Pattern Extraction

Pattern Extraction Skill

Extract reusable patterns from conversations and code, determine appropriate artifact types, and suggest organization locations.

When to Use

• Use when extracting patterns from conversations
• Use when identifying reusable patterns in code
• Use when determining what artifacts to create
• Use when organizing patterns into categories

Instructions

Pattern Extraction Process

1. Identify patterns: Find recurring themes, instructions, or workflows
2. Abstract patterns: Extract generalizable patterns from specific instances
3. Determine type: Decide if pattern should be rule, skill, command, or subagent
4. Suggest location: Recommend where artifact should be placed
5. Document pattern: Document the pattern clearly

Pattern Types

Instruction Patterns

• Repeated instructions or corrections
• User preferences and style choices
• Domain-specific conventions
• Team standards

Workflow Patterns

• Common sequences of actions
• Standardized processes
• Multi-step procedures
• Recurring task flows

Error Patterns

• Systematic errors and corrections
• Common mistakes and fixes
• Prevention strategies
• Error handling approaches

Context Patterns

• Context-specific behaviors
• File-type specific patterns
• Domain-specific knowledge
• Environment-specific configurations

Hook Patterns (Cursor agent loop)

• Gating: What the agent is allowed or forbidden to do (shell commands, file reads, MCP calls) → beforeShellExecution, beforeReadFile, beforeMCPExecution, etc.
• Post-edit automation: Run something after every file edit (format, lint) → afterFileEdit, afterTabFileEdit
• Auditing: Log or observe agent actions → postToolUse, afterShellExecution, afterMCPExecution
• Session context: Inject env or context at conversation start → sessionStart
• Tab vs Agent policy: Different rules for inline Tab vs Agent → Tab-specific hooks (beforeTabFileRead, afterTabFileEdit)

When a pattern fits hook behavior, suggest hooks as the outcome (with event(s) and command-based vs prompt-based). See Cursor Hooks.

Artifact Type Determination

Use decision framework:

• Rule: Persistent guidance, domain knowledge, standards
• Skill: Domain knowledge with scripts, automation needs
• Command: Manual workflows, checklists, step-by-step
• Subagent: Complex tasks, context isolation, parallel work
• Hooks: Behavior that must run automatically in the agent loop (gate, format after edit, audit, inject context)—configure via .cursor/hooks.json and scripts

Organization Suggestions

• Category selection: Choose appropriate category (meta, organization, etc.)
• Naming: Suggest name following conventions
• Location: Recommend full path for artifact
• Relationships: Identify related artifacts

Examples

Example 1: Instruction Pattern

Pattern: "Always use async/await instead of promises"

Extraction:

• Type: Rule
• Location: cursor/rules/organization/async-await.mdc
• Content: Rule enforcing async/await usage
• Related: JavaScript/TypeScript rules

Example 2: Workflow Pattern

Pattern: Standard code review process with checklist

Extraction:

• Type: Command
• Location: cursor/commands/meta/code-review-checklist.md
• Content: Step-by-step review checklist
• Related: Review-related artifacts

Best Practices

• Extract generalizable patterns, not one-off instances
• Consider frequency and significance
• Validate pattern worth creating artifact
• Document pattern clearly
• Suggest appropriate organization

Related Artifacts

• Conversation Analysis Skill
• Artifact Creation Skill
• Artifact Creation Rule
• Cursor Hooks – When to recommend hooks (agent-loop automation, gating, auditing)
• Pattern Extraction Research