fix: classify notebooks as trimmable data

CHANGELOG.md

@@ -3,6 +3,11 @@
 All notable changes to this project are documented here, following
 [Keep a Changelog](https://keepachangelog.com/) and semantic versioning.
 
+## [0.1.20] - 2026-09-23
+
+### Fixed
+
+- Classify `.ipynb` notebooks as data trim candidates, including notebooks below the large-data threshold.
 ## [0.1.19] - 2026-09-23
 
 ### Fixed

README.md

@@ -67,7 +67,11 @@ ctxtrim  ·  my-repo  ·  412 files
 
 Known binary media, including AVIF, HEIC/HEIF, APNG, WebM, Ogg, FLAC and M4A, are not read as text and contribute no text tokens. They remain explicit ignore candidates.
 
+<<<<<<< HEAD
+Jupyter notebooks (`.ipynb`) are treated as data and suggested for trimming regardless of size. This excludes the whole notebook, including its source cells, not only its outputs; review these suggestions before using `--write`.
+=======
 Loose `.snap` files are treated as generated output and suggested for trimming, just like files inside `__snapshots__/` directories.
+>>>>>>> origin/main
 
 Estimates use the widely-cited ~4-chars-per-token rule (great for ranking and relative savings; pass `--price` to match your model).
 

package.json

@@ -1,6 +1,6 @@
 {
   "name": "ctxtrim",
-  "version": "0.1.19",
+  "version": "0.1.20",
   "description": "Trim what bloats your AI coding context. Scan a repo, find the high-cost/low-value files ballooning your Claude Code / Cursor / Codex context, and write ignore files to cut token cost. Zero dependencies.",
   "type": "module",
   "bin": {

src/classify.js

@@ -21,7 +21,7 @@ const LOCKFILES = new Set([
 ]);
 
 const DATA_EXT = new Set([
-  ".csv", ".tsv", ".parquet", ".ndjson", ".jsonl", ".sqlite", ".sqlite3",
+  ".ipynb", ".csv", ".tsv", ".parquet", ".ndjson", ".jsonl", ".sqlite", ".sqlite3",
   ".db", ".dump", ".log", ".pkl", ".npy", ".npz", ".arrow", ".feather", ".geojson",
 ]);
 

test/notebooks.test.js

@@ -0,0 +1,37 @@
+import { mkdtempSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { test } from "node:test";
+import assert from "node:assert/strict";
+import { classify } from "../src/classify.js";
+import { scanRepo } from "../src/scan.js";
+
+test("notebooks are data trim candidates below the large-data threshold", () => {
+  for (const name of ["analysis.ipynb", "analysis.IPYNB"]) {
+    const result = classify(name, { tokens: 1510, maxTokens: 2000 });
+    assert.equal(result.category, "data", name);
+    assert.equal(result.trim, true, name);
+    assert.equal(result.binary, false, name);
+  }
+  assert.equal(classify("analysis.py", { tokens: 1510 }).trim, false);
+});
+
+test("scan accounts for output-heavy notebooks and suggests their ignore pattern", (t) => {
+  const root = mkdtempSync(join(tmpdir(), "ctxtrim-notebook-"));
+  t.after(() => rmSync(root, { recursive: true, force: true }));
+  const notebook = {
+    cells: [{ cell_type: "code", execution_count: 1, metadata: {}, source: ["print('x')"],
+      outputs: [{ output_type: "stream", name: "stdout", text: ["x".repeat(6000)] }] }],
+    metadata: {}, nbformat: 4, nbformat_minor: 5,
+  };
+  writeFileSync(join(root, "analysis.ipynb"), JSON.stringify(notebook));
+  writeFileSync(join(root, "analysis.py"), "print('x')\n");
+  const result = scanRepo(root);
+  const file = result.files.find((entry) => entry.rel === "analysis.ipynb");
+  assert.ok(file.tokens > 1500 && file.tokens < 2000);
+  assert.equal(file.category, "data");
+  assert.equal(file.trim, true);
+  assert.equal(result.totals.trimTokens, file.tokens);
+  assert.deepEqual(result.patterns, ["analysis.ipynb"]);
+  assert.equal(result.files.find((entry) => entry.rel === "analysis.py").trim, false);
+});