skl-x

Install, audit and manage AI agent skills for Claude Code and Codex.

Install skills

npx -y skl-x add sentimony/skills

skl-x add runs npx -y skills add with the same arguments, so every source format and option of npx skills add works as is.

# List skills in a repository
npx -y skl-x add sentimony/skills -l

# Specific skills to specific agents, non-interactive
npx -y skl-x add sentimony/skills -s tdd cross-review -a codex claude-code -y

# Hide the npx skills output, print a progress line, a table and a summary (needs -y)
npx -y skl-x add sentimony/skills -s tdd -a codex claude-code -y -m md

# The summary line only (needs -y)
npx -y skl-x add sentimony/skills -s tdd -a codex claude-code -y -m sm

Installing 1 skill from https://github.com/sentimony/skills to Project scope ██████████ 100%
skill    .agents    .claude  skills-lock.json      cost
tdd    universal  symlinked         +          ~125 tok
Project scope · +1 skill installed · tdd

Without -m (lg), the npx skills add output is shown as is, followed by the same lines when it changed any skill.

In a terminal the bar appears empty once the npm registry resolves (or npx prints anything): the first seven cells fill on a timer, one every 150 ms, then the eighth when npx starts printing, the ninth when every skill named with -s is on disk, the tenth when npx exits. Every value stays on screen for at least 150 ms: one that arrives sooner waits, and the final line waits for the last one. In scripts and CI only the final line is printed. If npx fails, its full output is printed.

npx skills add replaces a skill only in the dirs of -a. After it succeeds, add removes what is left of each added skill in the other agent dir, such as a skl sln link or a copy for the other agent, as skl rm before the install would. The target of a removed link is not touched. Entries in the dirs of -a stay; without -a, so does an entry that resolves to the skill's entry in the other dir. If an added skill is missing for an agent of -a, add exits 1.

Local skills

skl-x sln <path> symlinks <path>/skills/<name> into .agents/skills and .claude/skills. The links point straight at the clone, so edits there show up without reinstalling. <path> may also be the skills folder itself, such as ../skills/plugins/devflow/skills. When neither holds skills, sln searches <path>: every <name>/SKILL.md whose frontmatter name is <name> is a skill, so in a plugin repo sln ../skills -s tdd dashfix links skills from two plugins (plugins/devflow/skills/tdd, plugins/writing/skills/dashfix). The search goes five folders deep, skips hidden folders (.agents, .claude, .git), node_modules, dist, coverage and symlinks, and does not look inside a skill. A name found in two folders exits 1.

Before linking, sln removes each selected skill the way skl rm does: a copy or another link in the agent dir outside -a and its skills-lock.json entry, which would otherwise bring a copy back over the link on the next add. Without -y it lists these with the replaced copies and asks once.

# Every skill in the clone
npx -y skl-x sln ../skills -a codex claude-code -y

# Every skill except the listed ones
npx -y skl-x sln ../skills -x scope-check echarts -y

# Only the listed skills, only .claude/skills
npx -y skl-x sln ../skills -s tdd cross-review -a claude-code

Symlinking 2 skills from ../skills to Project scope ██████████ 100%
skill           .agents    .claude  skills-lock.json      cost
cross-review  symlinked  symlinked         -          ~186 tok
tdd           symlinked  symlinked         -          ~125 tok
Project scope · +2 skills symlinked · cross-review tdd

• Without -s, every <path>/skills/<name>/SKILL.md is linked (or <path>/<name>/SKILL.md when <path>/skills has no skills), except skills with metadata.internal: true (as npx skills add does).
• An existing copy or symlink with the same name is replaced after a Replace N existing skills? prompt, or at once with -y.
• skills-lock.json is not changed. Global scope (-g) is not supported yet.
• In a terminal the bar fills one cell every 50 ms, about half a second in all; in scripts and CI only the final line is printed.

Option	Description
--skill, -s <names...>	Only these skills; a name starting with a space is skipped
--reject, -x <names...>	Every skill except these (not with -s)
--agent, -a <agents...>	codex (.agents/skills), claude-code (.claude/skills); default: both
--yes, -y	Skip npx prompts (add); replace and remove existing copies and lock entries without asking (sln)
--mode, -m <mode>	lg (large, default), md (medium), sm (small); md and sm in add need -y

Mode	add	sln	rm
lg	npx output, then the md lines	as md	plan and summary
md	progress line, table, <scope> · +N skills installed · <names>	progress line, table, <scope> · +N skills symlinked · <names>	<scope> · -N skills removed · ~X tok
sm	<scope> · +N skills installed · <names>	<scope> · +N skills symlinked · <names>	<scope> · -N skills removed · ~X tok

<scope> is Project scope, or Global scope with -g. rm counts the skills removed from at least one place and sums their cost; with nothing to remove it prints <scope> · No skills. The .agents/.claude/lock breakdown is in the lg summary only. In a terminal these lines are gray, except the green bar and the skill names (cyan).

