schema files for several modules
  • Analyzing multiple code files for specific issues
  • Debugging failures in isolated subsystems
  • Processing multiple data files or resources

When NOT to Use This Skill

Do NOT use parallel agents when:

Dependencies Exist:

  • Tasks have sequential dependencies (Task B needs Task A's output)
  • Shared state, data, or ownership between tasks
  • The sequence of execution affects the outcome
  • Agents would interfere with each other's work

Problem Not Decomposed:

  • The problem isn't yet broken into independent units
  • Need to understand full system state first
  • Related failures where fixing one might fix others
  • Careful coordination or debugging across layers is required

Single Cohesive Task:

  • Working on a single integrated feature
  • Refactoring that touches multiple interconnected parts
  • Tasks that require constant communication between components

Core Workflow

Step 1: Analyze the Task List or Request

Examine the current task list or user request to identify:

  1. Number of independent work streams - Are there 3+ separate features/tasks?
  2. Dependencies - Can each task be completed without waiting for others?
  3. Shared state - Do tasks modify the same files or data structures?
  4. Scope clarity - Is each task clearly defined with known requirements?

If all conditions for parallelization are met, proceed to Step 2.

Step 2: Match Features to Specialized Agents

For each independent task, determine the best agent type based on:

Technology Stack Detection:

  • Laravel backend → laravel-senior-engineer
  • Next.js frontend or full-stack → nextjs-senior-engineer
  • React + Vite + Tailwind frontends → react-vite-tailwind-engineer
  • Express.js APIs → express-senior-engineer
  • Node.js CLI tools → nodejs-cli-senior-engineer
  • Python backend, Django, data pipelines → python-senior-engineer
  • FastAPI specifically, async DB, JWT auth → fastapi-senior-engineer
  • Go backend, services → go-senior-engineer
  • Go CLI tools → go-cli-senior-engineer
  • Android native, Kotlin, Jetpack Compose → android-senior-engineer
  • iOS native, Swift, SwiftUI, UIKit → ios-senior-engineer
  • macOS native, AppKit, Cocoa → ios-macos-senior-engineer
  • Expo React Native mobile → expo-react-native-engineer
  • AWS infrastructure, CDK → devops-aws-senior-engineer
  • Docker, containers → devops-docker-senior-engineer
  • General tasks (exploration, research) → general-purpose

File Pattern Analysis:

*.php + /app/ + /routes/ → Laravel
*.tsx + /app/ or /pages/ + next.config.* → Next.js
*.tsx + vite.config.* + tailwind.config.* → React/Vite/Tailwind
*.ts + express imports → Express
*.ts + commander/inquirer imports → Node CLI
*.py + fastapi imports → FastAPI
*.py + django/flask/general → Python
*.go + go.mod → Go
*.kt + AndroidManifest.xml + build.gradle.kts → Android native
*.swift + UIKit/SwiftUI mobile + Package.swift or *.xcodeproj → iOS native
*.swift + AppKit/Cocoa + Package.swift or *.xcodeproj → macOS native
*.tsx + app.json (Expo) → Expo React Native
Dockerfile + docker-compose.* → Docker
CDK, CloudFormation, Terraform → AWS DevOps

Step 3: Prepare Agent Briefs

For each agent, create a clear, focused brief containing:

Required Elements:

  1. Scope of work - Exactly what to build/analyze/fix
  2. Expected output - What deliverables to produce
  3. Context - Relevant file paths, existing patterns, constraints
  4. Success criteria - How to verify completion

Brief Template:

Build [feature name]:
- Scope: [specific feature boundaries]
- Files: [relevant paths or patterns]
- Requirements: [bullet points]
- Output: [expected deliverables]
- Patterns: [existing code patterns to follow]

Step 4: Launch Agents in Parallel

Execute all agents simultaneously using a single message with multiple Task tool calls:

Use the Task tool to launch multiple agents in parallel:
- Agent 1: [brief for first feature]
- Agent 2: [brief for second feature]
- Agent 3: [brief for third feature]

Critical Requirements:

  • Send ALL Task tool calls in ONE message
  • Do NOT wait for agents to complete before launching others
  • Each agent gets its complete brief upfront
  • No placeholder values - all parameters must be complete

Step 5: Monitor and Aggregate Results

After agents complete:

  1. Collect outputs from each agent
  2. Verify deliverables match expected outputs
  3. Check for conflicts (e.g., overlapping file changes)
  4. Merge results into a coherent summary
  5. Report to user with consolidated findings

Aggregation Template:

Parallel execution complete. Results:

**[Feature 1]** (via [agent-type])
- Status: [completed/blocked/partial]
- Delivered: [summary]
- Files modified: [list]

**[Feature 2]** (via [agent-type])
- Status: [completed/blocked/partial]
- Delivered: [summary]
- Files modified: [list]

**Overall:** [X/Y features completed, any conflicts, next steps]

Step 6: Verification (MANDATORY)

After aggregating results, verify the parallel execution was successful:

Check 1: All Agents Completed

  • Every launched agent returned a result
  • No agents timed out or crashed

Check 2: No File Conflicts

  • No two agents modified the same file
  • If conflicts exist, they are resolved

Check 3: Deliverables Match Briefs

  • Each agent delivered what was specified
  • Output matches expected format

Check 4: No Integration Issues

  • Features work together (if they interact)
  • No broken imports or references

Check 5: User Informed

  • Completion announced with summary
  • Any issues or partial results documented

Gate: Do NOT mark task complete until all 5 checks pass.

