Push and pull learnings

The self-learning plugin stores captured patterns in ~/.closedloop-ai/learnings/org-patterns.toon. Two commands keep these patterns synchronized across a team.

Prerequisites

• self-learning plugin installed.
• CLAUDE_ORG_ID exported in the environment.
• Access to the shared team repo that stores the organization's learnings.

Push

claude /self-learning:push-learnings

Exports your local patterns to the shared team repo. Echo prevention means patterns you originated in a project do not loop back into that project when you later pull.

Pull

claude /self-learning:pull-learnings

Imports the organization's patterns. New patterns merge into your local org-patterns.toon; existing patterns are updated if the organization copy is newer.

Process before pushing

If you have pending learnings from recent runs, process them first:

claude /self-learning:process-learnings

This classifies, validates, deduplicates, and merges pending learnings into org-patterns.toon, with low-performing patterns pruned per retention.yaml.

Inspect patterns

• ~/.closedloop-ai/learnings/org-patterns.toon — user-level pattern store.
• ~/.closedloop-ai/learnings/closedloop-learnings.json — ClosedLoop-specific learnings.

Patterns carry flags that tell you their status:

• [REVIEW] — success rate below 40%
• [STALE] — not applied in the last 10 iterations
• [UNTESTED] — no applications yet
• [PRUNE] — more than 20 applications with success rate below 40%

Goal stats

claude /self-learning:goal-stats

Shows pass rate, top patterns, and trends. Requires at least 5 runs.

When to push

• After a successful sprint where you captured valuable conventions.
• When a recurring pitfall deserves organizational attention.
• Periodically (weekly or biweekly) to keep the shared store current.

When to pull

• Before starting a new project.
• After joining a new team.
• When you notice your agents missing obvious conventions — a fresh pull may have what you need.

Why sharing patterns compounds

Pattern stores are the organization's institutional memory. Without them, each engineer repeats the same mistakes. With them, the second engineer to encounter a pitfall finds it already rejected by a pattern.

See Self-learning for the full mechanism.