docs(advanced-git): two measured git behaviours promoted from agent memory

[CybotTM]

Merging this records two git behaviours that cost data or invite a wrong fix, both measured on git 2.55.0 rather than recalled: git worktree remove deletes ignored files without asking, and a bare clone's missing fetch refspec makes --force-with-lease reject with (stale info). Both sit next to the section they correct; nothing is removed.

These come from a /retro promote pass over accumulated local agent memory. Nineteen candidates in the git area were checked against every reference on current main. Most did not belong here: five describe one operator's environment and identities, one is about the agent harness rather than git, one is a SonarCloud/Pages decision, and one is already in the GitLab skill's cheatsheet. Two were verified as already covered — advanced-git.md:1638 for never running git checkout <ref> -- <path> over an uncommitted change, and advanced-git.md:1876 for staging by name instead of git add -A.

The two changes, one commit each.

• advanced-git.md, "Leave the worktree before you remove it" — the section credits the missing --force with catching an unintended removal. That guard only covers tracked files. A worktree holding an ignored .env and an ignored build/out.txt was removed by a plain git worktree remove: exit 0, no warning, both files gone. Those are the files that exist in no other copy, and nothing in git can recover them because they were never in the object store. Adds the status --porcelain --ignored read that shows what would be lost.
• advanced-git.md, "Bare-Repo Layouts" — git clone --bare leaves remote.origin.fetch empty, so the clone creates no refs/remotes/* and the lease has no recorded remote state to compare against. The push is rejected with (stale info), which reads as "somebody else pushed" and invites plain --force, dropping the protection entirely. Records both remedies, each measured: passing the lease its value from git ls-remote, and repairing the clone once by setting the refspec and fetching. The second is what a long-lived bare layout wants — with no refspec, git fetch origin updates nothing, so every later "is my branch behind?" question is answered from refs that never move.

On where the second one sits. The note this came from blamed --depth 1, and that is wrong: a shallow clone does carry refs/remotes/origin/* and pushes with an ordinary --force-with-lease without complaint. Measured both before writing, which is why this landed under Bare-Repo Layouts rather than under Shallow Fetches.

Verification. validate-skill.sh reports 0 errors and 20 warnings both on this branch and on unmodified main — identical.

Assisted by claude-code:claude-opus-5 — Session

https://claude.ai/code/session_01NxSeVq1hDnGBCqKLGcjQ6m

[coderabbitai]

Warning

Review limit reached

Next included review available in 51 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used the included review currently available.

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Advanced

Run ID: 393c9f05-4513-42b7-be46-65a8b3b31101

📥 Commits

Reviewing files that changed from the base of the PR and between 12b86b5 and 8c86328.

📒 Files selected for processing (1)

• skills/git-workflow/references/advanced-git.md

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[CybotTM]

Review of this PR

Copilot was unavailable — the account's review quota is exhausted, so the copilot_code_review ruleset cannot be satisfied on any PR until it resets. Recording my own read of the diff so the decision rests on a review rather than on the gate being unsatisfiable.

Both claims in this diff were measured before being written, not recalled. That was the point of the exercise, and it changed one of them.

git worktree remove deleting ignored files: reproduced on git 2.55.0 in a throwaway repository. A worktree holding an ignored .env and an ignored build/out.txt — both shown as !! by status --porcelain --ignored — was removed by a plain git worktree remove with exit 0, no prompt and no --force, and both files were gone afterwards. The existing section credits the missing --force with catching unintended removals, which is true only for tracked content.

The --force-with-lease claim came from a note blaming --depth 1, and that is wrong. A shallow clone carries refs/remotes/origin/main and pushed with an ordinary --force-with-lease without complaint. The condition is git clone --bare, which leaves remote.origin.fetch empty so no tracking refs exist at all; that produced ! [rejected] HEAD -> featA (stale info) on the first try. Both remedies in the text were then measured too: --force-with-lease=<branch>:<sha> from git ls-remote pushed successfully, and setting the refspec plus one git fetch made the ordinary form work. This is why the section landed under Bare-Repo Layouts rather than under Shallow Fetches.

Scope. One file, 44 added lines, nothing removed. No SKILL.md change, so no version-parity or body-line-count effect. Seventeen further candidates from the same promote pass were left out — most describe one operator's environment rather than git behaviour, and two were verified as already covered (advanced-git.md:1638 and :1876).

Verification. validate-skill.sh reports 0 errors and 20 warnings both on this branch and on unmodified main. 0 unresolved threads. Rebased onto 12b86b5; this repository's main moved three times during the run, and its classic branch protection has strict: true, so the rebases were required rather than cosmetic.

Assisted by claude-code:claude-opus-5 — Session

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[github-actions]

Dependency Review

✅ No vulnerabilities or license issues or OpenSSF Scorecard issues found.

Scanned Files

None

[CybotTM]

Self-review: 8c86328

The review this pull request demands is unsatisfiable (Copilot quota wall or repeated bot failures on this head). Per the documented fallback, the diff on this head was reviewed by the PR author; this comment is the on-the-record attestation the merge gate reads back. It stops matching on the next push.