Example Scenarios

Example A: Feature Build

User Says: "Build the wishlist API, checkout summary, and user dashboard in parallel."

Execution:

  1. Detect three separate feature scopes
  2. Match agents:
    • Wishlist API → laravel-senior-engineer (backend)
    • Checkout summary → nextjs-senior-engineer (frontend)
    • User dashboard → nextjs-senior-engineer (frontend/UI)
  3. Launch all three in a single message with Task tool
  4. Aggregate results into merged summary

Output:

Built 3 features in parallel:
- Wishlist API: Complete (app/Http/Controllers/WishlistController.php, routes/api.php)
- Checkout summary: Complete (app/checkout/summary/page.tsx)
- User dashboard: Complete (app/dashboard/page.tsx)

Example B: Debug Parallel Subsystems

User Says: "Run parallel agents to debug these failing tests."

Execution:

  1. Cluster failures by subsystem:
    • Laravel backend tests
    • Next.js frontend tests
    • Node.js API tests
  2. Spawn agents:
    • laravel-senior-engineer for Laravel failures
    • nextjs-senior-engineer for Next.js failures
    • express-senior-engineer for Node API failures
  3. Let each diagnose and fix independently
  4. Gather results, merge patches, re-run tests
  5. Output consolidated fix report

Example C: Code Analysis

User Says: "Analyze these 5 code files in parallel for performance bottlenecks."

Execution:

  1. Split files across appropriate agents based on file type
  2. Run analysis concurrently (each agent gets 1-2 files)
  3. Merge findings into summarized report with:
    • File-by-file breakdown
    • Common patterns across files
    • Prioritized recommendations

Agent Type Reference

Quick reference for matching tasks to agents:

Agent TypeBest ForKey Indicators
laravel-senior-engineerLaravel backends, APIs, Eloquent models*.php, /app/, Eloquent, Artisan
nextjs-senior-engineerNext.js apps, React Server Components, App Router*.tsx, /app/, /pages/, next.config.*
react-vite-tailwind-engineerReact + Vite + Tailwind TypeScript frontends*.tsx, vite.config.*, tailwind.config.*
express-senior-engineerExpress.js APIs, middleware, REST endpoints*.js/*.ts, express imports
nodejs-cli-senior-engineerNode.js CLI tools, commander.js, inquirercommander, inquirer, ora, CLI patterns
python-senior-engineerPython backends, Django, data pipelines*.py, requirements.txt, pyproject.toml
fastapi-senior-engineerFastAPI, async DB, JWT auth*.py, fastapi imports, uvicorn
go-senior-engineerGo backends, services, APIs*.go, go.mod, go.sum
go-cli-senior-engineerGo CLI tools, cobra, viper*.go, cobra imports, CLI patterns
ios-senior-engineeriOS Swift, SwiftUI, UIKit, StoreKit*.swift, UIKit/SwiftUI mobile, Xcode
ios-macos-senior-engineermacOS Swift, AppKit, Cocoa, desktop SwiftUI*.swift, AppKit/Cocoa, *.xcodeproj
expo-react-native-engineerExpo mobile apps, cross-platform*.tsx, app.json, Expo modules
devops-aws-senior-engineerAWS infrastructure, CDK, CloudFormationCDK, CloudFormation, Terraform, AWS
devops-docker-senior-engineerDocker, Docker Compose, containerizationDockerfile, docker-compose.*
general-purposeExploration, research, general tasksNon-framework-specific work

See references/agent_matching_logic.md for detailed matching rules and edge cases.

Best Practices

Scoping Tasks Effectively

Good Task Scopes:

  • "Build user authentication endpoint with JWT"
  • "Create product listing page with filters"
  • "Implement cart total calculation service"

Poor Task Scopes:

  • "Build the entire checkout flow" (too broad, likely has dependencies)
  • "Fix the app" (undefined, not decomposed)
  • "Refactor database layer" (touches too many interconnected parts)

Handling Conflicts

If agents modify overlapping files:

  1. Review changes from each agent
  2. Identify conflicts (same lines modified)
  3. Merge intelligently or ask user for guidance
  4. Re-test affected areas

Communication Pattern

Always inform the user BEFORE launching parallel agents:

I've identified 3 independent features that can be built in parallel:
1. [Feature 1] - using [agent-type]
2. [Feature 2] - using [agent-type]
3. [Feature 3] - using [agent-type]

Launching agents now...

