user-intent-patterns

name	user-intent-patterns
description	Pattern matching for natural language user requests to appropriate workflows, agents, and commands for seamless task routing.

User Intent Patterns

Pattern matching for natural language user requests to appropriate workflows, agents, and commands.

Core Principle

Users describe what they want in plain English. Classify intent, select appropriate action, execute without asking users to memorize commands.

Intent Categories

Feature Development

Triggers: "I want...", "Add...", "Build...", "Create...", "Enable...", "Implement...", "Make it so..." Route to: /full-feature skill Examples:

• "I want users to log in" → full-feature with auth-specialist
• "Add a checkout flow" → main agent implements via full-feature
• "Build a dashboard" → main agent implements (optionally with architect for complex features)

Bug Fixes

Triggers: "Fix...", "Broken...", "Error...", "Bug...", "Not working...", "Issue with...", "Problem with..." Route to: /quick-fix skill Examples:

• "Fix the login button" → quick-fix workflow
• "The checkout is broken" → quick-fix workflow
• "Error on the homepage" → quick-fix workflow

Code Review

Triggers: "Review...", "Check...", "Is this okay?", "Look at...", "Audit...", "Examine..." Route to: /review skill Examples:

• "Review my changes" → review
• "Check if this is secure" → review
• "Is this code good?" → code-reviewer agent

Security Audit

Triggers: "Secure?", "Safe?", "Vulnerable?", "Security...", "Hack...", "Exploit..." Route to: /review skill Examples:

• "Is this secure?" → review
• "Check for vulnerabilities" → review
• "Audit security" → review

Performance Optimization

Triggers: "Slow...", "Fast...", "Optimize...", "Speed up...", "Performance...", "Faster..." Route to: frontend-specialist agent Examples:

• "Make the page faster" → frontend-specialist
• "Optimize database queries" → backend-specialist
• "Speed up the app" → frontend-specialist

Testing

Triggers: "Test...", "Coverage...", "TDD...", "Verify...", "E2E...", "Unit test..." Route to: /test-coverage, test-writer, or test-writer Examples:

• "Add tests" → test-coverage
• "Write unit tests" → test-writer agent
• "TDD for login" → test-writer agent
• "E2E tests" → test-writer agent

Documentation

Triggers: "Document...", "README...", "API docs...", "Comment...", "Explain..." Route to: /update-docs or doc-updater agent Examples:

• "Update the README" → update-docs
• "Document the API" → backend-specialist agent
• "Add comments" → doc-updater agent

Refactoring

Triggers: "Clean up...", "Simplify...", "Dead code...", "Modernize...", "Refactor..." Route to: /refactor-clean or code-reviewer agent Examples:

• "Clean up the code" → refactor-clean
• "Remove dead code" → code-reviewer agent
• "Simplify this" → code-reviewer agent

Quality Improvements

Triggers: "Lint...", "Format...", "Type errors...", "TypeScript...", "ESLint..." Route to: /checkpoint or /type-check Examples:

• "Fix linting errors" → checkpoint
• "Fix type errors" → type-check
• "Format the code" → checkpoint

Deployment

Triggers: "Deploy...", "Release...", "Ship...", "Go live...", "Publish..." Route to: release workflow or devops-specialist Examples:

• "Deploy to production" → release workflow
• "Ship this feature" → release workflow

Component Creation

Triggers: "New component...", "Add a button/form/modal...", "Create component..." Route to: /new-component skill Examples:

• "Create a login form" → new-component
• "Add a modal component" → new-component

Database Changes

Triggers: "Migration...", "Schema...", "Add field...", "Database...", "SQL..." Route to: /create-migration or backend-specialist Examples:

• "Add a field to users table" → create-migration
• "Create new table" → backend-specialist agent

Context-Based Auto-Delegation

By Technology Mentioned

• Authentication/login → auth-specialist
• Database/schema → backend-specialist
• API/endpoints → backend-specialist
• GraphQL → realtime-specialist
• WebSocket/real-time → realtime-specialist
• Docker/containers → devops-specialist
• CI/CD/pipeline → devops-specialist
• Infrastructure → devops-specialist
• AI/LLM/RAG → realtime-specialist

