Merge pull request #10733 from Kilo-Org/feature-read-ipynb-content

feat(cli): read Jupyter notebooks as cell content

.changeset/read-notebook-cells.md

@@ -0,0 +1,5 @@
+---
+"@kilocode/cli": patch
+---
+
+Read Jupyter notebooks as ordered markdown and code cell content instead of raw notebook payloads.

packages/opencode/src/kilocode/tool/notebook.ts

@@ -0,0 +1,46 @@
+import * as path from "path"
+import { Readable } from "stream"
+import * as Encoding from "../encoding"
+
+type ObjectValue = Record<string, unknown>
+
+const object = (value: unknown): value is ObjectValue => typeof value === "object" && value !== null && !Array.isArray(value)
+
+const parse = (text: string): unknown => {
+  try {
+    return JSON.parse(text)
+  } catch {
+    return undefined
+  }
+}
+
+const source = (value: unknown): string | undefined => {
+  if (typeof value === "string") return value
+  if (!Array.isArray(value) || !value.every((line) => typeof line === "string")) return undefined
+  return value.join("")
+}
+
+const render = (kind: "markdown" | "code", text: string) => {
+  const body = text.endsWith("\n") ? text : `${text}\n`
+  return `<${kind}_cell>\n${body}</${kind}_cell>`
+}
+
+export async function open(filepath: string): Promise<Readable | undefined> {
+  if (path.extname(filepath).toLowerCase() !== ".ipynb") return undefined
+
+  const raw = (await Encoding.read(filepath)).text
+  const data = parse(raw)
+  if (!object(data) || !Array.isArray(data.cells)) return Readable.from([raw])
+
+  const cells: string[] = []
+  for (const cell of data.cells) {
+    if (!object(cell)) continue
+    if (cell.cell_type !== "markdown" && cell.cell_type !== "code") continue
+
+    const text = source(cell.source)
+    if (text === undefined) continue
+    cells.push(render(cell.cell_type, text))
+  }
+
+  return Readable.from([cells.length ? cells.join("\n\n") : "(Notebook contains no markdown or code cell content.)"])
+}

packages/opencode/src/tool/read.ts

@@ -15,6 +15,7 @@ import { isPdfAttachment, sniffAttachmentMime } from "@/util/media"
 // kilocode_change start
 import * as Encoding from "../kilocode/encoding"
 import * as TextStream from "../kilocode/text-stream"
+import * as Notebook from "../kilocode/tool/notebook"
 // kilocode_change end
 
 const DEFAULT_READ_LIMIT = 2000
@@ -358,6 +359,8 @@ export const ReadTool = Tool.define(
 // routed through TextStream.withFallback so non-UTF-8 files are decoded via
 // iconv. The body otherwise matches upstream.
 export async function lines(filepath: string, opts: { limit: number; offset: number }) {
+  const extracted = await Notebook.open(filepath) // kilocode_change - extract readable notebook cells before paging
+  if (extracted) return readLines(extracted, opts) // kilocode_change
   return TextStream.withFallback(filepath, (stream) => readLines(stream, opts))
 }
 

