Skip to main content
Generalmajesticlabs-dev

learn

Extract cross-session patterns from git history, PRs, and handoffs, then recommend artifacts (skill/rule/hook/agent) based on frequency thresholds. Use when consolidating recurring patterns into durable artifacts. For single-session reflection, use `reflect` instead.

Stars
39
Source
majesticlabs-dev/majestic-marketplace
Updated
2026-05-13
Slug
majesticlabs-dev--majestic-marketplace--learn
View on GitHubRaw SKILL.md

// install — copy + paste into any project

mkdir -p .claude/skills && curl -fsSL https://raw.githubusercontent.com/majesticlabs-dev/majestic-marketplace/HEAD/plugins/majestic-tools/skills/learn/SKILL.md -o .claude/skills/learn.md

Drops the SKILL.md into .claude/skills/learn.md. Works with Claude Code, Cursor, and any agent that loads SKILL.md files from .claude/skills/.

Learn

Extract patterns from git history, PRs, and handoffs to create durable artifacts (skills, rules, hooks).

For single-session reflection, use Skill("reflect") instead.

Methodology: Use compound-learnings skill for frequency thresholds and artifact categorization logic.

Data Sources

Source Command What to Extract
Git commits git log --oneline -100 Repeated fix types, patterns
PR descriptions gh pr list --state merged -L 20 --json title,body Lessons, decisions
Handoffs .agents/handoffs/*.md (main worktree) Patterns, What Worked/Failed
Existing learnings CLAUDE.md Key Learnings section Already encoded patterns

Process

Step 1: Locate Main Worktree

MAIN_WORKTREE=$(git worktree list --porcelain | grep "^worktree" | head -1 | cut -d' ' -f2)
HANDOFF_DIR="$MAIN_WORKTREE/.agents/handoffs"

Step 2: Gather Patterns

Git history:

git log --oneline -100 | cut -d' ' -f2-

Merged PRs (if gh available):

gh pr list --state merged -L 20 --json title,body 2>/dev/null

Handoff files:

ls "$HANDOFF_DIR"/*.md 2>/dev/null && cat "$HANDOFF_DIR"/*.md

Existing Key Learnings:

grep -A 50 "Key Learnings" CLAUDE.md 2>/dev/null

Step 3: Analyze Patterns

Apply compound-learnings skill methodology:

  1. Consolidate similar patterns by semantic meaning
  2. Apply frequency thresholds (skip 1x, note 2x, recommend 3+, strong recommend 4+)
  3. Categorize into artifact types using the decision tree

Step 4: Present Findings

## Compound Learnings Analysis

**Sources scanned:**
- Git commits: X
- PRs: Y
- Handoffs: Z

### Strong Signal (4+ occurrences)

| Pattern | Count | Artifact | Rationale |
|---------|-------|----------|-----------|
| ... | ... | ... | ... |

### Emerging Patterns (2-3 occurrences)

| Pattern | Count | Potential | Notes |
|---------|-------|-----------|-------|
| ... | ... | ... | ... |

### Recommended Actions

1. **[Artifact Type]**: `name`
   - Draft content: ...

Step 5: Create Artifacts

Use AskUserQuestion to confirm each recommendation:

  • "Create these artifacts?"

If approved, create using appropriate tools:

  • Rules → Edit CLAUDE.md or create lesson in .agents/lessons/
  • Skills → Create in appropriate plugin's skills directory
  • Hooks → Create in .claude/hooks/
  • Commands → Create in appropriate plugin's commands directory