agentsclimarketplace

Git

Skill sherifabdlnaby/skills/skills/git

🍣 Hand-rolled AI Skills distilled from my everyday experience. Git & Mise (...and others).

Install
npx -y skills add sherifabdlnaby/skills --skill git

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

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 6 stars6 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

Load when planning or about to git commit, branch, push, rebase, resolve a merge conflict, open PR, stack PR, or write/update the description of a Pull Request (PR / stacked diff), respond to review comments, review a PR, or watch/babysit a PR's CI. Carries the user's git conventions and style you are expected to match. Load as early as possible, the moment git or GitHub work is anywhere in the chat's future.

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

5.1 KB, as published. Nobody here has run it

Git

Route first: each action's actual rules live in a references/ file. Open the matching one before planning or acting; rules like commit grouping and branch naming shape the plan, so reading them late means redoing it. Several actions, several references.

Router

Branches (create, name, stack) -> references/branches.md Switching branches, naming rules, ticket references, switching with a dirty tree, stacked PRs, restacking (after merge, chain of 3+, mid-stack push), force-pushing.

Committing -> references/commits.md Pre-staging walk, file confirmation, message style, hook-failure handling, scoping, empty commits, CI refresh.

Pull Requests (open, update) -> references/pull-requests.md Title format (including stacked PRs), body format, how to write descriptions, AI footers, post-create flow, linking PRs.

Responding to review comments (on your own/target PR) -> references/review-responses.md Classify the reviewer (automated bot/AI tool, an AI-disclosed agent behind a human account, or a human), then fix / push back / escalate; replies, thread resolution, re-requesting review.

Rebasing, squashing, resolving conflicts -> references/rebase.md Backup-branch convention, force-with-lease, range-diff verification, conflict procedure, non-interactive squash recipes. Stacked-PR restack mechanics and --onto (drop intermediate commits) live in references/branches.md, read that too.

Reviewing someone else's PR -> references/reviewing.md Procedure, review lenses, severity tiers, posting mechanics.

Watch a PR's CI and automated reviews -> references/watch.md Spawning the cheap background watcher sub-agent (and the background-task fallback), hot/cold polling, what to relay vs hold, auto-addressing bot reviews, the final digest. Uses scripts/pr-watch.py, never a sleep loop.

Always

Disclose AI. Anything posted on GitHub on the user's behalf (PR body, comment, issue, ticket update) carries an AI footer. The PR-body template lives in references/pull-requests.md; the post templates, chosen by who made the specific decision (Agent Decided vs Human Guided), live in AI Disclosure below. Use them as verbatim as possible, do not write from memory.

Parallelize read-only calls. Batch independent read-only git/gh calls (status, diff, log, gh pr view/diff/checks) into a single tool call; each sequential read is a roundtrip for nothing. Mutating commands (commit, push, rebase, gh pr create/edit/merge) stay sequential.

Voice for everything public (commit messages, PR titles and bodies, comments, issues):

  • Omit needless words. Concise sentences, no padding paragraphs.
  • Emojis sparingly, only where one helps the reader catch something while glancing.
  • NO em dashes in any public-facing text. Use commas, parentheses, or periods. (why: em dashes are a classic AI tell and repel readers.)
  • No vague reaffirmations like "for accountability", "for performance", "for resiliency" unless that reason is already in the conversation. (why: invented justification misrepresents the change.)
  • Dry, low-key humor.
  • Don't overdo formatting. Keep it balanced. (why: visual noise crowds out the content.)
  • A user-supplied Human Note is exempt: it goes in the PR body verbatim.

AI Disclosure

Every post on the user's behalf (comment, reply, issue) ends with this footer, after a ---. Pick the variant by who made the specific decision behind the post. A request to handle a task, fix an issue, or open a PR does not count as guidance on the decisions the agent makes while doing it.

Agent Decided (🤖): the agent chose the position, change, or response without the user's direction on that specific decision.

  • <Claude|Cursor|OpenCode>: the tool you're running as.
  • <MODEL>: the friendly name of the model you're running, e.g. Opus 4.8.
---

_<sub>🤖 Agent Decided: Posted by <Claude|Cursor|OpenCode> (<MODEL>) autonomously on behalf of @<GITHUB_USERNAME>.</sub>_

Human Guided (🤝): the user chose or materially directed the specific decision. This does not mean they reviewed the final wording or implementation.

---

_<sub>🤝 Human Guided Response: Posted by <Claude|Cursor|OpenCode> (<MODEL>) on behalf of @<GITHUB_USERNAME>.</sub>_

When unsure, use Agent Decided.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.