trunk-merge/pr-3491/69b8faac-1968-4840-8dc3-5e774280a33a

.claude/skills/quill-code/SKILL.md

@@ -0,0 +1,93 @@
+---
+name: quill-code
+description: Edit the @posthog/quill design system locally and consume the change in this repo (posthog-code) before it is published to npm. Use when changing quill components/primitives/tokens, when a quill change must be tested inside the Code app, or when the user mentions quill, the design system, the .local-quill tarball, or the @posthog/quill pnpm override.
+---
+
+# quill-code
+
+`@posthog/quill` is **not** in this repo. It is a published catalog dependency whose
+source lives in the main PostHog monorepo at `../posthog/packages/quill`. To test an
+unpublished quill change inside this repo (posthog-code), you build quill, pack it to a
+tarball, and point a pnpm `overrides` entry at that tarball. This is a **temporary
+local-dev** state — revert before merging (see below).
+
+## Quill layout (where to edit)
+
+`../posthog/packages/quill` is the **workspace** (`@posthog/quill-workspace`). It
+contains sub-packages, each a layer of the design system:
+
+- `packages/primitives/src` — base components (Card, Badge, Button, Progress, …)
+- `packages/components/src` — composed components (DataTable, DateTimePicker, Metric)
+- `packages/blocks/src` — **product-level blocks** (e.g. `ExperimentCard`). Add the
+  file here and export it from `packages/blocks/src/index.ts`.
+- `packages/quill/src` — the **aggregate** that re-exports all layers as `@posthog/quill`.
+
+A new export in any sub-package flows to `@posthog/quill` automatically on build.
+
+## The loop (every quill change)
+
+1. Edit/add the component in the right sub-package (above) and export it from that
+   package's `src/index.ts`.
+2. Re-sync into this repo manually — build → pack → point the override → reinstall:
+
+   ```bash
+   QUILL_DIR="${QUILL_DIR:-../posthog/packages/quill/packages/quill}"
+
+   # a. Build the WHOLE quill workspace (two levels up from the aggregate), so the
+   #    sub-packages rebuild BEFORE the aggregate bundles them.
+   ( cd "$QUILL_DIR/../.." && pnpm build )
+
+   # b. Pack into .local-quill/ under a UNIQUE filename. pnpm pins a tarball by
+   #    integrity, so a stable name caches stale across re-syncs — drop old local
+   #    tarballs first, then rename the packed file to a unique local name.
+   rm -f .local-quill/posthog-quill-local-*.tgz
+   ( cd "$QUILL_DIR" && npm pack --pack-destination "$(git rev-parse --show-toplevel)/.local-quill" )
+   mv .local-quill/posthog-quill-[0-9]*.tgz ".local-quill/posthog-quill-local-$(git rev-parse --short HEAD)-$$.tgz"
+
+   # c. Point the override at the new tarball, then reinstall.
+   #    Edit pnpm-workspace.yaml so overrides['@posthog/quill'] = file:./.local-quill/<new file>
+   pnpm install
+   ```
+
+   > Building only the aggregate (`packages/quill/packages/quill`) re-bundles the
+   > sub-packages' **stale** `dist/`, so edits to primitives/components/blocks are
+   > silently dropped. Always build at the workspace root.
+3. Verify in the Code app (`pnpm dev`, or the `test-electron-app` skill). Repeat from 1.
+
+After every quill edit you **must** re-run the sync — the app consumes the tarball, not
+the quill source, so unsynced edits are invisible here.
+
+If quill lives elsewhere, set `QUILL_DIR=/abs/path/to/posthog/packages/quill/packages/quill`.
+
+## Why a tarball, not `link:`
+
+`link:` symlinks into the mono's `node_modules` and drags in its **React 18** types,
+colliding with this repo's **React 19** (dual-React → broken typecheck +
+invalid-hook-call at runtime). The tarball is copied into this repo's store and deduped
+against React 19. The filename is **content-hashed** because pnpm pins a tarball by
+integrity, so a stable filename gets cached stale across re-syncs.
+
+## The override (what the script rewrites)
+
+In `pnpm-workspace.yaml`, under `overrides:`:
+
+```yaml
+'@posthog/quill': file:./.local-quill/posthog-quill-local-<hash>.tgz
+```
+
+There is also a permanent pin you should leave alone:
+
+```yaml
+'@posthog/quill>@base-ui/react': ^1.3.0   # quill ships a broken catalog: dep; do not remove
+```
+
+## Reverting (before merge)
+
+The override is local-dev only. Once the quill change is published to npm:
+
+1. Bump the catalog version in `pnpm-workspace.yaml` (`'@posthog/quill': 0.3.0-beta.x`)
+   to the published version.
+2. Restore the override line to point back at the catalog, or remove the `file:` override.
+3. `pnpm install`.
+
+Do not commit a `file:./.local-quill/...` override or the `.local-quill/` tarballs.

