extract

This skill, invoked as /si:extract, turns a proven pattern or debugging solution into a standalone, portable Claude skill complete with SKILL.md, a README, and optional reference examples. It is part of a self-improving-agent toolkit and solves the problem of hard-won fixes being forgotten or trapped in one project: when a solution is recurring, non-obvious, broadly applicable, or user-flagged, it packages that knowledge into a reusable skill installable anywhere.

The single-file SKILL.md defines a seven-step workflow: identify the pattern (optionally searching the project's auto-memory directory with grep), determine scope with at most two clarifying questions, generate a compliant skill name, spawn a skill-extractor agent to create the files, structure the SKILL.md to a prescribed template, run quality gates, and report. It enforces naming rules — lowercase and hyphenated, and explicitly forbidding the reserved fragments "claude" and "anthropic" in favor of a cc- prefix — and its quality gates include a check that no project-specific hardcoded values such as paths, URLs, or credentials leak into the generated skill.

The target users are developers and teams building a personal or shared library of reusable skills from their debugging history. Typical use cases include packaging a Docker-on-Apple-Silicon fix, an API-client regeneration workflow, or any multi-step solution worth reusing across projects. The only shell activity described is a read-only grep over a local memory directory to find related notes; the skill creates local files and performs no destructive or network operations.