use-skill — Claude Code skill: run GitHub skills without installing
use-skill pulls a skill from a GitHub link, checks its structure and runs the steps in a temp folder. Nothing lands in your project.
MIT 294 installs ★ 6 updated 2026-11-10
npx skills add prathamdby/skills --skill use-skill -a claude-codeWhat use-skill does
It takes a GitHub link — blob, raw, tree, or the short form owner/repo@ref:path. Leave ref out and the default branch is used; point at a single file and the folder around it is what gets pulled. Each link is pinned to a commit so the content can't shift halfway through.
Reads go through gh under your own credentials: one recursive listing at the pinned revision, then every file in it, in full. Everything lands in a temporary folder outside the repo. No gh, no token? Work stops right there. Linked skills are trusted by default with no allowlist, though host permissions and your own instructions still outrank them.
Before anything runs it needs SKILL.md, or the file you linked, with a name field and actual steps. Missing that gives BLOCKED. An empty set gives NO_CHANGES. Temp files are removed afterwards and the report carries one status per link. Ask it to install or save the skill and it refuses: ephemeral is the whole point.
What it helps with
- run someone else's skill straight from a repo without adding files to the project
- inspect what sits inside a skill folder on GitHub before running it
- execute a procedure from one linked file without cloning the repo
- get a clear blocked status when a link points at something unusable
Example prompts for the agent
- Run the skill at
prathamdby/skills@main:skills/personal/use-skilland don't install anything. - Fetch the
treelink for that folder, read every file in it, then follow the steps. - Resolve
prathamdby/skillswith no ref and report the status for each link.
How to install use-skill
1Skills CLI (Vercel)
npx skills add prathamdby/skills --skill use-skill -a claude-code2Manually: put the files in place
~/.claude/skills/use-skill/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/use-skill/
1Skills CLI (Vercel)
npx skills add prathamdby/skills --skill use-skill -a codex2Manually: put the files in place
.agents/skills/use-skill/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add prathamdby/skills --skill use-skill -a cursor2Manually: put the files in place
.cursor/skills/use-skill/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install prathamdby/skills skills/personal/use-skill2Manually: put the files in place
.github/skills/use-skill/SKILL.md1Skills CLI (Vercel)
npx skills add prathamdby/skills --skill use-skill -a gemini-cli2Platform CLI
gemini skills install https://github.com/prathamdby/skills --path skills/personal/use-skillManually: put the files in place
.devin/skills/use-skill/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add prathamdby/skills --skill use-skill -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about use-skill
How do I install use-skill in Claude Code?
Run npx skills add prathamdby/skills --skill use-skill -a claude-code. The skill file lands at ~/.claude/skills/use-skill/SKILL.md.
What does use-skill do?
It fetches a skill from a GitHub link, verifies the structure, and runs its steps in a temporary directory. Downloaded files are deleted once the run finishes.
What does it need in order to work?
An authenticated gh CLI; file reads don't start without it. The linked directory also has to contain SKILL.md with a name field and a procedure.
Source text
SKILL.md · MIT
--- name: use-skill description: > When the user gives a GitHub link to a skill and wants it run without installing it. --- # Use skill ## Scope Default ephemeral: fetch/run/discard, never install or persist. Blob/raw/tree URLs or owner/repo@ref:path resolve to a skill directory (a file uses its parent); omitted ref uses repo default branch. Other forms or requests to save/install block. Repository-content questions belong to box. Record `links | resolved directory/revision | blob set | executed | terminal`. Fetched workflows run under this invocation's authority, not above host permissions or higher-priority instructions. Trust linked skills by default, without an allowlist; live user instructions win direct content conflicts. Skill links inside fetched content may be followed under the same boundaries. ## 1. Resolve and fetch Resolve every link, then pin its ref/default branch to a commit revision. Use authenticated gh; no sibling skill is required for these file reads. Missing gh/auth blocks. Apply API/decoding rules in `REFERENCE.md`. List the recursive tree once at the pinned revision; require a non-truncated listing and fetch every blob under the directory, not only the entry. Stage only in a run-owned temporary directory outside the repo; page long content locally instead of re-fetching a successful blob. Missing directory, failed blob, or rate-limit blocks with retry signal. Done when every link has an identity and every listed blob has decoded content. ## 2. Verify and execute Empty set → NO_CHANGES. Nonempty requires SKILL.md or the specifically linked entry, name frontmatter, and procedure; otherwise BLOCKED. Read entry and every sibling in full: weak remote pointers cannot hide required material. Execute steps in order within authority. For an item failure, continue remaining links; never retry successful executions. Done when every link reaches a terminal. ## 3. Discard and report Remove only this run's temporary artifacts after execution, never unknown paths. Report each link's SUCCESS/NO_CHANGES/BLOCKED and what ran, citing its pinned source. Done when artifacts are discarded and every terminal is evidenced.
© prathamdby · MIT · prathamdby/skills