By File Context

• .tsx, .jsx files → Frontend work
• /api routes → Backend work
• .sql, supabase/migrations/ → Database work
• .yml in .github → CI/CD work
• .test.ts → Testing work

By Modifiers

• "simple", "quick", "just" → Less ceremony, skip planning
• "robust", "production" → More thorough approach
• "urgent", "asap" → Skip optional steps
• "carefully", "thoroughly" → Comprehensive approach

Disambiguation Patterns

When intent is unclear, ask ONE clear question:

Vague Request Patterns:

• "Help with X" → Ask: add functionality / fix issue / improve / review?
• "Change X" → Ask: add features / fix bug / refactor / change appearance?
• "Update X" → Ask: add functionality / fix / modernize / update docs?
• "Work on X" → Ask: implement / debug / optimize / review?

Disambiguation Format:

What would you like to do?
1. Build something new (feature)
2. Fix something broken (bug)
3. Improve existing code (refactor)
4. Check code quality (review)

Confirmation Rules

Confirm Before

• Creating >3 new files
• Modifying >5 files
• Database migrations
• Deployment to production
• Deleting files/code
• Major refactoring

Skip Confirmation

• Code reviews (read-only)
• Linting/formatting
• Single file edits
• Running tests
• Reading/exploring code

Confirmation Format

I'll [action]. This will:
- [Change 1]
- [Change 2]
Proceed? (y/n)

Natural Language Translation

Technical → Plain English

• "Running /full-feature" → "Building your new feature..."
• "Delegating to agent" → "Analyzing/checking/designing..."
• "git commit" → "Saving changes..."
• "git push" → "Uploading changes..."
• "Creating PR" → "Preparing for review..."
• "Build failed" → "There's an error to fix..."
• "Tests passing" → "All checks passed!"

Avoid Jargon

Never say technical terms when simpler words work:

• API → "Connection to [service]"
• Endpoint → "Feature/function"
• Component → "Part of the page"
• Repository → "Project"
• Deploy → "Put online"
• Environment variable → "Setting"

Multi-Intent Detection

Sequential Intents

Detect multiple requests in one message:

• "Add login and update the README" → feature + docs
• "Fix the bug and add tests" → fix + testing
• "Review the code and deploy" → review + deployment

Parallel Execution

When intents are independent:

• "Check security and performance" → Run both in parallel
• "Add tests and lint" → Run both in parallel

Confidence Levels

High Confidence (Execute)

Clear, unambiguous intent with known pattern

Medium Confidence (Confirm)

Intent is clear but action is significant (>5 file changes, database changes)

Low Confidence (Clarify)

Vague or ambiguous intent, need more information

Edge Cases

No Clear Intent

"What should I do next?" → Suggest based on project state (tests missing, docs outdated, etc.)

Multiple Possible Interpretations

Offer options with context for each

Contradictory Requests

"Make it simple but add all features" → Clarify priority

Best Practices

DO

• Accept natural language
• Infer intent from context
• Execute when confident
• Confirm significant actions
• Use plain English in responses
• Adapt to user's technical level

DON'T

• Force users to memorize commands
• Use jargon unless user does
• Ask "which agent?" or "which command?"
• Require precise syntax
• Assume user knows technical terms
• Execute destructive actions without confirmation

Examples

Good Intent Matching

• User: "Users need to reset passwords" → Feature: auth-specialist
• User: "The cart total is wrong" → Bug: quick-fix
• User: "Make the dashboard load faster" → Optimize: frontend-specialist
• User: "Is this PR ready?" → Review: review
• User: "Ship it" → Deploy: release workflow

Handling Vagueness

• User: "Do something with auth" → Clarify: add feature / fix issue / review?
• User: "Update the app" → Clarify: new features / bug fixes / dependencies?
• User: "Fix it" → Clarify: what needs fixing?