apps/code/snapshots.yml

@@ -96,6 +96,26 @@ snapshots:
         hash: v1.k4693efd2.b76d39ed8142d4449cb5befcde538bb11a6bffcbed66e81bdc3aa8e2746b7e06.GSIfBsgE-3RYuqlLUpluZ87C6-KjBUaTj2eu_a5elZ8
     autoresearch-runtime-stats--with-context-usage--light:
         hash: v1.k4693efd2.18889b5eda942cd19b0f7be00a87925d632529ca4111fac180d6339f7d34d1ca.X3s9IXrx2-TdQaJEjf1td3hwvNzeIp3Xq3uQh9Gwt5A
+    billing-usagemeter--exceeded--dark:
+        hash: v1.k4693efd2.dc026c5cace4e44c4cc713b95654e278d7bf0439a4862e9ceea39a033974815d.m5Be4dAOwwmt-UlEnhbhrKiym1rlGUiGZ4oyUonUoVg
+    billing-usagemeter--exceeded--light:
+        hash: v1.k4693efd2.850bf1e451a18981d9e3be653837dd282c6be289ac9c0826ff6a981ac67e1e38.1GwpfJ_4dUlMO7TXp8AOM5YmJuItGGB50ihKkkeIFgg
+    billing-usagemeter--free-tier--dark:
+        hash: v1.k4693efd2.164d54cac22cfa05dbbf25fae1a62e77c496b1981dcaff6b8e1ab07fc802f5e0.Gh75uI1vHbGWAgKun41BYoX6LZBYX_FVI7Qu-rFlDV4
+    billing-usagemeter--free-tier--light:
+        hash: v1.k4693efd2.5e220fe66fac4c54b4cb9f5ac6ec1054e0b709374fee2c58fdba782f5b2888df.W5la88YA6uOSc1tg30raKaRy12plPGxXl5o7uowL2gc
+    billing-usagemeter--subscribed-past-included--dark:
+        hash: v1.k4693efd2.ac987b7184aaa4f8895d9b75c6b7e1177efcb9a21cca0668caf533ee0d5eaab3.2Mq2GLXV83ebM8LKhTQY8csv6BeskWv5yPu8NSAQ20Y
+    billing-usagemeter--subscribed-past-included--light:
+        hash: v1.k4693efd2.0c4a37beaceb37baab57e6da98321b236c20127731fd5eaeb61677de4580c204.fZ2eDGlkEijZjHQyatiHtP39EFYT6EO1Vn_f8Tz7b0E
+    billing-usagemeter--subscribed-with-breakdown--dark:
+        hash: v1.k4693efd2.f74f8f76e51ee14e427423e1757859e3f001b13e6a43d1e7310b5b0cbbb570b5.eMwK2csA86oBarnAya99W87MfP6Z6PMfdjwbCmnPrw4
+    billing-usagemeter--subscribed-with-breakdown--light:
+        hash: v1.k4693efd2.1335dba8bf16980c987882641619e114584343ec50bb69b3c01e63507a13756b._v7_6tsVMmW3WixSeFN-puHQVeIxf9AAGxFZ8Ni97YQ
+    billing-usagemeter--zero-spend-limit--dark:
+        hash: v1.k4693efd2.1d199b2c4bba8034cb18fb5b49866b64a6eb3311591add5f58ec288710dce2e9.FCc9egSaO1Onih417yciSYKNROe8zqDZyyS6xuXCZGw
+    billing-usagemeter--zero-spend-limit--light:
+        hash: v1.k4693efd2.b50c18736bbde45fc89f4c0ac7fd616286e5c1ecc2a56910fffca13eaa13d8a2.31z4F0JmvMeR4yqk5pjs_OQ3MypnM2miZI9al4wuc8c
     components-permissions-permissionselector--create-new-file--dark:
         hash: v1.k4693efd2.c54203a4e636b83b3d24d7ed9c4ace8659db87cd8f231d9dc2ecc03320e31646.epDm7LebiLzlp0uuZBrE-Obt_anAn0xsE8bHFnm5vos
     components-permissions-permissionselector--create-new-file--light:

