learn

Extracts novel session learnings and persists them to the correct memory layer.

1|Updated Feb 8, 2026
One-click install
npx skills add https://github.com/blamechris/skill-templates --skill learn-blamechris
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: learn
Source: https://github.com/blamechris/skill-templates/tree/main/.claude/skills/learn
Command: npx skills add https://github.com/blamechris/skill-templates --skill learn-blamechris

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Valuable insights discovered during a coding session are usually lost when the session ends, forcing you to rediscover the same constraints and pitfalls later. This Skill captures only genuinely novel, behavior-changing learnings and routes them to the right memory layer (CLAUDE.md, .claude/rules/, CLAUDE.local.md, or auto memory) without bloating or silently mutating your project conventions. ## Core Features & Use Cases - Behavioral gate check: Filters out topic recall, value judgments, and reinforcement so most sessions correctly produce "nothing to persist" instead of low-quality memory noise. - Deduplication and conflict detection: Checks candidates against CLAUDE.md, .claude/rules/, and CLAUDE.local.md, dropping duplicates and flagging contradictions for user decision. - Governance-safe routing: Proposes (never auto-applies) changes to CLAUDE.md and rules files, while applying low-risk entries to CLAUDE.local.md or auto memory directly. - Use Case: After debugging a silent WebSocket failure caused by a dropped tunnel, run the skill to persist "check tunnel connectivity before debugging WS issues" as a proposed CLAUDE.md addition, with the exact diff shown for your approval. ## Quick Start After a session where you discovered something non-obvious, ask the AI to run the learn skill to capture and persist the insight to the appropriate memory file.

Frequently Asked Questions about learn

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I save learnings from a Claude Code session?▼

Run the learn skill at the end of a session. It extracts up to three behavior-changing insights, deduplicates them against CLAUDE.md and .claude/rules/, and routes each to the correct memory layer, asking for your approval before touching governance files.

How do I capture a specific insight directly without full extraction?▼

Pass the complete insight as an argument, for example a statement about a bash portability constraint. The skill skips discovery and goes straight to deduplication and placement, but approval gates for CLAUDE.md and rules files still apply.

Does the learn skill modify CLAUDE.md automatically?▼

No. Changes to CLAUDE.md and .claude/rules/ are always proposed with the exact text to add and require explicit per-invocation user approval. Only CLAUDE.local.md and auto memory entries are applied directly.

Why does the learn skill say nothing to persist?▼

That is the expected outcome for routine sessions. The skill applies a behavioral test and only persists insights that would change future behavior, filtering out topic recall, value judgments, and knowledge already documented in existing memory files.

What happens when a new learning contradicts an existing rule?▼

The skill flags it as a conflict, showing both the existing text and the new finding side by side. You decide whether to keep the existing entry, replace it, or keep both; the skill never edits existing lines on its own.