illo-skill: one slash command that turns a sentence into a print-style illustration

The prompt is the whole interface, and a named character is part of it

There is no parameter list here. The skill is invoked with a slash command and a sentence, and the sentence is the entire specification. The examples the project puts forward are deliberately varied rather than a template:

/illo we replatform with zero downtime
/illo how a book gets published explainer diagram
/illo using Blip, make a flowchart of how a newsletter issue ships
/illo stuck → slice → shipped as a mini-comic
/illo blot cutout waving
/illo surprise me

Two of those lines carry an instruction about form rather than subject, one asks for a flowchart, one asks for a mini-comic, one asks for a character pose, and the last asks for nothing at all. That range is the design. The recurring element across outputs is a mascot, and the mascot has a name you supply, as Blip is in the flowchart example. Companion characters live in a separate repository, tmchow/illo-characters, so the character set is something you extend rather than something baked in. The one thing the interface does not offer is a size, colour, or style flag, which is why the examples lean on natural language instead.

Ten install rows spanning five different plugin systems

The install table is the largest single object in the repository, and it exists because agent hosts have not converged on one packaging format. The rows fall into families. Claude Code, Codex, and Grok CLI all use a marketplace pattern with a two step add followed by an install of a named component:

/plugin marketplace add tmchow/illo-skill
/plugin install illo@illo-skill

Codex and Grok CLI use the same shape with their own binary, `codex plugin marketplace add` and `grok plugin marketplace add`, and Grok CLI adds a trust flag to the install step. Gemini CLI and Copilot take a URL or a repository slug directly, and Hermes installs a path inside the repository rather than the repository root. Cursor and the catch-all row both fall through to a generic installer:

npx skills add tmchow/illo-skill --skill illo

The Copilot row names the same command with a cross-agent flag, `gh skill install tmchow/illo-skill illo`, and a dash agent option. So five distinct mechanisms are on offer: a plugin marketplace, a repository URL, a subdirectory path, a generic skills registry, and a template tap.

Updates are not one command, and three hosts have no update verb at all

The update column is where the ten rows stop looking like variations of one thing. Claude Code has an explicit update command and an alternative in the form of a marketplace auto-update toggle. Codex has a marketplace upgrade. Grok CLI has a plugin update. Gemini CLI has an extensions update, Copilot has a skill update, and Hermes has a skills update. OpenClaw has no update verb at all: the instruction is to reinstall with the same command. The Cursor row and the catch-all row both come down to re-running an installer.

One row is worse than that. The Grok Bot entry is not a command but a template link, and its update instruction is to add the template again after updates. That row also carries its own footnote further down, where the flow is spelled out: open the illo bot template and tap Add to Grok Bot, which creates the bot on your account. In practice that host has no way to pull a new version, so a user on it stays on whatever was current when they tapped the link.

The Cursor row has a second, quieter problem. It is marked as pending review in the marketplace listing, which means the native path for that host is not the generic installer, it is a route that may not resolve yet.

The repository root carries a manifest for every host format it targets

The top level of the repository is a good map of which hosts are first class. There is a .claude-plugin directory, a .codex-plugin directory, a .grok-plugin directory, and a .cursor-plugin directory, each a manifest in the convention its host expects. Gemini uses a single gemini-extension.json file at the root rather than a directory. A shared .agents directory appears to serve the hosts with no native format, and skills/ is where the actual skill content lives, since the engine and cost documentation is filed at skills/illo/README.md.

Alongside those sit the files a release toolchain produces: version.txt, a .release-please-manifest.json, a release-please-config.json, and a CHANGELOG.md. That combination means versions are cut by an automated tool rather than by hand, and the tag you install from is generated rather than chosen. There are also two agent instruction files at the root, AGENTS.md and CLAUDE.md, plus docs/, tests/, and an _assets directory. The presence of tests/ is worth noting for a project whose visible product is a drawing: the code that is tested is the plugin plumbing, not the artwork.

Three tags in five weeks, and a hot number in the version

The release history is dense and the version number is high for a project this young. v0.35.0 shipped on 2026-08-23, v0.36.0 on 2026-09-17, and v0.37.0 on 2026-09-21. The last two are four days apart. The final push to the repository matches the newest tag, 2026-09-21, which is the same date the last release went out, so the tree and the artifact are aligned.

The number itself deserves a caveat rather than an interpretation. A version at 0.37 for a skill with a short public history usually means the version counts the whole plugin surface across hosts, so a change to one manifest can move it. Nothing in the repository states a compatibility range for a given host, and the changelog is the place to look before assuming an install row still reads the same way it did a month ago. The licence is MIT, attributed to Trevin Chow, and there is a separate NOTICE file inside the skill directory, which in this layout is the file to read if you are redistributing the character assets rather than the code.

Engines, models, cost, and keys are pushed one click deeper

The most consequential sentence in the documentation is also the shortest. Engines, models, cost, and API keys are all deferred to skills/illo/README.md. That single line is where the real cost of the skill lives, and it is not in the page a search result or a repository visit will show you.

What the top level does tell you is that keys are part of the picture at all, which separates this from a purely local transform. It also says the intended install method is your runtime's native plugin or skill manager, so the project is not distributing a binary and is not asking you to edit a host config by hand, apart from the generic installer fallback. The public site at illo-skill.com is positioned as the place for live examples, character packs, and copy-paste installs, and the repository treats it as the companion to the table rather than a duplicate of it.

What the top level does not state is which drawing engine runs by default, what a single illustration costs, whether a key is required for every host or only some, or what happens when a quota runs out mid-publish. For a tool aimed at a publishing workflow, that is the gap worth filling before you build a habit around it.

Editorial conclusion

illo-skill is worth a try if you publish articles or newsletters and want illustrations that share one visual voice instead of a new style per image. It is a poor fit if you need deterministic output, offline operation, or a fixed set of characters, because the character set is a choice you make per prompt and the engines behind the drawing are documented in a file the top level never summarises. Before installing, check three things: that your host has a real plugin or skill manager rather than a copy-paste fallback, which row of the install table actually applies to you, and what the drawing calls cost, since that lives in skills/illo/README.md rather than in the page most people land on. Version 0.37.0 shipped on 2026-09-21, after 0.36.0 four days earlier, so the install row you copy may change shape between releases.

Frequently asked questions

Official sources

1. License: MIT
2. Project website
3. README
4. Releases
5. tmchow/illo-skill on GitHub