pluck

        ______            ______
___________  /___  __________  /__
___  __ \_  /_  / / /  ___/_  //_/
__  /_/ /  / / /_/ // /__ _  ,<
_  .___//_/  \__,_/ \___/ /_/|_|
/_/

Download a file, folder, or whole repo from a Git host URL — without the .git directory.

Paste a GitHub URL. Get the contents. git clone drags .git along, the GitHub CLI needs auth even for public repos, and copy-pasting raw.githubusercontent.com links by hand is a chore. pluck parses the URL you already have and does the right thing.

Install

curl -fsSL https://raw.githubusercontent.com/pbshgthm/stash/main/install | bash -s pluck

Drops a standalone copy of pluck into ~/.local/bin/ — no clone required, and nothing breaks if you move the repo later. Single bash script, no runtime deps beyond curl, tar, and (for --sparse) git. Make sure ~/.local/bin is on your PATH.

Usage

pluck <url> [destination] [options]

URL forms

All of these work (GitHub):

REF is any branch name, tag, or commit SHA.

If you omit the ref (shorthand or bare /OWNER/REPO URL), pluck queries the GitHub API for the repo’s default branch. Falls back to HEAD if the API call fails.

Destination

The second positional argument decides where the output lands. --out / -o is the equivalent flag; you can’t use both.

• Omitted. Drops into the current directory using the default name:

  • Repo download → ./<repo>
  • Folder or file download → ./<basename of path>

• Ends with / or points at an existing directory. Treated as a parent; the default name is appended. Mirrors git clone semantics.

• Anything else. Treated as the full output path (directory or file).

pluck vercel/ai                      # → ./ai
pluck vercel/ai ai-sdk               # → ./ai-sdk
pluck vercel/ai ~/Downloads/         # → ~/Downloads/ai   (trailing slash = parent)
pluck vercel/ai ~/Downloads          # → ~/Downloads/ai   IF ~/Downloads exists
                                     # → file named Downloads otherwise

Options

When the destination already exists

pluck mirrors git clone: if the target path already exists (a file, or a non-empty directory), it refuses to continue. Pass -f / --force to remove it first and proceed. Empty directories are accepted as-is — the download fills them in.

When to use --ref

URLs like github.com/foo/bar/tree/feat/baz/qux are ambiguous. Is the ref feat and the path baz/qux, or is the ref feat/baz and the path qux? The URL alone doesn’t say. pluck defaults to splitting on the first slash — if that’s wrong, name the ref explicitly:

pluck https://github.com/foo/bar/tree/feat/baz/qux --ref feat/baz

How it works

Two download paths:

1. Default: streaming tarball. Fetches https://codeload.github.com/OWNER/REPO/tar.gz/REF and pipes it directly into tar -xz — no intermediate file. Simple, usually fastest, and works with any ref GitHub recognizes. Downloads the whole repo though, which is why --sparse exists.

2. --sparse: partial clone + sparse-checkout. For large repos where you only need a subfolder. Uses git clone --filter=blob:none --no-checkout so blob content is fetched lazily, then narrows the working tree to your subpath via git sparse-checkout set --no-cone. Requires git. Commit SHAs (7–40 hex chars) are detected and use git checkout <sha>; anything else is treated as a branch with --branch.

After either path, pluck copies the contents to the destination and (for tree/repo downloads) strips .git unless --keep-git is set.

Single-file downloads (blob, raw, raw.githubusercontent.com) always go through a single curl hit to raw.githubusercontent.com — no tarball, no clone.

Examples

# Whole repo into ./next.js
pluck https://github.com/vercel/next.js

# A single file
pluck https://github.com/vercel/next.js/blob/canary/README.md

# One folder out of a large repo
pluck https://github.com/torvalds/linux/tree/master/kernel/sched --sparse

# Shorthand into a specific directory
pluck vercel/ai ~/code/

# Named output
pluck vercel/ai my-vendored-ai-sdk

Limitations

• GitHub only. GitLab, Bitbucket, Gitea, etc. aren’t supported — different URL shapes and different tarball endpoints.
• No auth. Public repos only. (Private repos would need a token; out of scope for a small bash tool.)
• Without --sparse, the full repo is downloaded even if you only want one folder. For small repos this is usually faster than cloning; for multi-GB repos it isn’t.
• The default-branch lookup is a plain HTTPS request to the GitHub API and is subject to unauthenticated rate limits (60/hour/IP). If you’re hitting the limit, spell the ref out in the URL.