Record, search, and prune per-project learnings in .claude/learnings.jsonl.
npx skills add https://github.com/shenxingy/Clade --skill learn-shenxingy
Please help me install this Agent Skill. Skill: learn Source: https://github.com/shenxingy/Clade/tree/main/configs/skills/learn Command: npx skills add https://github.com/shenxingy/Clade --skill learn-shenxingy
It helps you capture recurring project lessons (patterns, pitfalls, preferences, and tooling choices) so your future sessions can reuse what you learned instead of relearning it.
Ask your AI assistant to record a learning for your current repo by saying: " /learn Avoid this failure when wiring deployment for compatibility gaps, and use the verified workaround instead."
High-intent search queries and answers about installing and using this skill.
To save project learnings across coding sessions, record typed, confidence-scored notes into a .claude/learnings.jsonl file. This builds a reusable per-repo memory by capturing pitfalls, preferences, and architecture choices for future search and retrieval.
The best way to search past coding pitfalls is using search-by-overlap retrieval against saved JSONL entries. It matches your query against prior typed learnings to return the top relevant workarounds and prevents repeating the same failure modes.
Yes, you can prune stale architecture choices and preferences from project memory. The retention model uses a pruned tombstone flag to remove low-value or outdated entries from the JSONL file while preserving the historical record.
Capturing project memory requires the JSONL file format and a confidence scoring system. Typed entries must be saved into .claude/learnings.jsonl to ensure structured, long-term reuse of coding patterns and tooling choices across sessions.
You need a per-project knowledge capture system when you repeatedly encounter recurring build or deploy issues. Documenting these pitfalls and verified workarounds allows future coding sessions to reuse solutions instead of relearning the same failures.
Your assistant keeps relearning deployment compatibility gaps because it lacks a persistent project memory. Recording these pitfalls as typed entries with confidence scores in a JSONL file enables search-by-overlap retrieval to immediately surface past verified workarounds.
oaustegardInitialize Git repositories and commit changes for skill directories.
ZouZouImplement full-text search and multi-criteria filtering across courses and content.
kimasplundLog failures, corrections, and missing capabilities into structured learning memory.
JamesFincherRead, list, search, and create Markdown notes in an Obsidian vault.