Cost

npx -y skl-x cost       # project scope
npx -y skl-x cost -g    # global scope

Per-skill ambient token cost (name + description, chars/3), sorted descending. Skills with disable-model-invocation: true show - and stay out of the total.

Usage

Skill usage from agent session logs, multiplied by cost.

npx -y skl-x usage -a claude -p 7d
npx -y skl-x usage -a codex --mode activations

Project scope · Usage 22 times by 7d
skill         .agents  .claude      cost       total
cross-review        9        2  ~186 tok  ~2,046 tok
tdd                 0       11  ~125 tok  ~1,375 tok
2 skills            9       13  ~311 tok  ~3,421 tok

.agents counts Codex runs, .claude - Claude Code runs; total is cost × runs. A skill that is no longer installed is shown in red with ~? tok.

Flag	Default	Description
--agent, -a	both	claude-code (claude), codex
--period, -p	all	60s, 30m, 24h, 7d, 2w, 6mo, all (1m is a minute, 1mo is 30 days)
--since	-	yyyy-mm-dd, overrides --period
--mode	merged (claude) / activations (codex)	merged, attributed, activations, mentions
--format	text	text, json
--scan-all-files	-	Ignore file mtime, read everything

• merged - per-session union of attributed and activations (max per skill).
• attributed - entries with an attributionSkill field set by Claude Code.
• activations - Skill tool calls (Claude), read-like exec_command_end events or <skill> XML (Codex).
• mentions - foo/SKILL.md paths or superpowers:name strings anywhere. The broadest signal; can match prompts, specs or docs.

Other commands

Command	Alias	Description
skl-x list	ls	Table of skills: .agents, .claude, lock and cost per skill
skl-x remove	rm	Remove skills from disk and skills-lock.json
skl-x update	up	Install the latest skl-x globally (npm i -g); only a global npm install
skl-x uninstall		Remove the global skl-x (npm uninstall -g) and its update-check cache; skills stay
skl-x completion		Print a shell completion script

skl-x with no command opens a menu in a terminal and prints cost otherwise.

skl-x list

Project scope
skill              .agents    .claude  skills-lock.json      cost
cross-review     universal  symlinked         +          ~186 tok
echarts                  -     copied         -           ~98 tok
old-skill                -          -         +            ~? tok
webapp-debugger  symlinked     broken         -            ~? tok
4 skills                 2          3         2          ~284 tok

Cell	Meaning
universal	a folder in .agents/skills (the canonical copy of npx skills)
copied	a folder in .claude/skills
symlinked	a symlink, to .agents/skills or to a local clone
broken	a symlink whose target is gone
-	not there
+	in the lock; red when the skill is on neither disk

skl-x ls --names prints one name per line, for scripts and completion. skl-x ls -m sm prints one line, Project scope · 30 skills · ~3,333 tok (<scope> · No skills when empty); md and lg print the table.

skl-x remove

skl-x rm tdd                     # plan, Proceed? [y/n], then Clean lock? [y/n]
skl-x rm .                       # every skill in scope
skl-x rm . -x tdd cross-review   # every skill except the listed ones
skl-x rm tdd -y                  # no prompts, no plan
skl-x rm . -y -m sm              # one line: Project scope · -30 skills removed · ~3,333 tok
skl-x rm tdd --lock-only         # also --agents-only, --claude-only
skl-x rm tdd -g                  # global scope

skl-x completion

Tab-completes subcommands and skill names for rm. Needs a global install.

skl-x completion bash >> ~/.bashrc
skl-x completion zsh >> ~/.zshrc
skl-x completion fish > ~/.config/fish/completions/skl.fish

Scope

Where you run it	Scope
inside a git repo (any subdirectory)	the repo root
anywhere else, $HOME included	that directory
cost, ls, rm with --global / -g	global: ~/.agents/.skill-lock.json, ~/.claude/skills, ~/.agents/skills

skl usage is always project scope. skl add and skl sln write into the current directory.

Global install

npm i -g skl-x    # provides `skl-x` and the short `skl`
skl update        # later: install the latest version globally (alias `skl up`)
skl uninstall     # remove it: asks first, -y skips the question

Moving from @sentimony/sklx or skillio: all three packages install the skl bin, so remove the old one first:

npm rm -g @sentimony/sklx skillio && npm i -g skl-x

To skip the daily update check set SKL_X_NO_UPDATE_CHECK=1 (SKLX_NO_UPDATE_CHECK still works).

Requires Node.js 22 or newer.

License

MIT