skillex

A TUI for browsing and reading your installed Claude Code skills.

Features

• Discovers all installed skills from Claude Code plugins as well as local skills
• Split-pane layout with filterable skill list and rendered markdown preview
• Vim-style hjkl navigation
• Per-skill enable/disable: press space to toggle a skill on or off by renaming SKILL.md to SKILL.md.disabled (start a new Claude session to apply)
• Activation health indicators: colored tag per skill shows whether its description is likely to auto-invoke
• Description budget meter: tracks total description length against the 16,000-character limit before skills silently stop loading

Per-skill enable/disable

Claude Code supports enabling/disabling entire plugins but has no built-in way to toggle individual skills. Skillex fills this gap: press space on any skill to toggle it.

Under the hood, skillex renames SKILL.md to SKILL.md.disabled. Claude Code only discovers files named SKILL.md, so renamed files are invisible to it. No external state file or config is needed.

Disabled skills appear dimmed in the list with a red disabled tag. The budget meter excludes them from the total since Claude won't load them. Start a new Claude Code session after toggling for changes to take effect.

Skill health indicators

Activation style dot

Each skill in the list shows a ● dot next to the plugin name:

Dot	Meaning
● (green)	Directive:description uses imperative language (ALWAYS, MUST, NEVER, DO NOT). Community benchmarks show ~98% auto-activation rate.
● (dim)	Neutral:no strong signal either way.
● (orange)	Passive:description uses descriptive language (Use when, Helps, Can be used, Useful for). Community benchmarks show ~69% auto-activation rate.

Auto-invocation in Claude Code is unreliable by default. Testing by Scott Spence across 200+ prompts found a ~50% baseline activation rate:essentially a coin flip. Ivan Seleznov's 650-trial study confirmed that description wording is the primary lever: passive descriptions scored as low as 69% while directive descriptions using ALWAYS invoke... / DO NOT ... directly reached 98–100%.

Example rewrites:

Before (passive, orange dot)	After (directive, green dot)
Helps you write commit messages.	ALWAYS invoke this skill when writing a git commit message.
Use when reviewing pull requests.	NEVER review a pull request without invoking this skill first.

Description budget meter

The bottom bar shows desc budget: X/16k tracking the total length of all skill descriptions combined:

Color	Meaning
Green	Under 12,800 chars (80%):plenty of room
Orange	12,800–15,999 chars:approaching the limit
Red	16,000+ chars:over the limit

Claude Code loads all skill descriptions into its system prompt at startup under an available_skills section. The budget for that section is 16,000 characters (or 2% of the model's context window, whichever is larger). When the combined total of all skill descriptions exceeds the budget, skills are silently excluded:no error, no warning, they just stop appearing to Claude. The limit was first documented empirically in GitHub issue #13099, where researchers found 42 of 63 installed skills invisible once the total crossed ~15,500 chars. It is now officially documented in the Claude Code troubleshooting guide and can be raised by setting the SLASH_COMMAND_TOOL_CHAR_BUDGET environment variable.

Installation

Homebrew

brew install smauermann/tap/skillex

From source

go install github.com/smauermann/skillex@latest

Usage

skillex

Keybindings

Key	Action
j/k	Navigate list / scroll preview
space	Toggle skill enabled/disabled
l	Focus preview pane
h	Back to skill list
/	Filter skills
q	Quit