Landing a PR With One Command

The interesting part of shipping software is writing the code. The boring part is everything that happens after the PR is open: watching CI, fixing whatever it flags, squash-merging, deleting branches, and pulling main.

It’s necessary, repetitive, and exactly the kind of thing AI agents should handle for us.

In my previous post, I introduced cc-forge: my personal collection of Claude Code skills and subagents designed to automate the repetitive parts of my development workflow. I already had a skill called /land, that helped to log previous actions in a localized CLAUDE.md file. This skill seemed like the perfect one to append a merge-and-sync step to.

I am sharing a short breakdown of /land here so that you can [hopefully] be inspired and use it as a reference to build your own related skill. You can find the full implementation in the cc-forge repository.

The Breakdown

In brief, here is what happens when you run /land:

Stamp provenance: It adds an entry to the relevant CLAUDE.md with the PR link and a summary. This builds a running record of why a directory looks the way it does and helps actually implement the compound part that cc-forge was inspired by.
Verify: It runs any tests mentioned in the PR and watches GitHub Actions (gh pr checks --watch).
Fix: If CI fails, the agent diagnoses the issue, proposes a fix, and commits it. (This loops up to three times, but always asks for your confirmation before pushing).
Merge: Once green, it squash-merges the PR and deletes the remote branch.
Sync: It checks out main, pulls the latest, and then deletes the local PR branch.

Automating the Predictable

By automating the predictable sequence, /land closes the loop. From /brainstorm to a merged PR, you spend less time clicking through the GitHub UI and more time just building.