diff --git a/plugins/overengineering/.claude-plugin/plugin.json b/plugins/overengineering/.claude-plugin/plugin.json
index f6958413e3..ad3d5289dc 100644
--- a/plugins/overengineering/.claude-plugin/plugin.json
+++ b/plugins/overengineering/.claude-plugin/plugin.json
@@ -1,7 +1,7 @@
 {
   "$schema": "https://json.schemastore.org/claude-code-plugin-manifest.json",
   "name": "overengineering",
-  "version": "0.4.14",
+  "version": "0.4.15",
   "description": "Evidence-earned-keep audit of an existing enforcement surface, covering agent hooks and standing instructions, repository and version-control hooks, CI lanes and gate scripts, branch protections, forge apps, and declared external integrations. It treats every incumbent mechanism as a retirement candidate until empirical evidence earns its keep, argues every verdict in cost of carry, caps retirement-direction verdicts on security-class artifacts at FLAG-FOR-HUMAN, and realigns to the simplest adequate solution behind an explicit per-item human gate. The audit is read-only and emits a diffable findings artifact; realignment is a separate, explicitly invoked skill; and a third read-only lane re-runs the audit on whatever cadence the consumer wires and reports only what moved since the last run, above a configurable noise budget. A justification lane applies the same method to whatever single artifact you point at, a decision record, a document, a component, a dependency, or a code construct, asking whether a reason existed for it and whether that reason still holds, and reporting how much evidence each verdict actually rests on.",
   "author": {
     "name": "Melodic Software",
diff --git a/plugins/overengineering/CHANGELOG.md b/plugins/overengineering/CHANGELOG.md
index 563f655144..8977a4f82d 100644
--- a/plugins/overengineering/CHANGELOG.md
+++ b/plugins/overengineering/CHANGELOG.md
@@ -3,6 +3,12 @@
 All notable changes to the `overengineering` plugin are documented here. Format follows
 [Keep a Changelog](https://keepachangelog.com/en/1.1.0/); this plugin uses semantic versioning.
 
+## [0.4.15] - 2026-09-28
+
+### Changed
+
+- **`audit` report groups out-of-repo findings by owner** ([#4596](https://github.com/melodic-software/claude-code-plugins/issues/4596)). The inline report lists in-repo findings first and in full, then one summary row per out-of-repo owner (count, verdict mix, pointer to the artifact). Finding ids and artifact rows are unchanged. #4597 already scoped the playbooks repo-sweep entry off the four org-level layers; a direct audit still walks every settings scope, so this is a report view, not a walk filter. Eval 13 covers the rollup.
+
 ## [0.4.14] - 2026-09-27
 
 ### Fixed
diff --git a/plugins/overengineering/skills/audit/SKILL.md b/plugins/overengineering/skills/audit/SKILL.md
index 38fab7d870..e519a533fc 100644
--- a/plugins/overengineering/skills/audit/SKILL.md
+++ b/plugins/overengineering/skills/audit/SKILL.md
@@ -121,6 +121,16 @@ Parse `$ARGUMENTS`:
 - Anything else, a free-text focus hint (a path, a mechanism name). Narrow attention with it; it
   does not change the layer scope, and a hint that matches nothing is reported, not silently dropped.
 
+The inline report groups by custody: in-repo findings first and in full; out-of-repo findings
+(user- and machine-scope settings, plugin hook manifests owned upstream) as one summary row per
+owner. The findings artifact is unchanged. See
+[context/report-template.md](context/report-template.md) "Custody grouping".
+**Claim:** the walk still covers every settings scope; only the inline report groups out-of-repo
+findings per owner. **Basis:** issue #4596 after #4597 scoped the playbooks repo-sweep entry off
+the four org-level layers; a direct `/overengineering:audit` still walks every harness-merged
+scope. **As of:** 2026-09-28. **Recheck:** when this skill gains a custody-scope argument or stops
+walking user or machine settings.
+
 ## Before the walk
 
 1. **Resolve the branch identity, then the artifact home.** The branch call above yields a branch
diff --git a/plugins/overengineering/skills/audit/context/report-template.md b/plugins/overengineering/skills/audit/context/report-template.md
index 6566bf969b..005b7fff52 100644
--- a/plugins/overengineering/skills/audit/context/report-template.md
+++ b/plugins/overengineering/skills/audit/context/report-template.md
@@ -70,6 +70,19 @@ it.
    Cap the inline list **only when the artifact was written**; the artifact carries the rest.
    When no branch identity resolved and this summary is the only record, emit **every** finding
    inline. A cap here would discard the tail of a scheduled detached run.
+   **Custody grouping (report view only).** In-repo findings render first, in full. Findings whose
+   custody is out-of-repo (a `settings:` identifier outside the tree, a user- or machine-scope
+   plugin hook manifest the repo does not own) collapse to **one summary row per owner**: the
+   owning plugin, marketplace, or settings scope, the count of findings, the verdict mix, and a
+   pointer to the underlying artifact rows. Artifact rows and finding ids are unchanged; only this
+   view groups them. #4597 scoped the playbooks `repo-sweep` hygiene entry off the four org-level
+   layers; a direct `/overengineering:audit` still walks every settings scope the harness merges,
+   so this grouping is what keeps those rows from burying the in-repo list.
+   **Claim:** a custody *filter* is declined; the walk stays complete and only the inline report
+   groups by owner. **Basis:** #4597 already took the org-level layers off the repo-sweep entry;
+   a walk filter would hide user- and machine-scope incumbents this skill still has to name
+   (§12). **As of:** 2026-09-28. **Recheck:** when a consumer-config key scopes the walk itself,
+   or when the harness stops merging user and machine settings into a repo session.
 5. **The proposed ablation batch**, when one was produced: its items, an owner and a re-check date
    each, and the observation window's end date.
 6. **Open checkpoints**: the intent questions awaiting an answer (attended), or the count of
diff --git a/plugins/overengineering/skills/audit/evals/evals.json b/plugins/overengineering/skills/audit/evals/evals.json
index 92d7e62d92..e404d415c3 100644
--- a/plugins/overengineering/skills/audit/evals/evals.json
+++ b/plugins/overengineering/skills/audit/evals/evals.json
@@ -169,6 +169,20 @@
         "Writes a checkpoint through that route after each completed layer, re-reading and merging first, with scope naming only the layers completed so far",
         "Never uses a shell content-write for the artifact"
       ]
+    },
+    {
+      "id": 13,
+      "name": "out-of-repo-findings-rollup-per-owner-in-the-report",
+      "prompt": "/overengineering:audit — this is a 37-file repo. Walk everything; I need the report, not a filter.",
+      "expected_output": "Walks every settings scope the harness merges, including user and machine, and writes one artifact row per plugin hook manifest as before. The inline report lists in-repo findings first, in full. Out-of-repo findings (user-scope plugin hook manifests owned upstream, ~/.claude/settings.json, machine-scope settings) collapse to one summary row per owner: count, verdict mix, and a pointer at the artifact rows. Finding ids in the artifact are identical to a run without the rollup. It does not add a custody=repo argument and does not skip out-of-repo layers.",
+      "files": [],
+      "narration": true,
+      "expectations": [
+        "Renders in-repo findings before any out-of-repo content in the inline report",
+        "Gives each out-of-repo owner at most one row in the report (count, verdict mix, pointer to artifact rows)",
+        "Leaves the findings artifact rows and finding ids unchanged",
+        "Does not skip out-of-repo layers and does not introduce a custody-scope argument"
+      ]
     }
   ]
 }