Skills

Reusable agent skills for repository setup, code review, and refactoring. The instructions use the standard SKILL.md format and are independent of any programming language, project, model provider, or host-specific tool names.

• simplify: simplify recently written modules lossily and from the top down. A fresh explainer describes the simplest model of the behavior; incidental behavior is cut and reported; stages, representations, and stored state are trimmed before functions, which target 10 lines, in the style of Bitwise. Whole-program simplification requires explicit user authorization.
• install-repository-skills: install or repair this collection in a repository, pin it at .agents/skills, and share that checkout with Claude Code through a symlink.

Third-party skills, pinned as submodules and credited to their authors:

• humanizer (MIT, blader): rewrite AI-sounding prose so it reads like the writer, without changing what it says.
• hallmark (MIT, Nutlope): design new pages that look designed rather than generated. Its optional image step calls Together AI only when TOGETHER_API_KEY is set.
• unslop (CC BY 4.0, Damir Sokolovsky): turn an existing AI-looking UI into a designed interface.

humanizer keeps its SKILL.md at its repository root, so its submodule sits at the root. The other two keep theirs under skills/<name>/, so their repositories live in vendor/ and a root-level symlink exposes each skill folder. Agents that do not follow symlinked skill folders will not see those two. Update a pin with git submodule update --remote <path> and review the log before committing it.

Each skill folder lives at the repository root. In a consuming Git repository, add this repository directly at .agents/skills and link Claude Code's discovery directory to the same checkout:

git submodule add https://github.com/tsnl/skills.git .agents/skills
git submodule update --init --recursive .agents/skills
mkdir -p .claude
ln -s ../.agents/skills .claude/skills
git add .gitmodules .agents/skills .claude/skills

An agent that can read install-repository-skills/SKILL.md can perform this setup, including checking existing installations before changing them. Once available, invoke it as $install-repository-skills in Codex or /install-repository-skills in Claude Code, giving the target repository path.

These commands assume those paths are unused. Commit the submodule and symlink in the consumer repository so everyone uses the same skill revision. The result is .agents/skills/simplify/SKILL.md, also available through .claude/skills/simplify/SKILL.md.

After cloning a consumer repository, initialize its pinned skills, including the third-party submodules, with:

git submodule update --init --recursive .agents/skills

To adopt a newer version, review and commit the updated submodule reference:

git submodule update --remote .agents/skills
git submodule update --init --recursive .agents/skills
git diff --submodule=log -- .agents/skills
git add .agents/skills

For simplification, use $simplify in Codex or /simplify in Claude Code. Include the files to review, behavior to preserve, and any pass limits or model preferences in your request. For example:

Use simplify on the new parser module. Preserve its public API. Use a fresh explainer for each pass and repeat until further changes would not improve clarity. Keep all changes within the parser module and its tests.

For personal use across projects, keep a separate checkout and link individual skills into each tool's personal skills directory:

git clone --recurse-submodules https://github.com/tsnl/skills.git "$HOME/Developer/skills"
mkdir -p "$HOME/.agents/skills" "$HOME/.claude/skills"
ln -s "$HOME/Developer/skills/simplify" "$HOME/.agents/skills/"
ln -s "$HOME/Developer/skills/simplify" "$HOME/.claude/skills/"

Update that personal checkout with git -C "$HOME/Developer/skills" pull --ff-only --recurse-submodules.

The simplify workflow honors the model and reasoning effort available and requested in its host. If isolated delegation is unavailable, the skill requires that the limitation be disclosed; a local critique is not presented as an independent review.

See the official Codex skill documentation and Claude Code skill documentation.