compound-learning

name	compound-learning
description	Extract and store learnings from completed work to make the next task easier. Use after any significant deliverable — demos, emails, code reviews, research. Triggers: 'what did we learn', 'capture what worked', 'extract learnings', 'that didn't go well', after high-stakes work.
metadata	{"openclaw":{"emoji":"🧠"}}

Compound Learning

Use this skill to turn completed work into reusable knowledge. The goal is simple: make future work faster and better by storing concrete, searchable lessons.

Setup

Canonical script location:

SKILL_DIR="$HOME/.openclaw/skills/compound-learning-openclaw-skill"

Optional compatibility wrappers for legacy callers:

bash "$SKILL_DIR/scripts/install-compat.sh"

Environment overrides:

• COMPOUND_LEARNINGS_DIR (default ~/.openclaw/learnings)
• COMPOUND_LEARNINGS_SCRIPTS_DIR (optional script path override)

When to Trigger

Trigger after any meaningful output:

• feature delivery
• incident response
• code review
• sales or customer conversations
• research synthesis
• operational runbooks

Natural prompts: "what did we learn", "capture this", "what should future me remember", "why did this fail".

Process

1. Analyze outcome through lenses:
   • pattern: repeatable behavior that worked or failed
   • template: reusable structure
   • preference: stakeholder/user style preference
   • failure: preventable miss and guardrail
   • insight: non-obvious connection
2. Produce 2-5 discrete insights max.
3. Store each insight as its own file with store-insight.sh.
4. Suggest search hooks for next similar task.

Store Example

cat > /tmp/insight-body.md << 'EOF'
## Context
[What happened]

## The Learning
[What matters and why]

## When to Apply
[Specific trigger conditions]

## Related
[Optional related files]
EOF

SKILL_DIR="$HOME/.openclaw/skills/compound-learning-openclaw-skill"

bash "$SKILL_DIR/scripts/store-insight.sh" \
  --type pattern \
  --category operations \
  --task "Release deploy with rollback drills" \
  --outcome success \
  --tags "release,rollback,runbook" \
  --takeaway "Pre-deploy rollback rehearsal cuts incident recovery time." \
  --title "Rehearse Rollback Before High-Risk Deploys" \
  --content-file /tmp/insight-body.md

Optional strict duplicate protection:

bash "$SKILL_DIR/scripts/store-insight.sh" ... --reject-duplicates

Search Before Similar Work

SKILL_DIR="$HOME/.openclaw/skills/compound-learning-openclaw-skill"

bash "$SKILL_DIR/scripts/search-learnings.sh" release
bash "$SKILL_DIR/scripts/search-learnings.sh" --category code --type failure test-flake
bash "$SKILL_DIR/scripts/search-learnings.sh" --limit 10

Anti-Patterns

• Do not store vague statements.
• Do not combine multiple insights into one file.
• Do not skip failure learnings.
• Do not exceed 2-5 insights per task.

Scripts

Script	Purpose
extract-learnings.sh	Prompt scaffold for extracting insights
store-insight.sh	Persist one insight file and update index
search-learnings.sh	Search insights by keyword/category/type
install-compat.sh	Create/update legacy wrapper path symlinks

See references/insight-format.md for file schema and examples.