PatchPilot: Run Copilot diff apply verify (Copilot CLI Challenge)

#cli #devchallenge #ai #github

This is my submission for the GitHub Copilot CLI Challenge.

What I built

I built PatchPilot, a small command-line tool that closes the loop:

run a failing build/test command
ask GitHub Copilot CLI to make the smallest fix inside the repo
capture a real git diff
re-run the same command to verify it’s green
write a short report you can share/review

Repo: https://github.com/aisamposhen/patchpilot

Why it’s useful

A lot of “AI in terminal” demos stop at suggestions. PatchPilot is opinionated: no celebration until the command actually passes again.

This makes it great for:

quickly unblocking on small test/build failures
creating minimal, reviewable patches
teaching the habit of verifying fixes

Demo (90 seconds)

# install (dev)
python3 -m venv .venv
source .venv/bin/activate
pip install -e .

# run the intentionally broken sample project
cd sample_projects/python_pytest_broken
pip install -r requirements.txt
patchpilot run --show-prompts --branch patchpilot/demo -- pytest -q

# view the generated report
sed -n '1,140p' PATCHPILOT_REPORT.md

How I used Copilot CLI (explicit)

PatchPilot calls Copilot CLI non-interactively via gh copilot.

The key design choice is that Copilot edits the repo in-place (with permission restricted to the repo directory), and PatchPilot then captures the patch using git diff. That keeps the result audit-friendly.

Example (simplified):

gh copilot -- --add-dir <repo> --allow-all-tools -p "...make pytest -q pass..."

UX / safety

Requires a clean worktree before running (so diffs are trustworthy)
Creates a new branch (defaults to patchpilot/fix)
Never pushes automatically
Stops after one attempt (keeps it predictable)

What’s next

If I had more time:

add one retry loop (bounded)
optional gh pr create to open a PR with the report
support multiple languages/framework templates

Thanks for reading!

DEV Community