Most Claude Code skill packs are thin prompt wrappers: they dump a pile of instructions into context and hope the model picks the right one. This project inverts that bet. Its center of gravity isn't the 66 domain skills at all — it's a context-engineering layer called 'Common Ground' that interrogates a codebase's hidden assumptions before a single line is written, while the skills themselves load lazily, only when a request actually touches their domain.
What Sets It Apart
- Lazy, context-aware activation — instead of pinning 66 skill definitions into the window at once, only the references a request needs are surfaced. That keeps the model focused and the token bill down on long sessions.
- Assumption surfacing before code — the 'Common Ground' step validates stack versions, conventions, and constraints up front, catching the mismatches that normally only show up at review time.
- Breadth that stays one toolset — 66 skills across 12 categories (languages, frameworks, infra/DevOps, APIs, testing, security, data/ML) let a multi-disciplinary task stay inside a single coherent system rather than swapping tools mid-flow.
- Atlassian in the loop — nine workflow commands wire into Jira and Confluence, so epic and ticket management happen in the same conversation rather than a separate tab.
Who It's For
Great fit if you're a full-stack or platform engineer juggling several stacks and want Claude Code to behave consistently across all of them — especially in an Atlassian-centric shop where the Jira/Confluence commands pay off. Look elsewhere if you live in one narrow stack, where the breadth is mostly overhead, or if you can't run a large community plugin you haven't fully audited: it's MIT-licensed and maintained by a single developer, so treat it like any third-party automation that touches your repository and issue tracker, and read what each skill actually does before trusting it.
