Pre Deploy Code Drift Detection

Pre-Deploy Code-Drift-Detection

When to use

Before ANY deploy (rsync, docker compose up, git pull-and-restart) where:

• The codebase has >1 active push-source (Mac + work-laptop + server-direct-edit + remote Claude session)
• The Server has a long-lived Working-Tree (not ephemeral CI/CD pipeline)
• The last deploy was >3 days ago (drift compounds)

When NOT to use

• First-time-Setup-Deploy (no parallel sources yet)
• Single-developer-single-machine workflows
• Server-only-via-CI deploys (CI = sole push-source)
• In-Container-Hot-Reload-iteration (no rsync)

The 3-Way-Compare Pre-Flight (90 seconds, mandatory)

# 1) Mac-Side fetch+state
git -C <mac-repo> fetch origin <branch>
git -C <mac-repo> log --oneline <branch>..origin/<branch>     # origin AHEAD of Mac?
git -C <mac-repo> log --oneline origin/<branch>..<branch>     # Mac AHEAD of origin?
git -C <mac-repo> status --short --branch                     # Mac uncommitted?

# 2) Server-Side fetch+state
ssh <user>@<server> "cd <server-repo> && git fetch origin <branch> && \
  echo '=== origin ahead of server ===' && git log --oneline <branch>..origin/<branch> && \
  echo '=== server ahead of origin ===' && git log --oneline origin/<branch>..<branch> && \
  echo '=== server uncommitted ===' && git status --short --branch && \
  echo '=== server untracked ===' && git status --porcelain | grep '^??'"

# 3) Interpret 4 scenarios:
#   A) Both Mac+Server = origin (clean) → safe rsync/pull, proceed
#   B) Mac ahead of origin, Server = origin → Mac push + Server pull, proceed
#   C) Server has uncommitted/untracked → STOP, classify in buckets, decide:
#        c1) commit them (if real work) — backup-branch first
#        c2) discard them (if temp/test) — reset --hard with explicit user confirm
#        c3) preserve them out-of-band (cp to /tmp) — then deploy
#   D) Mac ahead of origin AND origin ahead of Mac (= 3rd push-source exists) → STOP, the dangerous case

The 3rd-Push-Source Discovery (Scenario D)

This is the dangerous case. Symptoms:

• Mac main..origin/main shows commits
• Mac origin/main..main ALSO shows commits
• → Mac and origin have diverged because something else pushed to origin

Discovery questions:

1. Are there commit-messages on origin that look like they came from a different model / different style than usual?
2. Are there work-laptop / iPad-Claude-Session / CI-bot pushes?
3. Check git log origin/main --format='%h %ci %an %s' -10 for author/timestamp patterns

Resolution options:

• (a) Mac rebase onto origin — preserves Mac-local + integrates origin-newer (safest if Mac-local commits don't conflict)
• (b) Mac merge origin — creates merge-commit but preserves both histories
• (c) Cherry-pick selective — if Mac-local is mostly junk, cherry-pick the few good commits onto origin

The Server-Direct-Edit Anti-Pattern

If git status on server shows >10 modified files + >5 untracked files with mtime spread over multiple days:

Diagnosis: someone is editing on the Server directly (a Claude session running on the Server, or SSH-edit). This is an anti-pattern because:

• Server-Direct-Edits are typically uncommitted → no audit trail
• They block rsync because --delete would erase untracked work
• They duplicate work already in flight on Mac → 3-source-Drift escalates

Response:

1. DO NOT rsync over the Server-state — back up first
2. Triage the Server-uncommitted into thematic buckets (mtime-cluster + content-grep)
3. Commit each bucket atomically (with proper author/co-author attribution) → push origin → pull on Mac
4. THEN Mac is the canonical truth and rsync/pull proceeds safely

Output-Format

After the 3-Way-Compare, surface to user:

## Drift-Status
- Mac → origin: {N commits ahead | clean | M behind}
- Server → origin: {N commits ahead | clean | M behind}
- Server Working-Tree: {clean | {N modified, M untracked}}
- Scenario: {A | B | C | D}
- Recommendation: {proceed | commit-then-deploy | stop-decide}

## Risks if we just rsync now
- {list of concrete files / commits that would be lost}

Anti-Patterns

• ❌ Skipping Pre-Flight because "I just deployed yesterday" — Server-Direct-Edits can happen in hours
• ❌ git pull without checking --rebase vs --merge — depends on whether Mac has local commits
• ❌ rsync --delete without Pre-Flight — destroys untracked Server-Direct-Edit work
• ❌ git reset --hard origin/main on Server without backup-branch first — irreversible loss of Direct-Edit work
• ❌ Assuming origin = Mac — only true if Mac just pushed; in Multi-Source-Setup origin can diverge

Real-World Impact

• 4h lost through:
  • 15 server-commits made without pre-flight
  • Push attempt rejected → discovery: origin had 10 parallel local commits (3rd push-source) AND Mac had 10 local unpushed
  • Hard-reset to origin/main + backup-branch + re-commit of only gitignore-safety
• With pre-flight the outcome would be: 90-second discovery → decision to push Mac→origin first, then server pull, then normal rsync. Saves 4h.
• Reproducible: every push from a 3-source setup carries this risk. Without the skill: the next 4h-drift in 1-2 weeks.

Cross-references

• superpowers:using-git-worktrees — alternative when parallel branches should run more isolated
• Pair with a code-review chunk-dispatch workflow when a server-direct-edit backlog must be reviewed before merging