diff --git a/.intent/review-state.json b/.intent/review-state.json
index a7a27c5..809241e 100644
--- a/.intent/review-state.json
+++ b/.intent/review-state.json
@@ -294,6 +294,18 @@
       "evidence": [
         "update-readme-action.test.ts (8 tests) passes against a local bare repository and a stand-in gh: PR holds only the two documents update wrote; sources untouched; caller checkout unchanged; rerun pushes nothing; refresh on main's new tip; behind checkout still bases on tip; in-sync closes the PR; unreadable file= opens nothing; branch==base exits 2. check-sync-action.test.ts (8 tests) still passes after moving its helpers to .github/actions/shared/mdcode.mjs."
       ]
+    },
+    "source:.out-of-scope/extract-shrink-guard.md": {
+      "fingerprint": "b32c01d95ab489e49413e956f1cbb7e37fcd0474debe8d0925484b8a6db57d3b",
+      "snapshot": {
+        ".out-of-scope/extract-shrink-guard.md": "76388d6ffa05e0454781cc47a685bcf9b9e9d6c9837d8575857f3519d0ed452c"
+      },
+      "head": "238c32e5534770f961ab48e46c81674d88bec474",
+      "outcome": "no-change",
+      "reason": "Adds .out-of-scope/extract-shrink-guard.md, a triage record of a rejected enhancement (#25). It documents existing extract behaviour and changes no skill guidance.",
+      "evidence": [
+        "Changed: .out-of-scope/extract-shrink-guard.md at 238c32e5534770f961ab48e46c81674d88bec474"
+      ]
     }
   }
 }
diff --git a/.out-of-scope/extract-shrink-guard.md b/.out-of-scope/extract-shrink-guard.md
new file mode 100644
index 0000000..05028b0
--- /dev/null
+++ b/.out-of-scope/extract-shrink-guard.md
@@ -0,0 +1,37 @@
+# Extract shrink guard
+
+`mdcode extract` does not refuse, or ask for an extra flag, when a write would make an existing file
+smaller.
+
+## Why this is out of scope
+
+The idea came from #21. There, a plain `extract` on a README whose blocks were excerpts
+(`file=… region=…`) replaced 27 real source files with just those excerpts, and a "refuse writes that
+shrink a file" check would have caught every one. That data loss is now prevented by what `extract`
+writes, not by a size check:
+
+- A block with `region=` is spliced into the existing file. Code outside the region, and regions the
+  Markdown doesn't mention, are kept.
+- Ambiguous mappings, such as several blocks for one file without distinct `region=` values, broken
+  markers or unsafe paths, are refused before anything is written.
+- Without `--force`, an existing file whose block has no `region=` is never overwritten; it is
+  skipped and `extract` exits 2.
+
+That leaves one way to shrink a file: `extract --force` on a whole-file block, where the block is the
+entire file. `--force` already is the explicit opt-in to replace an existing file with the block's
+contents. A shrink guard there would only mean a second flag (`--force --allow-shrink`) for the same
+intent, and a block that legitimately removes code is exactly the case `--force` is for.
+
+To see what `--force` would change before running it, use the check, which writes nothing:
+
+```bash
+mdcode extract --check --force README.md
+# ✗ Out of sync: line 3: src/whole.ts differs from this block; extract would overwrite it
+```
+
+Splices that make a region shorter are ordinary code edits, so any threshold there (bytes, lines or a
+ratio) would refuse normal changes or let real loss through.
+
+## Prior requests
+
+- #25: "extract: refuse writes that would shrink an existing file unless --force", split out of #21