Skill Assistant Pro

Problem

In Agent Skill workflows, teams often face the same friction: it is hard to know which skill is worth installing, sources are scattered, quality varies, and security review before installation is inconsistent. Existing skills may also have weak descriptions, unclear trigger boundaries, and limited validation. skill-assistant turns discovery, selection, installation, creation, inspection, and diagnosis into a repeatable process.

How it works

The skill is route-based. It starts with prechecks in config/sources.yaml, performs configuration migration or first-run setup, then dispatches user intent into search, recommendation, install, creation, inspection, or diagnosis modules.

• Search: runs parallel recall across platforms and GitHub repositories, deduplicates results, normalizes quality signals, and ranks candidates by a unified score.
• Secure installation: performs multi-engine review before install, combining script scanning, external audit, and AI judgment; high-risk patterns can stop the flow, while low-trust sources require confirmation.
• Creation and quality checks: uses shared best-practice references and checklists to validate frontmatter, description, boundaries, security, and verifiability, with static, dynamic, hybrid, or blind evaluation modes.
• Diagnosis and optimization: analyzes instructions, constraints, and redundancy, supports iterative evaluation with regression protection, and outputs actionable improvement reports.

Boundaries

It is focused on Agent Skills, not general package managers such as npm, pip, or brew, and not IDE plugin installation. The diagnostic module can suggest changes or produce refactored files, but it does not overwrite originals automatically. Sources and API credentials must be configured; channel failures are reported through explicit fallback rules rather than silent retries.