I build tools that coding agents can run on a machine you control, then I send the hard fixes back upstream. Python, TypeScript, MCP. Building in public.
/reimagine-it · book-guide-mcp · research · sponsor
| Project | What it is |
|---|---|
| reimagine-it · v2.3 | Claude Code plugin and Cursor / Codex / Copilot / Gemini CLI skill. Any file (webpage · PDF · doc · deck · mobi · epub · code · protocol · prose). A content-aware design leap in one command. Ships an inline WebGL2 shader hero, a craft floor (:focus-visible, prefers-reduced-motion decompose, compositor-only motion, INP < 200 ms), an open creative engine across seven axes, and a first-class infographic pack (paper poster: common-scale timeline, ISOTYPE counts, lossless table — not a dashboard). Ten tested case studies, all rendered locally from real .html files in the repo. No CDN, no paid API, no third-party service. |
| book-guide-mcp | Playbooks and tutors your agents run locally (Cursor, Claude, VS Code, and others). Citations from books you own. No API keys for the core loop. |
| research | Dated public notes on local-first agents, MCP, and the failure modes that show up when tools actually have to run. |
/reimagine-it — same source, same command, three completely different reader registers. Every visual is a real .html file in the repo.
v2.3 · /reimagine-it infographic · statistical poster, not a dashboard.
Hard, reviewable work — not drive-by typo farms.
| Repository | Change | Status |
|---|---|---|
| simonw/llm | Accept JSON Schema type names in schema_dsl |
Open — #1612 |
| simonw/llm | Set conversation_id correctly in llm logs --data-ids |
Open — #1613 |
| pypa/hatch | Build a wheel from the sdist when both targets are requested | Open — #2386 |
AI agents help me move faster. I own every upstream diff I open.
GitHub is the right place: @Kayforkind