Use when asked to "analyze learnings", "find patterns", "what should become rules", "compound my learnings", or after accumulating 5+ session learnings. Analyzes session learnings to detect recurring patterns (3+ occurrences), categorizes them as rule/skill/hook candidates, and generates proposals for user approval.
Transform ephemeral session learnings into permanent, compounding capabilities.
This skill analyzes accumulated learnings from past sessions, detects recurring patterns (appearing in 3+ sessions), and generates proposals for new rules, skills, or hooks. User approval is REQUIRED before creating any permanent artifacts.
Core Principle: The system improves itself over time by learning from successes and failures.
# List learnings (most recent first)
ls -t $PROJECT_ROOT/.ring/cache/learnings/*.md 2>/dev/null | head -20
# Count total learnings
ls $PROJECT_ROOT/.ring/cache/learnings/*.md 2>/dev/null | wc -l
If fewer than 3 files: STOP. Insufficient data for pattern detection.
Read the most recent 5-10 files for analysis.
For each learnings file, extract entries from:
| Section Header | What to Extract |
|---|---|
## What Worked | Success patterns |
## What Failed | Anti-patterns (invert to rules) |
## Key Decisions | Design principles |
## Patterns | Direct pattern candidates |
Consolidate similar patterns before counting:
| Raw Patterns | Consolidated To |
|---|---|
| "Check artifacts before editing", "Verify outputs first", "Look before editing" | "Observe outputs before editing code" |
| "TDD approach", "Test-driven development", "Write tests first" | "Use TDD workflow" |
Use the most general formulation.
Signal thresholds:
| Occurrences | Action |
|---|---|
| 1 | Skip (insufficient signal) |
| 2 | Consider - present to user for information only |
| 3+ | Strong signal - recommend creating permanent artifact |
| 4+ | Definitely create |
Only patterns appearing in 3+ DIFFERENT sessions qualify for proposals.
For each qualifying pattern, determine artifact type:
Is it a sequence of commands/steps?
→ YES → SKILL (executable process)
→ NO ↓
Should it run automatically on an event?
→ YES → HOOK (SessionEnd, PostToolUse, etc.)
→ NO ↓
Is it "when X, do Y" or "never do X"?
→ YES → RULE (behavioral heuristic)
→ NO → Skip (not worth capturing)
Examples:
| Pattern | Type | Why |
|---|---|---|
| "Run tests before committing" | Hook (PreCommit) | Automatic gate |
| "Step-by-step debugging process" | Skill | Manual sequence |
| "Always use explicit file paths" | Rule | Behavioral heuristic |
Present each proposal in this format:
---
## Pattern: [Generalized Name]
**Signal:** [N] sessions ([list session IDs])
**Category:** [rule / skill / hook]
**Rationale:** [Why this artifact type, why worth creating]
**Draft Content:**
[Preview of what would be created]
**Target Location:** `.ring/generated/rules/[name].md` or `.ring/generated/skills/[name]/SKILL.md`
**Your Action:** [APPROVE] / [REJECT] / [MODIFY]
---
CRITICAL: DO NOT create any files without explicit user approval.
User must explicitly say:
NEVER assume approval. NEVER create without explicit consent.
Ensure the target directories exist:
mkdir -p $PROJECT_ROOT/.ring/generated/rules
mkdir -p $PROJECT_ROOT/.ring/generated/skills
mkdir -p $PROJECT_ROOT/.ring/generated/hooks
Create file at .ring/generated/rules/<slug>.md with:
Create directory .ring/generated/skills/<slug>/ with SKILL.md containing:
Note: Generated skills are project-local. To make them discoverable, add to your project's CLAUDE.md:
## Project-Specific Skills
See `.ring/generated/skills/` for learnings-based skills.
Create in .ring/generated/hooks/<name>.sh following Ring's hook patterns.
Register in project's .claude/hooks.json (not plugin's hooks.json).
After creating artifacts, move processed learnings:
# Create archive directory
mkdir -p $PROJECT_ROOT/.ring/cache/learnings/archived
# Move processed files (preserving them for reference)
mv $PROJECT_ROOT/.ring/cache/learnings/2025-*.md $PROJECT_ROOT/.ring/cache/learnings/archived/
## Compounding Complete
**Learnings Analyzed:** [N] sessions
**Patterns Found:** [M]
**Artifacts Created:** [K]
### Created:
- Rule: `explicit-paths.md` - Always use explicit file paths
- Skill: `systematic-debugging` - Step-by-step debugging workflow
### Skipped (insufficient signal):
- "Pattern X" (2 occurrences - below threshold)
### Rejected by User:
- "Pattern Y" - User chose not to create
**Your setup is now permanently improved.**
Before creating any artifact:
.ring/generated/rules/ and .ring/generated/skills/ first| Mistake | Why It's Wrong | Correct Approach |
|---|---|---|
| Creating without approval | Violates user consent | ALWAYS wait for explicit approval |
| Low-signal patterns | 1-2 occurrences is noise | Require 3+ session occurrences |
| Too specific | Won't apply elsewhere | Generalize to broader principle |
| Too vague | No actionable guidance | Include concrete DO/DON'T |
| Duplicate artifacts | Wastes space, confuses | Check existing rules/skills first |
Project-local data (per-project):
.ring/cache/learnings/*.md.ring/cache/proposals/pending.json.ring/cache/proposals/history.json.ring/generated/rules/<name>.md.ring/generated/skills/<name>/SKILL.md.ring/generated/hooks/<name>.shPlugin code (shared, read-only):
default/lib/compound_learnings/default/skills/compound-learnings/SKILL.mddefault/commands/compound-learnings.md