Extract and Save

Master skill that analyzes a conversation and automatically extracts all patterns into your system-wide configuration.

How It Works

When you ask to extract and save:

1. Scan the conversation for extractable patterns
2. Categorize each pattern (skill, rule, subagent, hook, memory)
3. Create files in appropriate ~/.claude/ locations
4. Report what was extracted and where
5. Suggest usage in future projects

What I Look For

Skills (Repeating Workflows)

Signals:

• Same decision-making process used multiple times
• "I always do X when Y"
• Workflow with clear steps
• Can be applied in other projects

Rules (Path-Specific Requirements)

Signals:

• "In [directory], I always..."
• "For [file type], requirement is..."
• Requirement applies to a subset of files
• Same constraint in different contexts

Subagents (Specialist Expertise)

Signals:

• "Have the [role] review this"
• Task needs specific tools or perspective
• Same specialist role appears 2+ times
• Delegation to expertise

Hooks (Automated Validation)

Signals:

• "I wish this would automatically..."
• Manual check repeated multiple times
• "I always do X after doing Y"
• Validation/formatting that's repetitive

Memory (Personal Principles)

Signals:

• "I believe...", "I prefer...", "I always..."
• Personal philosophy or principle
• Applies to all projects
• Value or tradeoff decision

Prioritization

High Priority (extract these):

• User explicitly asks for extraction
• Same workflow appears 3+ times in conversation
• Personal principle stated 2+ times
• User expresses frustration with manual task

Medium Priority (suggest these):

• New workflow discovered
• Principle stated once but clearly important
• Path-specific requirement identified

Low Priority (probably skip):

• One-off solutions
• Temporary workarounds
• Project-specific quirks
• Edge cases

Extraction Criteria

Only extract if:

1. ✓ It's actually reusable (appears in multiple contexts or clearly would be)
2. ✓ It has a clear name and purpose
3. ✓ It's not redundant with existing skills/rules
4. ✓ It's well-documented enough to use later

Don't extract if:

1. ✗ It's a one-time solution for this project
2. ✗ It's too specific (not generalizable)
3. ✗ It would just replicate existing skills
4. ✗ Documentation would be unclear

Output Format

## Extraction Summary

Total patterns identified: X
Extracted: Y
Skipped: Z

---

### ✓ Extracted: Skills (Y/Z)

1. **python-async-patterns**
   Location: ~/.claude/skills/python-async-patterns/SKILL.md
   Based on: [lines from conversation]
   Use in future: /python-async-patterns

### ✓ Extracted: Rules (Y/Z)

1. **typescript-services**
   Location: ~/.claude/rules/typescript-services.md
   Applies to: services/**/*.ts
   Based on: [conversation context]

### ✓ Extracted: Subagents (Y/Z)

1. **code-reviewer**
   Location: ~/.claude/agents/code-reviewer/
   Memory: ~/.claude/agent-memory/code-reviewer/MEMORY.md
   Use in future: Have the code-reviewer review this

### ✓ Extracted: Hooks (Y/Z)

1. **auto-format-python**
   Location: ~/.claude/settings.json
   Trigger: PostToolUse on .py files
   Action: black --quiet

### ✓ Extracted: Memory (Y/Z)

1. **Personal principle: Prefer simplicity**
   Location: ~/.claude/CLAUDE.md
   Section: Core Philosophy > Code Clarity Over Cleverness

---

### ⊘ Skipped Patterns (Could extract later)

1. "How to choose between Redis and in-memory cache"
   Reason: Too context-specific, only mentioned in this project

2. "Error message formatting"
   Reason: Covered by existing backend-standards rule

---

## Next Steps

These extractions are now available across all projects:
- Run `/python-async-patterns` to use the skill
- Rules auto-apply to matching paths in projects
- Delegate with "Have the code-reviewer check this"
- Hooks auto-run on matching files
- Personal philosophy is loaded in all sessions

Execution Steps

1. Read the full conversation to understand context
2. Extract each pattern type using the appropriate /extract-as-* skill
3. Check for duplicates against existing ~/.claude/ content
4. Create files with Clear names and documentation
5. Test patterns if possible (e.g., verify hooks run)
6. Document what was extracted and why
7. Report the summary to the user

When to Use This

Best time: End of a productive session where you:

• Discovered a new workflow
• Stated preferences or principles
• Solved a recurring problem
• Built something you'd use again

Invoke with:

/extract-and-save

or

> Analyze this conversation and extract all patterns to ~/.claude/

Tips

• Run this monthly after multiple projects
• Dedicate a 10-minute session just to extraction
• Review what was extracted — it shows patterns in your thinking
• If something extracted doesn't feel right, you can delete it
• Your ~/.claude/ directory grows as you work — it becomes your personal framework

Learning Over Time

As you extract more:

1. Your personal framework grows
2. Future sessions start faster (less exploration needed)
3. You recognize patterns earlier
4. Skills/rules become more refined
5. Your Claude Code "knows you" better