Fix reverse sync missing human commits when mono-filtered advances

--ancestry-path returns empty when mono-filtered/main has advanced
past the last forward sync (any monorepo commit causes josh to create
a new filtered SHA not in the subrepo's ancestry). Use git merge-base
to find the last connected point instead.

Also return skip-dirty (not skip) when trees differ but no human
commits found, preventing state from being updated on false skips
which would permanently lose unsynced commits.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

CHANGELOG.md

@@ -6,10 +6,16 @@
 
 - **Linearize fallback for reverse sync**: When josh-proxy rejects a push due to unmappable merge commits, reverse sync automatically falls back to linearizing the history — cherry-picking regular commits individually and squashing only the problematic merge commits via `cherry-pick -m 1`. Preserves individual commit granularity while handling complex merge topologies that josh cannot map. PR body notes when linearization was used.
 
+### Fixes
+
+- **Reverse sync human commit detection**: Fixed `--ancestry-path` returning empty when `mono-filtered` has advanced past the last forward sync. Any monorepo commit (even outside the subfolder) causes josh to create a new filtered commit, breaking the ancestry path to subrepo HEAD. Now uses `git merge-base` to find the last connected point.
+- **State not updated on false skip**: When trees differ but no human commits are found, reverse sync now returns `skip-dirty` instead of `skip`. State is not updated, so the system retries on the next run instead of silently dropping commits.
+
 ### Docs
 
 - Expanded troubleshooting guide for "Josh rejected push" with root cause analysis and prevention advice.
 - Added ADR-011: Linearize fallback for reverse sync.
+- Added recommended Git workflow section to the guide.
 
 ## 1.2.0
 

bin/josh-sync

@@ -259,6 +259,10 @@ _sync_direction() {
         echo "::error::Target ${target_name}, branch ${branch}: josh rejected push — check proxy logs"
         continue
       fi
+      if [ "$result" = "skip-dirty" ]; then
+        log "WARN" "Trees differ but no human commits — skipping state update (will retry)"
+        continue
+      fi
 
       # Update state (only on success)
       local new_state

lib/sync.sh

@@ -226,7 +226,9 @@ ${BOT_TRAILER}: filter-change/${mono_branch}/$(date -u +%Y-%m-%dT%H:%M:%SZ)")
 # ─── Reverse Sync: Subrepo → Monorepo ──────────────────────────────
 #
 # Always creates a PR on the monorepo — never pushes directly.
-# Returns: skip | pr-created | josh-rejected
+# Returns: skip | skip-dirty | pr-created | josh-rejected
+#   skip       — trees identical, nothing to do
+#   skip-dirty — trees differ but no human commits found (don't update state)
 
 reverse_sync() {
   local mono_branch="$SYNC_BRANCH_MONO"
@@ -265,15 +267,27 @@ reverse_sync() {
   fi
 
   # 4. Find new human commits (excludes bot commits from forward sync)
-  #    Uses --ancestry-path to restrict to the direct lineage and avoid
-  #    leaking old history through reconciliation merge parents.
-  local human_commits
-  human_commits=$(git log --ancestry-path "mono-filtered/${mono_branch}..HEAD" \
+  #    Uses merge-base + --ancestry-path to restrict to the direct lineage.
+  #    merge-base is needed because mono-filtered may have advanced past the
+  #    last forward sync (any monorepo commit, even outside the subfolder,
+  #    causes josh to create a new filtered commit).  Using mono-filtered
+  #    directly with --ancestry-path returns empty when it's not an ancestor
+  #    of HEAD.  merge-base always finds the last connected point.
+  local base human_commits
+  base=$(git merge-base "mono-filtered/${mono_branch}" HEAD 2>/dev/null || echo "")
+
+  if [ -n "$base" ]; then
+    human_commits=$(git log --ancestry-path "${base}..HEAD" \
       --oneline --invert-grep --grep="^${BOT_TRAILER}:" 2>/dev/null || echo "")
+  else
+    # No common ancestor (first sync or fully diverged) — skip ancestry-path
+    human_commits=$(git log "mono-filtered/${mono_branch}..HEAD" \
+      --oneline --invert-grep --grep="^${BOT_TRAILER}:" 2>/dev/null || echo "")
+  fi
 
   if [ -z "$human_commits" ]; then
     log "INFO" "No new human commits in subrepo — nothing to sync"
-    echo "skip"
+    echo "skip-dirty"
     return
   fi
 
@@ -301,9 +315,16 @@ reverse_sync() {
     git checkout -b "tmp-linearize-$$" "mono-filtered/${mono_branch}" 2>/dev/null \
       || die "Failed to create linearize branch"
 
+    # Reuse merge-base from step 4 (same reasoning: mono-filtered may have
+    # advanced past the last forward sync, breaking --ancestry-path).
     local commit_shas
-    commit_shas=$(git log --ancestry-path "mono-filtered/${mono_branch}..${original_head}" \
+    if [ -n "$base" ]; then
+      commit_shas=$(git log --ancestry-path "${base}..${original_head}" \
         --reverse --format="%H" --invert-grep --grep="^${BOT_TRAILER}:" 2>/dev/null || echo "")
+    else
+      commit_shas=$(git log "mono-filtered/${mono_branch}..${original_head}" \
+        --reverse --format="%H" --invert-grep --grep="^${BOT_TRAILER}:" 2>/dev/null || echo "")
+    fi
 
     local sha
     for sha in $commit_shas; do