packages/opencode/test/kilocode/read-notebook.test.ts

@@ -0,0 +1,196 @@
+import { describe, expect } from "bun:test"
+import { Effect, Layer } from "effect"
+import path from "path"
+import { Agent } from "../../src/agent/agent"
+import * as CrossSpawnSpawner from "@opencode-ai/core/cross-spawn-spawner"
+import { AppFileSystem } from "@opencode-ai/core/filesystem"
+import { LSP } from "../../src/lsp/lsp"
+import { Instruction } from "../../src/session/instruction"
+import { MessageID, SessionID } from "../../src/session/schema"
+import { ReadTool } from "../../src/tool/read"
+import * as Tool from "../../src/tool/tool"
+import { Truncate } from "../../src/tool/truncate"
+import { provideInstance, tmpdirScoped } from "../fixture/fixture"
+import { testEffect } from "../lib/effect"
+
+const ctx = {
+  sessionID: SessionID.make("ses_test-notebook"),
+  messageID: MessageID.make(""),
+  callID: "",
+  agent: "code",
+  abort: AbortSignal.any([]),
+  messages: [],
+  metadata: () => Effect.void,
+  ask: () => Effect.void,
+}
+
+const it = testEffect(
+  Layer.mergeAll(
+    Agent.defaultLayer,
+    AppFileSystem.defaultLayer,
+    CrossSpawnSpawner.defaultLayer,
+    Instruction.defaultLayer,
+    LSP.defaultLayer,
+    Truncate.defaultLayer,
+  ),
+)
+
+const run = Effect.fn("NotebookReadTest.run")(function* (dir: string, args: Tool.InferParameters<typeof ReadTool>) {
+  return yield* provideInstance(dir)(
+    Effect.gen(function* () {
+      const info = yield* ReadTool
+      const tool = yield* Tool.init(info)
+      return yield* tool.execute(args, ctx)
+    }),
+  )
+})
+
+const put = Effect.fn("NotebookReadTest.put")(function* (filepath: string, content: string | Uint8Array) {
+  const fs = yield* AppFileSystem.Service
+  yield* fs.writeWithDirs(filepath, content)
+})
+
+const notebook = JSON.stringify({
+  metadata: { secret: "ignore-notebook-metadata" },
+  cells: [
+    {
+      cell_type: "markdown",
+      metadata: { private: "ignore-cell-metadata" },
+      source: ["# Analysis\n", "Useful introduction"],
+    },
+    {
+      cell_type: "raw",
+      source: ["ignore raw cell"],
+    },
+    {
+      cell_type: "code",
+      execution_count: 7,
+      metadata: {},
+      source: ["value = 42\n", "print(value)"],
+      outputs: [{ output_type: "stream", text: ["ignore-output-payload"] }],
+    },
+  ],
+})
+
+describe("kilocode notebook reads", () => {
+  it.live("extracts markdown and code cells without notebook payloads", () =>
+    Effect.gen(function* () {
+      const dir = yield* tmpdirScoped()
+      const filepath = path.join(dir, "analysis.ipynb")
+      yield* put(filepath, notebook)
+
+      const result = yield* run(dir, { filePath: filepath })
+
+      expect(result.output).toContain("<markdown_cell>")
+      expect(result.output).toContain("# Analysis")
+      expect(result.output).toContain("<code_cell>")
+      expect(result.output).toContain("value = 42")
+      expect(result.output.indexOf("# Analysis")).toBeLessThan(result.output.indexOf("value = 42"))
+      expect(result.output).not.toContain("ignore-output-payload")
+      expect(result.output).not.toContain("ignore-notebook-metadata")
+      expect(result.output).not.toContain("ignore-cell-metadata")
+      expect(result.output).not.toContain("ignore raw cell")
+    }),
+  )
+
+  it.live("skips invalid cells without exposing raw notebook payloads", () =>
+    Effect.gen(function* () {
+      const dir = yield* tmpdirScoped()
+      const filepath = path.join(dir, "partial.ipynb")
+      const content = JSON.stringify({
+        cells: [
+          null,
+          { cell_type: "code", source: null, outputs: ["INVALID_OUTPUT_SHOULD_NOT_APPEAR"] },
+          { cell_type: "markdown", source: ["Readable cell"], metadata: { marker: "CELL_METADATA_SHOULD_NOT_APPEAR" } },
+        ],
+      })
+      yield* put(filepath, content)
+
+      const result = yield* run(dir, { filePath: filepath })
+
+      expect(result.output).toContain("Readable cell")
+      expect(result.output).not.toContain("INVALID_OUTPUT_SHOULD_NOT_APPEAR")
+      expect(result.output).not.toContain("CELL_METADATA_SHOULD_NOT_APPEAR")
+    }),
+  )
+
+  it.live("reports valid notebooks with no readable cells without exposing payloads", () =>
+    Effect.gen(function* () {
+      const dir = yield* tmpdirScoped()
+      const filepath = path.join(dir, "empty-content.ipynb")
+      const content = JSON.stringify({
+        metadata: { marker: "NOTEBOOK_METADATA_SHOULD_NOT_APPEAR" },
+        cells: [
+          null,
+          { cell_type: "raw", source: ["RAW_CONTENT_SHOULD_NOT_APPEAR"] },
+          { cell_type: "code", source: null, outputs: ["INVALID_OUTPUT_SHOULD_NOT_APPEAR"] },
+        ],
+      })
+      yield* put(filepath, content)
+
+      const result = yield* run(dir, { filePath: filepath })
+
+      expect(result.output).toContain("Notebook contains no markdown or code cell content")
+      expect(result.output).not.toContain("NOTEBOOK_METADATA_SHOULD_NOT_APPEAR")
+      expect(result.output).not.toContain("RAW_CONTENT_SHOULD_NOT_APPEAR")
+      expect(result.output).not.toContain("INVALID_OUTPUT_SHOULD_NOT_APPEAR")
+    }),
+  )
+
+  it.live("applies read pagination to extracted cell text", () =>
+    Effect.gen(function* () {
+      const dir = yield* tmpdirScoped()
+      const filepath = path.join(dir, "paged.ipynb")
+      yield* put(filepath, notebook)
+
+      const result = yield* run(dir, { filePath: filepath, offset: 2, limit: 2 })
+
+      expect(result.output).toContain("2: # Analysis")
+      expect(result.output).toContain("3: Useful introduction")
+      expect(result.output).not.toContain("value = 42")
+      expect(result.metadata.preview).toBe("# Analysis\nUseful introduction")
+      expect(result.metadata.truncated).toBe(true)
+    }),
+  )
+
+  it.live("falls back to raw text for malformed notebooks", () =>
+    Effect.gen(function* () {
+      const dir = yield* tmpdirScoped()
+      const filepath = path.join(dir, "broken.ipynb")
+      const content = '{"cells":[{"cell_type":"markdown","source":["unfinished"]}'
+      yield* put(filepath, content)
+
+      const result = yield* run(dir, { filePath: filepath })
+
+      expect(result.output).toContain(content)
+      expect(result.output).not.toContain("<markdown_cell>")
+    }),
+  )
+
+  it.live("keeps ordinary text reads unchanged", () =>
+    Effect.gen(function* () {
+      const dir = yield* tmpdirScoped()
+      const filepath = path.join(dir, "notes.txt")
+      yield* put(filepath, "plain text")
+
+      const result = yield* run(dir, { filePath: filepath })
+
+      expect(result.output).toContain("1: plain text")
+      expect(result.output).not.toContain("<markdown_cell>")
+    }),
+  )
+
+  it.live("keeps PDF files as native attachments", () =>
+    Effect.gen(function* () {
+      const dir = yield* tmpdirScoped()
+      const filepath = path.join(dir, "document.pdf")
+      yield* put(filepath, "%PDF-1.4\nminimal content")
+
+      const result = yield* run(dir, { filePath: filepath })
+
+      expect(result.output).toBe("PDF read successfully")
+      expect(result.attachments?.[0].mime).toBe("application/pdf")
+      expect(result.metadata.truncated).toBe(false)
+    }),
+  )
+})