GitHub repository setup, deliberative backlog refinement, quick PBI delivery, ordered step execution, issue delivery with researched priority and effort, user-facing wiring and usability/accessibility reviews, repository learning and teach-back, scanner-and-agent GitHub PR review and finding fixes, guarded completion, maintenance, and marketplace release. Ships 23 skills and 14 agents.
23 skills · 14 agents
Advisory structural diagnostics with an optional plaintext readability boundary. Current-state reports and test-quality evidence remain read-only, and the plugin registers no hook. Ships 1 skill: /quality-refactor (responsibility-first refactoring with a zero-dependency scanner).
1 skill · 1 MCP server
Read-only MCP source navigation for one project. Provides search, focus, semantic relation following, history, and workspace status without project-editing tools.
1 MCP server
Read-only retrieval from Markdown in an externally configured knowledge-base vault. Browse chapters, sections, and source-linked entries.
1 MCP server
A Stop hook that checks the final message of each turn reads on its own for an owner who reads nothing else: no restating recent final messages, no hard-to-read phrasing, and no names or numbers left unexplained. A failing message goes back for a rewrite, at most twice per turn. Fails open with a visible warning. Claude Code only. No MCP server, no skills.
hooks
Two output styles. Natural: plain prose, common words, short sentences, active voice, no filler. Neurodivergent: tuned for an AuDHD reader with demand avoidance - choice-first, memoryless, no obligations, interest-first ordering. Both keep Claude Code's built-in coding behavior. No MCP server, no skills, no hooks.