apps/code/src/renderer/desktop-contributions.ts

@@ -1,6 +1,7 @@
 import { agentChatCoreModule } from "@posthog/core/agent-chat/agentChat.module";
 import { autoresearchCoreModule } from "@posthog/core/autoresearch/autoresearch.module";
 import { billingCoreModule } from "@posthog/core/billing/billing.module";
+import { taskThreadCoreModule } from "@posthog/core/canvas/taskThread.module";
 import { inboxCoreModule } from "@posthog/core/inbox/inbox.module";
 import { githubConnectModule } from "@posthog/core/integrations/githubConnect.module";
 import { localMcpCoreModule } from "@posthog/core/local-mcp/local-mcp.module";
@@ -37,6 +38,7 @@ export function registerDesktopContributions(): void {
     autoresearchCoreModule,
     billingUiModule,
     billingCoreModule,
+    taskThreadCoreModule,
     browserTabsUiModule,
     cloneUiModule,
     connectivityUiModule,

apps/web/src/web-container.ts

@@ -1,5 +1,6 @@
 import "reflect-metadata";
 import { TypedContainer } from "@inversifyjs/strongly-typed";
+import { taskThreadCoreModule } from "@posthog/core/canvas/taskThread.module";
 import { setRootContainer } from "@posthog/di/container";
 import { ROOT_LOGGER, type RootLogger } from "@posthog/di/logger";
 import {
@@ -96,4 +97,6 @@ container.bind(MCP_SANDBOX_PROXY_URL).toConstantValue(() => {
   return sandboxProxyUrl;
 });
 
+container.load(taskThreadCoreModule);
+
 setRootContainer(container);

packages/agent/src/adapters/claude/session/rtk.ts

@@ -15,15 +15,17 @@ import { gitSubcommand } from "../git-command";
 
 // Commands RTK compresses faithfully and that have no side effects, so wrapping
 // them changes only how much output reaches the model, never what runs.
-const RTK_PLAIN_COMMANDS = new Set(["grep", "find", "ls"]);
+// Exported so the instruction-level Codex guidance advertises the same set.
+export const RTK_PLAIN_COMMANDS = new Set(["grep", "find", "ls"]);
 
 // Git subcommands whose output is worth compressing and that RTK handles
 // faithfully. The criterion is compressible output, NOT read-only: RTK never
 // changes what runs, so a mutating form (`git tag -d`, `git remote add`,
 // `git reflog expire`) still executes its write — its output is just shorter.
 // Excludes commit/push: negligible output to compress, and the cloud
 // signed-commit guard keys on a leading `git` token that `rtk git …` would hide.
-const GIT_COMPRESSIBLE_SUBCOMMANDS = new Set([
+// Exported so the instruction-level Codex guidance advertises the same set.
+export const GIT_COMPRESSIBLE_SUBCOMMANDS = new Set([
   "status",
   "diff",
   "log",
@@ -44,7 +46,8 @@ const GIT_COMPRESSIBLE_SUBCOMMANDS = new Set([
 // wrapping only its head would change the meaning of the rest.
 const SHELL_OPERATORS = /[|&;<>`\n]|\$\(/;
 
