yrzhe/claude-skills/learn

Learn — Skill Auto-Extraction

Distill the current task's approach into a reusable skill, or improve an existing one.

When to Suggest

After completing a task, evaluate TWO conditions:

1. Complexity: 5+ tool calls used
2. Reuse potential: The approach applies to other projects/scenarios, not just this one

If both met, suggest: "This approach might be reusable — want to /learn it?" Never auto-execute. Only suggest.

Extraction Flow

1. Recall

Check if this knowledge already exists:

• Search existing skills for similar functionality
• If a memory system (supermemory, MEMORY.md, etc.) is available, search it too

2. Review

Scan the current conversation for the most recently completed task. Identify:

• Core workflow steps
• Key decisions and why they were made
• Non-obvious tricks or gotchas discovered
• Tools/libraries used and how

If /learn "description" was provided, focus on that specific aspect.

3. Dedup

Search ALL existing skills for similar functionality:

ls ~/.claude/skills/
grep -r "<relevant keywords>" ~/.claude/skills/*/SKILL.md

Three outcomes:

• Similar skill exists: Suggest /skill-improve <skill-name> instead. Do NOT create a duplicate.
• Similar learned skill exists: Suggest updating that skill instead of creating new.
• No match: Proceed to create new skill.

4. Assess Scope

Determine where this knowledge belongs:

• Cross-project reusable workflow → Create skill (continue to step 5)
• Project-specific pattern → Suggest saving to local project memory instead, stop here
• Single factual insight → Suggest saving to memory system instead, stop here

5. Generate

Create skill following skill-creator conventions:

File: ~/.claude/skills/<skill-name>/SKILL.md

Naming: use learned- prefix for auto-extracted skills (e.g., learned-pdf-merge).

Structure:

---
name: learned-<name>
description: <what it does and when to trigger — be specific>
---
 
# <Title>
 
<Concise steps. Only include what Claude doesn't already know.>

Rules:

• Under 200 lines — context window is shared
• Imperative form ("Run X", not "You should run X")
• No README, CHANGELOG, or extra docs
• Only add scripts/, references/, assets/ if genuinely needed

6. Confirm

Show the generated SKILL.md content to the user. Write to disk only after confirmation.

7. Log

Create or update the extraction log:

File: ~/.claude/skills/learn/learned/<skill-name>.md

---
skill: learned-<name>
created: <YYYY-MM-DD>
last_updated: <YYYY-MM-DD>
source: conversation
---
 
## Creation Record
-**Date**: <YYYY-MM-DD>
-**Task**: <what triggered the extraction>
-**Reason**: <why this is reusable>

Also update ~/.claude/skills/learn/learned/README.md index with a one-line entry.

8. Memory Sync (Optional)

If a memory system is available (supermemory, MEMORY.md, etc.) and the extraction revealed cross-project universal insights (API quirks, tool gotchas, non-obvious behaviors), save a concise one-line version.

Do NOT duplicate the full skill content into memory. Only save atomic insights that stand alone.