learn

Manage project learnings in a markdown knowledge base across sessions.

Updated Jul 10, 2026
One-click install
npx skills add https://github.com/Kaleb-Rupe/aurora --skill learn-kaleb-rupe
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: learn
Source: https://github.com/Kaleb-Rupe/aurora/tree/main/claude/skills/learn
Command: npx skills add https://github.com/Kaleb-Rupe/aurora --skill learn-kaleb-rupe

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Teams lose hard-won insights between sessions: patterns get rediscovered, pitfalls get hit twice, and preferences are forgotten. This Skill maintains a persistent, human-readable knowledge base of project learnings so past discoveries stay searchable and actionable. ## Core Features & Use Cases - Review and Search: Show the 20 most recent learnings or search entries by keyword across keys, insights, sources, and files. - Prune and Validate: Detect stale entries whose referenced files no longer exist and resolve contradictions with user confirmation. - Export and Stats: Format learnings for CLAUDE.md, save a standalone export, or view summary statistics like counts, confidence averages, and top sources. - Use Case: After a debugging session uncovers that CPI calls require PDA signer seeds, record it as a pitfall so the next session recalls it instead of repeating the mistake. ## Quick Start Ask the assistant to show what we have learned on this project so far.

Frequently Asked Questions about learn

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

FAQPage Schema
How do I save project learnings across sessions?

Use the /learn add command to record a learning with a type, kebab-case key, insight, confidence score, and optional file references. Entries are appended to .superstack/learnings.md, and other skills can also write learnings automatically using the documented protocol.

How do I search past project learnings by keyword?

Run /learn search followed by your query to match against entry keys, insights, sources, and file paths case-insensitively. Results are grouped by type, and a message suggests broadening the query if nothing matches.

Can learnings be exported to CLAUDE.md?

Yes, /learn export formats all entries into a clean markdown section grouped by type with confidence scores. You choose whether to append it to CLAUDE.md, save a standalone export file, or print it without saving.

How are stale or contradictory learnings handled?

The /learn prune command checks whether files referenced by each entry still exist and flags contradictions between entries. Each flagged entry is presented with options to remove, keep, or update it, and no change is applied without confirmation.

What happens when a learning key already exists?

The new entry is appended below the existing one rather than replacing it, following an append-only model where the latest entry by date wins. The older entry is preserved as history.