fix(cli): keep todo updates incremental

.changeset/incremental-todo-guidance.md

@@ -0,0 +1,5 @@
+---
+"@kilocode/cli": patch
+---
+
+Require the todo tool to update multi-step lists after each completed task.

packages/opencode/src/tool/todowrite.txt

@@ -23,6 +23,9 @@ Skip when:
 
 ## Rules
 - Update status in real time; don't batch completions
+- After creating a list, call this tool before starting the first item.
+- After completing each item, call this tool before starting the next item, with the completed item marked `completed` and the next item marked `in_progress`.
+- Do not complete multiple items or continue through multiple steps without updating the list between them.
 - Mark `completed` only after the required work is actually done, including any required verification. Never based on intent.
 - Keep exactly one `in_progress` while work remains
 - If blocked or partial, keep it `in_progress` and add a follow-up todo describing the blocker

packages/opencode/test/kilocode/todowrite-description.test.ts

@@ -0,0 +1,10 @@
+import { describe, expect, test } from "bun:test"
+import DESCRIPTION_WRITE from "../../src/tool/todowrite.txt"
+
+describe("todowrite description", () => {
+  test("requires an update between each task", () => {
+    expect(DESCRIPTION_WRITE).toContain("call this tool before starting the first item")
+    expect(DESCRIPTION_WRITE).toContain("After completing each item, call this tool before starting the next item")
+    expect(DESCRIPTION_WRITE).toContain("Do not complete multiple items or continue through multiple steps")
+  })
+})

packages/opencode/test/kilocode/todowrite-e2e.test.ts

@@ -0,0 +1,79 @@
+import { describe, expect } from "bun:test"
+import { createKiloClient } from "@kilocode/sdk/v2"
+import { Effect } from "effect"
+import { cliIt } from "../lib/cli-process"
+
+const todos = [
+  [
+    { content: "Inspect files", status: "in_progress", priority: "high" },
+    { content: "Implement fix", status: "pending", priority: "medium" },
+    { content: "Run checks", status: "pending", priority: "medium" },
+    { content: "Report results", status: "pending", priority: "low" },
+  ],
+  [
+    { content: "Inspect files", status: "completed", priority: "high" },
+    { content: "Implement fix", status: "in_progress", priority: "medium" },
+    { content: "Run checks", status: "pending", priority: "medium" },
+    { content: "Report results", status: "pending", priority: "low" },
+  ],
+  [
+    { content: "Inspect files", status: "completed", priority: "high" },
+    { content: "Implement fix", status: "completed", priority: "medium" },
+    { content: "Run checks", status: "in_progress", priority: "medium" },
+    { content: "Report results", status: "pending", priority: "low" },
+  ],
+  [
+    { content: "Inspect files", status: "completed", priority: "high" },
+    { content: "Implement fix", status: "completed", priority: "medium" },
+    { content: "Run checks", status: "completed", priority: "medium" },
+    { content: "Report results", status: "in_progress", priority: "low" },
+  ],
+  [
+    { content: "Inspect files", status: "completed", priority: "high" },
+    { content: "Implement fix", status: "completed", priority: "medium" },
+    { content: "Run checks", status: "completed", priority: "medium" },
+    { content: "Report results", status: "completed", priority: "low" },
+  ],
+] as const
+
+describe("todowrite end-to-end", () => {
+  cliIt.live(
+    "persists every sequential update through the real CLI session",
+    ({ llm, opencode }) =>
+      Effect.gen(function* () {
+        const server = yield* opencode.serve()
+        const client = createKiloClient({ baseUrl: server.url })
+        const session = yield* Effect.promise(() =>
+          client.session.create({
+            permission: [{ permission: "*", action: "allow", pattern: "*" }],
+          }),
+        )
+        const sessionID = session.data?.id
+        if (!sessionID) throw new Error("test session was not created")
+
+        for (const list of todos) yield* llm.tool("todowrite", { todos: list })
+        yield* llm.text("done")
+
+        const result = yield* opencode.run("complete this four-step task", {
+          extraArgs: ["--attach", server.url, "--session", sessionID, "--auto"],
+          timeoutMs: 90_000,
+        })
+        opencode.expectExit(result, 0)
+
+        expect(JSON.stringify(yield* llm.inputs)).toContain(
+          "After completing each item, call this tool before starting the next item",
+        )
+        const saved = yield* Effect.promise(() => client.session.todo({ sessionID }))
+        const final = todos[todos.length - 1]?.map((todo) => ({ ...todo }))
+        expect(saved.data).toEqual(final)
+
+        const messages = yield* Effect.promise(() => client.session.messages({ sessionID }))
+        const calls =
+          messages.data?.flatMap((message) =>
+            message.parts.filter((part) => part.type === "tool" && part.tool === "todowrite"),
+          ) ?? []
+        expect(calls).toHaveLength(todos.length)
+      }),
+    120_000,
+  )
+})