-function shQuote(value: string): string {
+// Exported so the instruction-level Codex guidance quotes the prefix the same way.
+export function shQuote(value: string): string {
   if (/^[\w./-]+$/.test(value)) return value;
   return `'${value.replace(/'/g, `'\\''`)}'`;
 }

packages/agent/src/adapters/codex-app-server/codex-app-server-agent.test.ts

@@ -1961,9 +1961,23 @@ describe("CodexAppServerAgent", () => {
     } as unknown as PromptRequest);
 
     // The single turn/completed resolves both the original and the folded prompt.
+    stub.emit("thread/tokenUsage/updated", {
+      tokenUsage: {
+        last: {
+          totalTokens: 45,
+          inputTokens: 30,
+          cachedInputTokens: 5,
+          outputTokens: 10,
+        },
+      },
+    });
     stub.emit("turn/completed", { turn: { status: "completed" } });
-    expect((await first).stopReason).toBe("end_turn");
-    expect((await second).stopReason).toBe("end_turn");
+    const [firstResult, secondResult] = await Promise.all([first, second]);
+    expect(firstResult).toMatchObject({
+      stopReason: "end_turn",
+      usage: { totalTokens: 45 },
+    });
+    expect(secondResult).toEqual({ stopReason: "end_turn" });
 
     const steer = stub.requests.find((r) => r.method === "turn/steer");
     expect(steer?.params).toMatchObject({
@@ -2126,7 +2140,19 @@ describe("CodexAppServerAgent", () => {
       },
     });
     stub.emit("turn/completed", { turn: { status: "completed" } });
-    await done;
+    const result = await done;
+
+    expect(result).toEqual({
+      stopReason: "end_turn",
+      usage: {
+        inputTokens: 60,
+        outputTokens: 30,
+        cachedReadTokens: 10,
+        cachedWriteTokens: 0,
+        thoughtTokens: 5,
+        totalTokens: 100,
+      },
+    });
 
     const turnComplete = extNotifications.find(
       (n) => n.method === "_posthog/turn_complete",
@@ -2818,6 +2844,16 @@ describe("CodexAppServerAgent", () => {
         text: "The implementation plan is ready.",
       },
     });
+    stub.emit("thread/tokenUsage/updated", {
+      tokenUsage: {
+        last: {
+          totalTokens: 30,
+          inputTokens: 20,
+          outputTokens: 10,
+          reasoningOutputTokens: 2,
+        },
+      },
+    });
     stub.emit("turn/completed", {
       turn: { id: "turn_1", status: "completed" },
     });
@@ -2826,10 +2862,31 @@ describe("CodexAppServerAgent", () => {
     await waitUntil(
       () => stub.requests.filter((r) => r.method === "turn/start").length >= 2,
     );
+    stub.emit("thread/tokenUsage/updated", {
+      tokenUsage: {
+        last: {
+          totalTokens: 50,
+          inputTokens: 35,
+          cachedInputTokens: 5,
+          outputTokens: 10,
+          reasoningOutputTokens: 3,
+        },
+      },
+    });
     stub.emit("turn/completed", {
       turn: { id: "turn_2", status: "completed" },
     });
-    expect((await done).stopReason).toBe("end_turn");
+    expect(await done).toEqual({
+      stopReason: "end_turn",
+      usage: {
+        inputTokens: 55,
+        outputTokens: 20,
+        cachedReadTokens: 5,
+        cachedWriteTokens: 0,
+        thoughtTokens: 5,
+        totalTokens: 80,
+      },
+    });
 
     // The approval renders as the plan-approval UI (switch_mode + the plan text).
     expect(permissionRequests).toHaveLength(1);