My Git + GitHub Workflow with Agent Skills

How I use three small skills to turn commits, PR creation, and PR feedback into a repeatable workflow.

#GIT#GITHUB#AI

I built these 3 skills to speed up and automate my Git and GitHub workflow which saves me a lot of time every day.

Skills Source

https://github.com/saadjs/agent-skills/tree/main/skills

This is the loop I run daily:

  • git-incremental-commits: split messy working trees into clean commits
  • gh-pr-create: push + open a PR with a consistent body
  • gh-address-pr-comments: fetch feedback, add regressions, fix, push, reply

1) git-incremental-commits

Goal: turn “a pile of changes” into small Conventional Commits, ending with a clean git status.

What it automates (for me): deciding commit boundaries, staging strategy, and commit messages.

Typical command pattern:

git status -sb
git diff --stat

git add -p
git commit -m "feat(scope): …"

# repeat until clean

Built-in guardrails I like:

  • Won’t commit on main/master (branch safety gate)
  • Encourages lockfiles to be committed with dependency changes
  • Prefers single-purpose commits (uses git add -p when needed)

2) gh-pr-create

Goal: after the branch is clean, push and open a PR with a predictable structure.

Preconditions:

git status -sb
gh auth status

Workflow:

# first push sets upstream
git push -u origin HEAD

# next pushes are just
git push

The value is the PR body format. I keep it the same every time:

  • ## Summary
  • ## Major changes
  • Optional: ## Screenshots, ## Tests, ## Additional info

That consistency makes PRs easier to scan and review.

After the PR is created, the skill asks who to request review from and applies the right action.

3) gh-address-pr-comments

Goal: turn PR feedback into an evidence-based checklist and ship the fixes end-to-end.

Required input: a PR number.

Core flow:

# checkout PR branch
gh pr checkout "$PR_NUMBER"

# fetch PR + comments (inline + issue comments)
gh pr view "$PR_NUMBER" --json number,title,url,headRefName,baseRefName,comments,reviews

What it enforces:

  • Classify feedback (meaningful vs subjective/unclear)
  • Add regression tests before fixes (when testable)
  • Run the full test suite before pushing
  • Reply to threads with what changed + commit SHA(s)
  • Re-request review only after everything is pushed

I usually use $<skill> to invoke the skill in Codex CLI or app.