Skip to content

Releases: LilMGenius/paperthin

Paperthin 0.17.5

Choose a tag to compare

@github-actions github-actions released this 06 Sep 18:57
v0.17.5
6f706e3

Consent for account actions

re0-upgrade uses a structured choice before adding a GitHub star, after reporting the upgrade and discovery-notice results. It waits for a submitted choice; skipping leaves the completed upgrade intact. Unattended runs and hosts without a structured question tool skip starring.

Also

Contributor instructions live in AGENTS.md, with CLAUDE.md pointing to the shared guide. Authoring links follow that home; the existing casebook preservation policy remains unchanged.

Install

npx skills@latest add LilMGenius/paperthin --global --agent '*'

Paperthin 0.17.4

Choose a tag to compare

@github-actions github-actions released this 18 Aug 21:19
v0.17.4
3bca079

A release nobody finished by hand

The push is the whole manual step. Running what the release workflow owns — publishing, titling the release page — produces something that looks like a release and quietly isn't: it drops the signed provenance attestation, or takes the raw tag as its title, and nothing fails. re0-release now refuses it as a rule and again at the moment a failed run tempts a rescue, where fixing the cause or rolling the version forward is the only way through.

Install

npx skills@latest add LilMGenius/paperthin --global --agent '*'

Paperthin 0.17.3

Choose a tag to compare

@LilMGenius LilMGenius released this 18 Aug 20:47
v0.17.3
b7f9f33

A commit body has a shape, even in a repo that never named one

commit-economy decided which changes earned a bullet but never what the body looked like, so a log with no settled convention let prose paragraphs through — explaining where a log should list. It now falls back to a subject line, a blank line, and one - bullet per change on a single unwrapped line. The fallback ranks below the existing resolution order, so a repo that already documents or demonstrates a convention still keeps its own.

Install

npx skills@latest add LilMGenius/paperthin --global --agent '*'

Paperthin 0.17.2

Choose a tag to compare

@github-actions github-actions released this 22 Jul 14:40
v0.17.2
13f98db

The de- family sits together

reorder, detool, dedash, and debloat now run contiguous across every roster surface and README, so the shared prefix scans as the one cleanup family it is. debloat takes its seat in the READMEs' Problem removes-list, and the checklist now says when a skill earns that line — or a Fixes entry — at all: only when it carries the thesis, both sections curated.

Install

npx skills@latest add LilMGenius/paperthin --global --agent '*'

Paperthin 0.17.1

Choose a tag to compare

@github-actions github-actions released this 21 Jul 08:11
v0.17.1
cdbb4a1

Adding a skill now includes putting it in order

The shipping checklist and re0-release now name the step that was always there in practice: a new skill or rename takes its place in the roster's one logical order across every surface — the README and its translations, plugin.json, scripts/catalog.cjs, and re0-upgrade's catalog — kept in lockstep with reorder. The set was already CI-guarded; the order is now a written step, not a habit.

Install

npx skills@latest add LilMGenius/paperthin --global --agent '*'

Paperthin 0.17.0

Choose a tag to compare

@github-actions github-actions released this 20 Jul 22:32
v0.17.0
05c92e8

A reflex for the bloat

debloat joins the cleanup skills: point it at an artifact that has accreted into padding, walls, and restated rules, and it compresses to the load-bearing density — cutting words, never a rule. It is the compression half of paperthin's Restraint triad that ssotize (duplication) leaves, and it is distinct from re0's full rewrite — a density pass on prose that is otherwise correct and current.

Also

re0-plan now always gates a cycle on keeping CLAUDE.md and README aligned, with their roles pinned: CLAUDE.md is dual (agent instructions and the contributor guide), README is human-only. And CLAUDE.md itself got the debloat treatment — its shipping, releasing, philosophy, and invocation sections lost their accretion, every rule kept once.

The catalog (28 skills)

depth/ gains debloat, the twelfth user-invoked skill.

Install

npx skills@latest add LilMGenius/paperthin --global --agent '*'

Paperthin 0.16.2

Choose a tag to compare

@github-actions github-actions released this 20 Jul 21:20
v0.16.2
07c1a4d

The discovery notice keeps its state in one place

The session-start discovery notice keeps its runtime state under ~/.re0/, defined once in catalog.cjs and imported by both the command-hook and OpenCode adapters instead of hardcoded in each. A CI drift-guard holds that single definition in place: it fails if the home is reconstructed anywhere else, or the retired path name resurfaces in the shipped runtime or re0-upgrade's wiring.

Install

npx skills@latest add LilMGenius/paperthin --global --agent '*'

Paperthin 0.16.1

Choose a tag to compare

@github-actions github-actions released this 19 Jul 09:08
v0.16.1
85e5bf7

An install that only looks installed

re0-upgrade now catches a shadow install — a git clone that Claude Code loads and runs, but the npx skills CLI never registered. Before it classifies anything, it checks the managed footprint under ~/.agents/skills/: if paperthin is loaded yet unmanaged, it stops and offers a clean managed reinstall or keeping the clone, instead of misreading the whole catalog as missing and installing a duplicate.

Also

The minor/patch boundary is now decided by kind, not size — was the old behavior wrong (a fix) or merely narrower (a new capability)? — applied by re0-release and re0-plan.

Install

npx skills@latest add LilMGenius/paperthin --global --agent '*'

Paperthin 0.16.0

Choose a tag to compare

@github-actions github-actions released this 19 Jul 05:29
v0.16.0
8593224

modelchk sizes how hard to think, not just which model

modelchk now recommends two coordinates from one risk read: the cheapest sufficient capability tier and the reasoning effort within it, on a neutral glance / measured / thorough / exhaustive scale that binds to whatever levels the active model exposes rather than any vendor's names. Effort is orthogonal to tier — a bounded-but-fiddly task can be fast and thorough, a quick expert call frontier and glance — and a boundary rule keeps the skill to the run's cognitive spend, not the whole run configuration.

Install

npx skills@latest add LilMGenius/paperthin --global --agent '*'

Paperthin 0.15.1

Choose a tag to compare

@github-actions github-actions released this 18 Jul 02:54
v0.15.1
a94aea8

The map and the tour catch up to the shipped suite

The four-quadrant map now lights mesh/ like the other three: prism made it a live quadrant, so it is no longer drawn as reserved. The README's own problem-and-fixes tour gains the reflexes that shipped alongside it — aim, feynman, prism, and reorder join the problem list, and the fixes gain worked reorder and re0-merge entries — across the root and all ten translations.

Install

npx skills@latest add LilMGenius/paperthin --global --agent '*'