SuperAPI AI Directory

clone-repo

Coding & development

View repository

Updated 10-15 19:54

About this skill

Clone Repo (vendor as subtree)

Pulls a library's source into the current project at repos/<name>/ so coding agents can read real source instead of relying on stale training data or doc-CLI shims. Move from Michael Arnaldi's "Vibe Engineering Effect Apps" — agents are post-trained on reading code, not prose, so feed them code.

The result: a dumb directory the agent treats as "more of my codebase". Distilling what's relevant into project-local patterns/*.md is a separate, on-demand step.

What this skill does NOT do

  • It does not generate patterns/*.md. That's a per-topic follow-up the user invokes explicitly so they self-select which subset of the library the project actually uses.
  • It does not configure ESLint / TypeScript strictness — orthogonal, project-wide setup.
  • It does not pull updates. Re-running on the same name will refuse; see "Updates" below for the manual upgrade command.

Arguments

$clone-repo <url> [name] [branch]
  • <url> — required. Any clone URL (https or ssh).
  • [name] — optional. Directory under repos/. Defaults to the URL's basename without .git.
  • [branch] — optional. Defaults to the remote's HEAD branch (queried via git ls-remote --symref).

If the user just gives a URL, infer name and branch.

Workflow

1. Preflight checks

Run all of these. If any fails, stop and tell the user exactly what to fix — do not try to fix it for them.

git rev-parse --is-inside-work-tree            # must print "true"
git rev-parse HEAD >/dev/null 2>&1              # must succeed — subtree needs ≥1 commit
test -z "$(git status --porcelain)"             # working tree must be clean
test ! -e repos/<name>                           # destination must not exist

Each failure has a concrete remedy to relay:

  • not in a git repo → git init && git commit --allow-empty -m "init"
  • no commits yet → git commit --allow-empty -m "init"
  • dirty tree → ask the user to commit/stash; do not stash on their behalf
  • repos/<name> exists → either pick a different [name], or run the update command (see Updates)

2. Resolve the default branch (if not given)

git ls-remote --symref <url> HEAD | awk '/^ref:/ {sub("refs/heads/", "", $2); print $2; exit}'

If the command fails (private repo, network), ask the user for the branch instead of guessing main.

3. Add the subtree

git subtree add --prefix=repos/<name> <url> <branch> --squash

Creates repos/<name>/ with the library's source at that branch's tip, plus a single squashed commit on the current branch. No history pollution; the SHA is pinned by the squash commit.

4. Wire it into AGENTS.md

Find the agent-instructions file in this order: AGENTS.md, CLAUDE.md, .cursor/rules/agents.md. Whichever exists first, append (do not overwrite) a "Reference repositories" section if one isn't already there, then add a bullet for the new subtree:

## Reference repositories

Source-of-truth code for libraries we depend on. Treat as **read-only reference material** — do not edit files under `repos/`. When asked about a library listed below, explore its source here first instead of guessing or relying on training data.

- `repos/<name>/` — <url> @ <branch> (squashed)

If none of those files exist, create AGENTS.md with that section as the only content.

If an entry for repos/<name> already exists, leave it alone.

5. Report

One short message to the user:

  • The subtree path, source URL, branch, and squash commit (git log -1 --oneline on the current branch).
  • The exact follow-up prompt to distill patterns:

    Explore repos/<name>/ for patterns on how to . Save your findings to patterns/<task>.md with concrete file references.

  • The update command (so they know where the upgrade path lives — see below).

Stop there. Do not write patterns yourself unless asked.

Updates (manual, not automated by the skill)

To pull upstream changes later:

git subtree pull --prefix=repos/<name> <url> <branch> --squash

Why subtree, not submodule, not plain clone

  • Submodule — extra .gitmodules machinery, requires git submodule update --init on every clone, agents often skip the contents.
  • Plain git clone + .gitignore — not reproducible across machines or CI runs; the cloned SHA isn't pinned anywhere.
  • Subtree --squash — files live in the repo, pinned to a SHA via the squash commit, no special tooling on clone. Agents see them as ordinary directories. Matches what Arnaldi uses in the talk.

Reference