> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
**Goal:** Build a new standalone Vendetta plugin, `TypingAvatars`, that replaces Discord's "X is typing..." indicator with a row of the avatars of whoever's currently typing.
**Architecture:** Reuses the proven-safe `TypingIndicatorInner` interception already verified in `role-color-everywhere/src/patches/typingWrapper.ts` (this repo, same session): catch the component via `registerTypeDetector`, wrap it so the original still renders exactly once (preserving its hooks), find the label node that used to hold the typing text, and replace its children with a new `AvatarStack` component built from `props.typingUserIds`.
**Tech Stack:** TypeScript, React (via `@vendetta/metro/common`), React Native (`View`/`Image`), this repo's shared `@shared/lib/createElementIntercept` and `@shared/lib/patcher` helpers, Rollup/SWC build (`node build.mjs`).
## Global Constraints
- No test runner exists in this repo (`package.json` has only a `build` script) — every plugin here is verified by `node build.mjs` (bundling/syntax correctness) plus live on-device verification via `revenge-devtools`, not unit tests. This plan follows that same convention.
- Every patch point must be wrapped in try/catch that leaves the original behavior intact on failure, matching every other plugin in this repo (see `role-color-everywhere`, `copy-role-color`, `server-drawer`).
-`channel.guild_id` (snake_case) is the confirmed field for a channel's guild id — verified in `typingWrapper.ts:46` and `rows.ts:39`. Do not use `channel.guildId`.
-`GuildMemberStore.getMember(guildId, userId)` and `UserStore` (via `findByStoreName`) are confirmed-working store lookups elsewhere in this repo — but the exact avatar-hash property name on the returned member/user record (`member.avatar` / `user.avatar`) has **not** been live-verified. Flag this clearly in code comments; confirm in Task 4.
- CDN avatar URLs are hand-built directly (`https://cdn.discordapp.com/...`), matching this repo's own established pattern in `server-drawer/src/components/GuildIcon.tsx:26` — not a guessed Discord internal "IconUtils" module lookup.
-`revenge-devtools` was disconnected while this plan was written. Task 4 (live verification) cannot run until it reconnects — do not claim the plugin "works" before that task is actually completed against a live device.
---
### Task 1: Scaffold the plugin
**Files:**
- Create: `plugins/typing-avatars/manifest.json`
- Create: `plugins/typing-avatars/src/index.ts`
**Interfaces:**
- Produces: a default-exported plugin object (`{ onLoad, onUnload }`) that later tasks wire their patch functions into.
- Consumes: `getTypingAvatarURL` from Task 2 (`../lib/avatarUrl`).
- Produces: `AvatarStack` (default export, props `{ typingUserIds: string[]; guildId: string | undefined }`) and `patchTypingIndicator` (default export, `() => () => void`, same shape as every other patch function in this repo).
This mirrors `role-color-everywhere/src/patches/typingWrapper.ts` exactly for the detection/wrapping part (already proven live in this repo), but replaces the label's content with `AvatarStack` instead of recoloring the existing per-user text nodes. Unlike the color-matching version, this doesn't need to zip `typingUserIds` against per-user text elements at all - it only needs `props.typingUserIds` directly, so it isn't affected by the "several people are typing..." collapsed-string case that made the color patch bail out.
```tsx
import { React } from "@vendetta/metro/common";
import { findInReactTree } from "@vendetta/utils";
import { registerTypeDetector, registerIntercept, patchCreateElement } from "@shared/lib/createElementIntercept";
import AvatarStack from "../components/AvatarStack";
export default function patchTypingIndicator(): () => void {
Via `revenge-devtools``eval`, install a one-off probe patching `TypingIndicatorInner`'s call count (same technique used earlier this session for `useIsMobileQuestDockRendered`), then have the user start typing in a channel with someone else. Confirm the patch's `console.log`/counter actually fires - if it doesn't, the type-name check (`type.name === "TypingIndicatorInner"`) needs re-verification against the current Discord build, the same way this session discovered `QuestDockContentExpanded` had silently drifted.
Via `eval`, call `GuildMemberStore.getMember(guildId, userId)` and `UserStore.getUser(userId)` for a real user who is currently typing, and inspect the returned object's keys directly (`Object.keys(member)`) rather than assuming `.avatar` is correct. Fix `avatarUrl.ts` if the real field name differs, per the Global Constraints note above.
Have the user get 1, 2, and 3+ people typing simultaneously in a test channel (or DM group) and screenshot each state. Confirm: avatars render (not broken images), the overlap direction looks right, and 3+ typers wrap to a second row rather than clipping off-screen.