Skill Development Super Assistant

Problem

Building agent skills often fails at the same concrete points: unclear SKILL.md structure, triggers that do not match how users actually phrase requests, missing publish fields, and code edits that appear to do nothing because the wrong file, project, or render path was changed. This skill turns skill authoring and publishing into checkable steps rather than ad hoc prompt writing.

How it works

It supports three main workflows:
- Creation wizard: starts from the problem to solve and the exact phrases users will say, then drafts a SKILL.md skeleton, triggers, and FAQ.
- Publishing checks: reviews fields such as slug, displayName, version, changelog, and description, then suggests concrete improvements.
- Learning mode: explains trigger design, FAQ coverage, experience capture, and evaluation optimization using an “answer + explanation + reminder” style.

It also includes lessons from real projects, such as checking selector scope before CSS replacements, avoiding Chinese text in code logic and error messages, preventing BOM issues when writing text files, and searching the whole codebase after changing shared logic. When instructions are ambiguous, it asks for scope, target values, and reference points before making changes.

Scope and caveats

It is useful for teams preparing to publish AI agent skills, refactoring SKILL.md, or standardizing pre-change confirmation and post-project lessons. It is process-oriented and does not replace architecture design or platform-specific validation.