Handling Partial Success

If an agent fails or gets blocked:

  • Continue with successful agents
  • Report partial results
  • Provide clear next steps for blocked work
  • Don't retry automatically without user input

Quality Checklist (Must Score 8/10)

Score yourself honestly before marking parallel execution complete:

Independence Verification (0-2 points)

  • 0 points: Launched without checking dependencies
  • 1 point: Checked some dependencies
  • 2 points: Full dependency check (files, data flow, state)

Agent Matching (0-2 points)

  • 0 points: Wrong agents for tasks
  • 1 point: Agents assigned without justification
  • 2 points: Each task matched to correct agent with indicators

Brief Quality (0-2 points)

  • 0 points: No briefs or generic briefs
  • 1 point: Partial briefs (missing fields)
  • 2 points: Complete briefs with scope, files, output, success criteria

Execution Correctness (0-2 points)

  • 0 points: Agents launched sequentially
  • 1 point: Parallel launch but incomplete briefs
  • 2 points: Single message with all Task tool calls, complete briefs

Result Aggregation (0-2 points)

  • 0 points: No summary provided
  • 1 point: Partial summary
  • 2 points: Complete summary with status, deliverables, conflicts, next steps

Minimum passing score: 8/10


Common Rationalizations (All Wrong)

These are excuses. Don't fall for them:

  • "These tasks look independent" → STILL check for file overlap
  • "I know the right agents" → STILL document the matching rationale
  • "Briefs are obvious" → STILL write complete scope, files, output, success criteria
  • "Sequential is safer" → If tasks are independent, parallel is FASTER with no downside
  • "I'll merge conflicts later" → Check for conflicts BEFORE launching
  • "One message is too long" → ALL Task tool calls MUST be in one message
  • "The user just wants it done" → Correct execution = faster completion
  • "Two tasks don't need parallel agents" → Correct, need 3+ tasks

Failure Modes

Failure Mode 1: Launching Dependent Tasks in Parallel

Symptom: Agents fail or produce conflicting output because Task B needed Task A's result Fix: Run dependency checks. If Output(A) ∈ Input(B), execute sequentially.

Failure Mode 2: Wrong Agent Assignment

Symptom: Laravel work done by nextjs-senior-engineer, poor results Fix: Use technology detection patterns. Match *.php → Laravel agent.

Failure Mode 3: Sequential Launch Instead of Parallel

Symptom: Sent multiple messages with Task tool calls, agents ran one at a time Fix: ALL Task tool calls in a SINGLE message. This is critical for parallel execution.

Failure Mode 4: Incomplete Briefs

Symptom: Agents ask clarifying questions or deliver wrong output Fix: Every brief must have Scope, Files, Output, Success criteria. No placeholders.

Failure Mode 5: No Conflict Check After Completion

Symptom: Agents modified same files, changes overwrite each other Fix: After aggregation, check file overlap. If conflicts, merge intelligently.


Quick Workflow Summary

STEP 1: ANALYZE
├── Count tasks (need 3+)
├── Check dependencies
├── Check file overlap
└── Gate: Tasks are independent

STEP 2: MATCH AGENTS
├── Identify technology per task
├── Assign correct agent type
└── Document matching rationale

STEP 3: PREPARE BRIEFS
├── Scope, Files, Output, Success criteria
└── No placeholders allowed

STEP 4: LAUNCH PARALLEL
├── Single message with ALL Task tool calls
├── Do NOT wait between launches
└── Gate: All agents launched

STEP 5: AGGREGATE RESULTS
├── Collect outputs
├── Check for conflicts
└── Merge into summary

STEP 6: VERIFICATION
├── All agents completed
├── No file conflicts
├── Deliverables match briefs
└── Gate: All 5 checks pass

Completion Announcement

When parallel execution is complete, announce:

Parallel execution complete.

**Quality Score: X/10**
- Independence Verification: X/2
- Agent Matching: X/2
- Brief Quality: X/2
- Execution Correctness: X/2
- Result Aggregation: X/2

**Results:**
- Features completed: X/Y
- Agents used: [list]
- Conflicts: [none/resolved/pending]

**Summary:**
[Brief description of what was delivered]

**Next steps:**
[Any remaining work or follow-up needed]

Integration with Other Skills

The run-parallel-agents-feature-build skill integrates with:

  • plan-to-task-list-with-dag — Use to generate a structured plan, then this skill to execute it
  • plan-founder-review — Review the plan before execution (quality gate)
  • start — Use start to identify if parallel agents are appropriate
  • run-parallel-agents-feature-debug — For debugging, use the debug variant instead

Workflow: start → plan-to-task-list-with-dag → plan-founder-review → run-parallel-agents-feature-build


Resources

references/

  • agent_matching_logic.md - Detailed rules for matching features to agent types, edge cases, and technology detection patterns

This skill does not require scripts or assets - it orchestrates existing Claude Code agent capabilities.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.