self-improve

Convert session corrections and failures into updated or new skill files.

Updated May 5, 2026
One-click install
npx skills add https://github.com/josippapez/ai-setup --skill self-improve-josippapez
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: self-improve
Source: https://github.com/josippapez/ai-setup/tree/main/claude/plugins/dev-core/skills/self-improve
Command: npx skills add https://github.com/josippapez/ai-setup --skill self-improve-josippapez

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When an AI assistant repeats a mistake or ignores an instruction, the fix often dies with the session. This Skill turns a correction or failure into a persistent rule or skill file so the next session starts with the lesson already learned. ## Core Features & Use Cases - Failure-to-file workflow: Names the failure, finds the owning file, decides whether the fix is a wording problem or a coverage gap, and writes the change. - Trigger-focused descriptions: Guides writing description and when_to_use frontmatter so the skill actually fires when it should, using the phrases users really type. - Cross-adapter mirroring: Ensures every change under claude/plugins/*/skills/ is mirrored to opencode/skills/ so the two adapters never drift. - Use Case: A user says "you keep committing without running the tests." This Skill locates the rule that should have caught it, sharpens its trigger description, mirrors the change to the opencode copy, and verifies the file reads correctly without session context. ## Quick Start Tell the assistant to turn the correction it just received into a skill or rule update and mirror it to the other adapter.

Frequently Asked Questions about self-improve

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

FAQPage Schema
How do I make an AI assistant remember a correction across sessions?

Turn the correction into a skill or rule file instead of relying on conversation context. Name the specific failure, find the file that should have caught it, write the fix as forward-looking guidance, and mirror it across your adapter directories.

Why did my skill not trigger when it should have?

A skill that exists but does not fire has a description problem. The description and when_to_use fields are the only text the model sees when deciding to load it, so they must open with what the skill does and include the exact phrases users type.

Should I add a rule or a skill for new guidance?

Use an always-on rule only for guidance that must hold on every turn, since rules load into every session and subagent. Default to a skill for guidance that applies to a kind of task and can be loaded on demand.

How do I keep Claude Code and OpenCode configurations in sync?

Every skill under claude/plugins/*/skills/ has a copy in opencode/skills/, and rules mirror to opencode/rules/. After editing one side, copy the change across immediately or the two adapters drift apart.

When should I not create a new skill file?

Do not create a new file when an existing skill is a near-miss you can sharpen, since two files covering the same behavior cause conflicts. Also skip writing anything if you cannot name a specific failure case from the session.