Paperthin 1.0.0

Type /paperthin, then your dream

Paperthin 1.0 is the skillbook for recursive self-improvement (RSI): 45 plain-Markdown skills an agent uses to check, cut and redo its own work until it holds up, on Codex, Claude Code and any agent that reads skills. You no longer need to know the catalog: type /paperthin and what you want, and it picks the skills, shows you their order and runs them, pausing where only you can act. Codex and Claude Code now install it as a plugin that keeps itself current, every other agent gets a daily background update, and a 0.x install moves over with one /re0-upgrade.

New

• One command. /paperthin hands your request to re0-workflow, which runs the right skills in order. re0-tutorial teaches one skill in three levels, practised on your own work.
• Think before building. elon questions every requirement and active commitment in Musk's five steps and proposes what to delete, for your approval. scopechk tells you whether a change belongs in a project, judged by what that project says it is for. crossthink borrows proven fixes from other fields when every option looks the same. multithink weighs several independent opinions by the evidence each cites, not by how many agree.
• Learn from what went wrong. backprop takes a correction or a failed run and writes the lesson into the rules, docs and skills that caused it. re0-memo now generalizes each lesson to the whole class of mistake, and catchup can report where a project stands against a goal.
• Cleaner code, not just cleaner prose. debloat now trims code too, with tests held green, and keeps a quickstart to one action and one link. re0-style fixes a change to the style its repository already uses. demock finds fake or placeholder code left on a path real users run. deid finds one person's or company's data that leaked into a shared repository. dedot cleans up Korean middle-dot lists, and dedash now rewrites only the dashes that sit inside sentences.
• Repositories and history. re0-readme keeps the README for people, AGENTS.md for agents and the package files on one name, and checks the GitHub page too. re0-import files new notes into the document that already covers them. re0-decide turns an open decision into a page that computes the answer from your own choices. re0-squash republishes a repository as one clean commit. ssotize-local keeps one real copy of a duplicated file and links the rest. re0-git can now move a forgotten commit to when the work was done. re0-merge lands a contribution as it is, with fixes added on top, or rebuilt, always keeping the contributor's credit. re0-work archives a project and restarts it layer by layer.
• Long runs. re0-watch watches a long agent job and tells you when it stalls.
• Repeatable steps run as scripts. Twenty skills now do their mechanical part with a tested script and keep only the judgment in the text, so a rerun gives the same result.

Also

• Renamed and retired. readchk is now readback, reorder is now re0-order, feynman is now whychk, and prism is retired in favour of multithink. /re0-upgrade swaps the old names in an existing install.
• No more start-up notice. The message that appeared at session start is gone; /re0-upgrade removes it and switches a 0.x install to self-updating.
• A new look. A new banner, icon and social card, in a black-and-white and a ceramic version. The README is rewritten: it opens with what Paperthin is and who builds it, The Proof shows NVIDIA's short-context results and the skills built on them, Quickstart takes 30 seconds, and the skill list puts the ones you will use most first. The ten translations moved from docs/readme/ to assets/i18n/, so update any bookmark.
• Web chats. Give ChatGPT or Claude on the web the address of llms-full.txt, and it reads the README and every skill at once.
• Publishing. Each npm release now shows where and how it was built (npm provenance), and the repository holds no npm token.
• Every translated README lists the skills in the same order as the English one (#18, @sergiobuilds).

The catalog (45 skills)

• depth/: re0, debloat, re0-order, detool, re0-style, dedash, dedot, readback, aim, modelchk, elon, scopechk, hate, macrothink, crossthink, re0-decide, whychk, autobahn, sip, shower, factchk, mandela, demock, deid, re0-git, re0-release, re0-merge, re0-squash
• breadth/: ssotize, ssotize-local, re0-readme, re0-import, re0-upgrade
• coil/: paperthin, re0-workflow, re0-memo, backprop, re0-work, catchup, nba, re0-plan, re0-loop, re0-watch, re0-tutorial
• mesh/: multithink

Install

Pick one route per agent; installing two loads every skill twice. Coming from an older install, run /re0-upgrade once in a fresh session.

Codex (updates itself at startup):

codex plugin marketplace add LilMGenius/paperthin && codex plugin add paperthin@paperthin

Claude Code (then, once, in Claude Code, run /plugin, open Marketplaces, select paperthin and choose Enable auto-update):

claude plugin marketplace add LilMGenius/paperthin && claude plugin install paperthin@paperthin

Every other agent (schedule.mjs install registers a per-user job, unelevated and windowless, that updates every skill and adds new ones at local 00:00):

npx skills@latest add LilMGenius/paperthin --global --agent '*' && node "$HOME/.agents/skills/re0-upgrade/scripts/schedule/schedule.mjs" install