Skip to content

2.0.0 — 🪦 RIP summoning-the-user: A Eulogy in Semver

Choose a tag to compare

@britt britt released this 06 Jul 21:53
· 18 commits to main since this release
78e89c4

This one's a major version bump, which around here means "we killed a couple of skills and made everyone else go to therapy." 29 skills went under the knife. Two didn't make it. The rest came out more self-aware.

Breaking Changes

  • summoning-the-user is dead. It has been removed from the marketplace entirely. It served bravely, in the sense that it existed.
  • user-story-template has been absorbed into writing-user-stories — the INVEST checklist and Given-When-Then acceptance criteria now live there. If you had this installed standalone, update your plugin list; it's not coming back as its own thing.
  • Marketplace and bundle plugins jump to 2.0.0 (claude-code-skills, project-foundations) to reflect the removals above. 26 individually published skills bump to 1.1.0 for the refinements below (additive, non-breaking).

What's New

  • project-foundations bundle — a curated plugin bundling eight skills (setup, issue decomposition, user stories, verification plans, and the architecture/dependency/Mermaid diagram trio) for taking a project from "idea" to "diagrammed plan," single-sourced via symlinks so nothing is duplicated.
  • sgai-goal — new skill for composing GOAL.md files for SGAI-driven repos.
  • idea-to-design and roughly 20 other skills imported from the old LLPM collection, covering planning, stakeholder tracking, at-risk detection, FAQ generation, and more.
  • Individual skill installation — every skill can now be installed on its own instead of forcing the whole collection on you.

Under the Hood: The Skill Review Overhaul

Every one of the 29 skills got audited against skill-authoring best practices, because apparently we'd been winging it. Highlights:

  • Stripped unsupported frontmatter fields (instructions, tags) from 16 skills and folded their trigger phrases into actual discoverable "Use when..." descriptions.
  • Every skill that fetches GitHub or external data now states its tool prerequisites (gh CLI, MCP connectors) with concrete setup commands, instead of silently failing and letting you guess why.
  • Replaced leftover references to a defunct LLPM notes system and phantom claude.ai artifacts with real file conventions (docs/stakeholders/, requirements-<project>.md).
  • Fixed broken cross-references (britt/*, obra/*), a name mismatch in daily planning, and a scan tool in project-analysis that was, to put it gently, fictional.
  • Bundled TDD.rules.md directly into setting-up-a-project instead of fetching it from main at runtime.
  • Restored the architecture-diagramming skill's CRITICAL Mermaid syntax section (no-parentheses, matching-brackets) after it got accidentally collapsed to a one-liner — this guards against a real GitHub rendering bug, so it's staying emphatic.

Fixes

  • Root plugin.json was pinned to 1.3.0 while the marketplace entry said otherwise; since plugin.json wins at install time, this was silently freezing users on an older version. Fixed, and claude plugin validate . --strict now passes.
  • Corrected the documented install syntax everywhere (README + 28 skill pages): it's /plugin install <name>@britt, not /plugin install britt/<name>. The owner/repo form only works for marketplace add.

Docs

  • Documentation site synced with the overhaul: new pages for sgai-goal and project-foundations, both linked from the homepage, bundle-install instructions documented, and drift fixed across 19 skill pages that had gone stale.
  • Added semantic-release commit examples (fix/feat/perf/breaking) to the contributing docs, because someone asked "what does a breaking change commit actually look like" one too many times.

Full Changelog: 1.1.0...2.0.0