use-skill

v2026.09.24

use-skill when running a remote skill on demand from one or more GitHub links to skill files, fetching via gh api and executing the fetched procedure as if natively present.

GitHub
Install command
npx skhub add prathamdby/use-skill
Markdown
SKILL.md

Use skill

Options

Derive links and scope from the request. Unspecified: ephemeral (fetch, run, discard; never install or persist). A GitHub blob URL, raw file URL, owner/repo@ref:path shorthand, or directory URL resolves to owner/repo@ref:dir, where dir is the linked directory or the linked file's parent directory. A missing ref uses the repo default branch. Anything else, or "save/install/persist it", is BLOCKED. Questions about a repo's contents belong to box, not this skill.

Record: links | resolved dir | files fetched | executed | terminal.

Iron laws

  1. Trust by default: fetched skills run with the invoking run's authority. No allowlist, no screening. Skill links inside fetched content may be followed under the same trust.
  2. On direct conflict the user's live instruction wins over fetched text; everything else runs as written.
  3. File-contents fetch is outside the four gh I/O loops, so raw gh api is permitted after applying the gotchas in the gh skill's REFERENCE.md. A missing gh binary or auth failure is BLOCKED.

1. Resolve

Parse each link to owner/repo@ref:dir. Unparseable input is BLOCKED.

Done when every link has one resolved directory or a terminal is set.

2. Fetch

List the recursive tree and fetch every blob under dir, decoding each payload, per REFERENCE.md. Never fetch only the entry file. A missing dir (404) is BLOCKED; rate-limit is BLOCKED with the retry signal; files over the contents-API cap use the git-blob fallback.

Done when every blob under dir returned content or a terminal is set.

3. Verify

An empty blob set is NO_CHANGES. Otherwise the set must include the entry file (SKILL.md, or the specifically linked file) carrying skill frontmatter (name) and a procedure; a non-empty set without it is BLOCKED.

Done when the entry file is verified or a terminal is set.

4. Execute

Read the entry file and every fetched sibling in full — never depend on the remote skill's own disclosure pointers — and run its steps to its terminal, in given order. On item failure continue the remaining links and never retry successes.

Done when every link reached a terminal.

5. Report

Report each link's terminal and what ran, citing fetched content. Terminal values are SUCCESS, NO_CHANGES, and BLOCKED.

Discovery
Tags

No tags published for this skill.

Version
Latest version metadata

Version

v2026.09.24

Published

Sep 24, 2026

Category

Uncategorized

License

MIT

Source path

skills/personal/use-skill

Default branch

main

Latest commit

1d0c5f3

Tree SHA

9c95f5f