self-improve

Crystallize session learnings into tracked lesson memory with approval-gated commits.

Updated Mar 22, 2026
One-click install
npx skills add https://github.com/diazMelgarejo/orama-system --skill self-improve-diazmelgarejo
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: self-improve
Source: https://github.com/diazMelgarejo/orama-system/tree/main/bin/orama-system/skills/self-improve
Command: npx skills add https://github.com/diazMelgarejo/orama-system --skill self-improve-diazmelgarejo

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Valuable insights discovered during a work session—confirmed facts, resolved errors, architectural decisions—are usually lost when the session ends. This Skill captures those learnings at session end and persists them into a tracked lesson memory so future sessions can reuse them. ## Core Features & Use Cases - Session Learning Extraction: Identifies new facts, patterns, decisions, resolved errors, and skill updates from the just-finished session and formats them as dated entries. - Approval-Gated Writes: Proposes a minimal, additive-only diff and waits for explicit user approval (approve/edit/skip) before writing or committing anything. - Idempotent Capture: Checks for duplicate topics, never overwrites existing entries, and includes a version guard to skip redundant writes. - Use Case: After finishing a debugging session that uncovered a root cause and fix pattern, invoke the skill to record the lesson into the PT .agent memory and commit it to git for future reuse. ## Quick Start Ask the AI to run the self-improve skill at the end of your session to review what was learned and propose a lesson entry for your approval.

Frequently Asked Questions about self-improve

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I save session learnings for future AI sessions?

Invoke the self-improve skill at session end or type /self-improve at any checkpoint. It extracts facts, patterns, decisions, and resolved errors from the session, proposes a dated lesson entry, and writes it to tracked lesson memory only after you approve.

How does the self-improve skill prevent duplicate lessons?

It checks existing lesson memory for duplicate topics before proposing additions and only appends new entries. A version guard also skips writes when the installed skill version is already current, and only the last run's learnings are committed per session.

Does the skill write or commit changes automatically?

No. It presents a proposed diff and stops at a hard approval gate with approve, edit, or skip options. Nothing is written or committed until you explicitly approve, and pushes to remote remain human-gated.

When should the self-improve skill not be triggered?

It should not run when the session was exploratory and produced no stable facts, or when /self-improve was already executed in the same session. It is intended for sessions that concluded meaningful work with reusable outcomes.

What kinds of learnings does the skill capture?

It captures confirmed facts such as endpoints and configs, reusable patterns, architectural decisions with rationale, error root causes with fix patterns, skill updates, and install or validation lessons, each formatted as a dated concise entry.