Skillx

Skill tsubasaogawa/skillx

skillx is a tool for AI agents that lets you try out skills directly from any GitHub repository without formal installation.

Install

npx -y skills add tsubasaogawa/skillx

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use this skill whenever the user wants to run a skill directly from a GitHub repository without pre-installing it, especially for requests involving `owner/repo`, GitHub URLs, temporary skill execution, or when a fetched skill should be cleaned from the local cache after it runs.

SKILL.md

3.6 KB, as published. Nobody here has run it

skillx

Use this skill to fetch and run a skill from the GitHub repository provided in the arguments. Do not assume the skill is installed already. Instead, use the bundled scripts/skillx.sh helper for the deterministic repository-handling steps so normalization, cache safety checks, and cleanup stay reproducible.

Usage

/skillx <REPO> [OPTIONS...]

Arguments

<REPO>: The GitHub repository containing the skill to run
[OPTIONS...]: Additional arguments to pass through to the downloaded skill

Accepted `<REPO>` formats

Accept the following forms:

github.com/owner/repo
https://github.com/owner/repo
[email protected]:owner/repo.git
owner/repo
https://github.com/owner/repo/tree/branch/path/to/skill
https://github.com/owner/repo/tree/branch/path/to/skill/SKILL.md
owner/repo/path/to/skill

You may strip a trailing / or .git. Reject any non-GitHub host.

Execution steps

Follow these steps in order.

1. Parse arguments

Treat the first argument as REPO
Preserve all remaining arguments, in order, as OPTIONS
Treat OPTIONS as literal string arguments, not shell code

If REPO is missing, print the following error and stop:

Error: No repository was specified.
Usage: /skillx <REPO> [OPTIONS...]

2. Prepare the cached skill with the helper script

Run:

bash scripts/skillx.sh prepare "$REPO"

Treat the helper script as the source of truth for the deterministic repository-handling steps. It is responsible for:

normalizing the input and extracting branch/path information
deriving the canonical cache metadata
creating ~/.local/share/skillx/skills/<owner>
cloning into a temporary directory before moving into place
verifying an existing cache's remote URL and clean git status before git pull --ff-only
confirming that SKILL.md exists after clone or update

On success it prints these values:

NORMALIZED_REPO=<owner>/<repo>
OWNER=<owner>
REPO_NAME=<repo>
DISPLAY_REPO=github.com/<owner>/<repo>
CLONE_URL=https://github.com/<owner>/<repo>.git
SKILL_DIR=$HOME/.local/share/skillx/skills/<owner>/<repo>[/<path>]

Read those values from the script output and carry them forward. If the helper exits non-zero, surface its error and stop.

3. Run the downloaded skill

Treat SKILL_DIR as the working directory
Read SKILL_DIR/SKILL.md and follow its instructions
Pass OPTIONS through as additional arguments exactly as provided
If execution fails, remember that failure so you can still perform cleanup in the next step

skillx is a wrapper. Do not replace the downloaded skill with your own summary or interpretation. Read the actual SKILL.md in the fetched repository and execute that skill.

4. Delete the used cache directory

After the downloaded skill finishes, remove only the cache entry for the repository that was just executed.

bash scripts/skillx.sh cleanup "$NORMALIZED_REPO"

Attempt this cleanup whether the downloaded skill succeeded or failed
Never delete the parent owner directory unless it is empty and you are sure it belongs only to this cache layout
If cleanup fails, clearly report that the cache could not be removed and show the location
If the downloaded skill failed earlier, report that original failure too instead of hiding it behind the cleanup result