stream-flutter — Claude Code skill for Stream in Flutter
This skill helps build Flutter apps with Stream Chat, Video, and Feeds. Use it to start a project, integrate Stream into an existing app, or migrate from Sendbird.
npx skills add GetStream/agent-skills --skill stream-flutter -a claude-codeWhat stream-flutter does
The skill routes requests to one of several workflows: create an app, add Stream to an existing project, set up packages and authentication, or migrate from the Sendbird Chat SDK. API and widget questions go through a reference lookup path. The agent reads RULES.md first.
For chat, it covers stream_chat_flutter with prebuilt UI and stream_chat_flutter_core for custom UI. Calls and livestreams use stream_video_flutter; activity feeds use stream_feed or stream_feed_flutter_core, which have no prebuilt UI. That matters if your project needs ready-made feed screens. The skill works in an existing Flutter project or when scaffolding a new one, and does not require Stream CLI.
What it helps with
- add Stream Chat to a Flutter app
- build calling or livestream features with Stream Video
- integrate an activity feed with Stream Feeds
- migrate a Flutter app from Sendbird Chat SDK to Stream
Example prompts for the agent
- Add Stream Chat to my Flutter project and wire up authentication.
- Create a Flutter video calling app with Stream.
- Migrate chat in my Flutter app from Sendbird to Stream.
How to install stream-flutter
1Skills CLI (Vercel)
npx skills add GetStream/agent-skills --skill stream-flutter -a claude-code2Manually: put the files in place
~/.claude/skills/stream-flutter/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/stream-flutter/
1Skills CLI (Vercel)
npx skills add GetStream/agent-skills --skill stream-flutter -a codex2Manually: put the files in place
.agents/skills/stream-flutter/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add GetStream/agent-skills --skill stream-flutter -a cursor2Manually: put the files in place
.cursor/skills/stream-flutter/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install GetStream/agent-skills stream-flutter2Manually: put the files in place
.github/skills/stream-flutter/SKILL.md1Skills CLI (Vercel)
npx skills add GetStream/agent-skills --skill stream-flutter -a gemini-cli2Platform CLI
gemini skills install https://github.com/GetStream/agent-skills --path skills/stream-flutterManually: put the files in place
.devin/skills/stream-flutter/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add GetStream/agent-skills --skill stream-flutter -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about stream-flutter
How do I install stream-flutter in Claude Code?
Run npx skills add GetStream/agent-skills --skill stream-flutter -a claude-code. The skill file is at ~/.claude/skills/stream-flutter/SKILL.md.
What does stream-flutter do?
It helps build Flutter apps with Stream Chat, Video, and Feeds, integrate those products into an existing project, and migrate from the Sendbird Chat SDK.
Does stream-flutter require Stream CLI?
No. It works with an existing Flutter project or can scaffold a new one without Stream CLI.
Source text
SKILL.md · Apache-2.0
---
name: stream-flutter
description: "Build and integrate Stream Chat, Video, and Feeds in Flutter apps. Use for Flutter/Dart project work with Stream package setup, auth wiring, and widget blueprints. Supports stream_chat_flutter (pre-built Chat UI), stream_chat_flutter_core (custom Chat UI), stream_video_flutter (Video calling and livestreaming), and stream_feed / stream_feed_flutter_core (Activity Feeds, no pre-built UI). Also migrates Flutter apps from the Sendbird Chat SDK (sendbird_chat_sdk / sendbird_uikit) to Stream Chat."
license: See LICENSE in repository root
compatibility: Works in an existing Flutter project, or scaffolds a new one. No Stream CLI required.
metadata:
author: GetStream
allowed-tools: >-
Read, Write, Edit, Glob, Grep,
Bash(ls *),
Bash(grep *),
Bash(find . *),
Bash(cat pubspec.yaml), Bash(cat pubspec.lock),
Bash(flutter pub *),
Bash(flutter create *),
Bash(flutter run *),
Bash(flutter analyze *), Bash(flutter analyze),
Bash(flutter build *),
Bash(flutter devices),
Bash(flutter emulators *), Bash(flutter emulators),
Bash(getstream token *),
Bash(getstream env *),
Bash(getstream api *),
Bash(getstream login *),
Bash(getstream init *)
---
# Stream Flutter - skill router + execution flow
**Rules:** Read **[`RULES.md`](RULES.md)** once per session - every non-negotiable rule is stated there, nowhere else.
This file is the **single entrypoint**: intent classification, local project detection, and module pointers for Stream work in Flutter apps.
---
## Step 0: Intent classifier (mandatory first - never skip)
Before any tool call, decide the **track** from the user's input alone - no probes first.
### Signals -> track
| Signal in user input | Track |
| ------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------- |
| Explicit package/widget token: `stream_chat_flutter`, `StreamChannelListView`, `StreamMessageListView`, `StreamChatClient`, etc. | **C - Reference lookup** |
| Explicit video token: `stream_video_flutter`, `StreamCallContainer`, `StreamVideo`, `StreamVideoRenderer`, `goLive`, `stopLive`, `livestream` call type | **C - Reference lookup** |
| Explicit feeds token: `stream_feed`, `stream_feed_flutter_core`, `StreamFeedClient`, `FlatFeedCore`, `FlatFeed`, `FeedBloc`, `activity feed`, `feeds flutter` | **C - Reference lookup** |
| Words "docs" or "documentation" around Stream Flutter work | **C - Reference lookup** |
| "How do I {X} in Flutter?", "What does {widget/method} do?" | **C - Reference lookup** |
| "Build me a new Flutter app", "create a Flutter chat app" + Stream | **A - New app** |
| "Build a Flutter video call app", "create a livestream app in Flutter" | **A - New app** (load `VIDEO-FLUTTER.md` + `VIDEO-FLUTTER-blueprints.md` or `LIVESTREAM-FLUTTER.md` + `LIVESTREAM-FLUTTER-blueprints.md`) |
| "Build an audio room / Twitter Spaces clone", "TikTok-style live feed", "call while livestreaming", "chat with video calls", "two calls at once" | **A or B** (load `VIDEO-ADVANCED-FLUTTER.md` + `VIDEO-ADVANCED-FLUTTER-blueprints.md` on top of the Video/Livestream pair) |
| "Add ringing / incoming calls", "video call with push notifications", "CallKit", "VoIP push", "FCM ringing", "missed call notification" | **A or B** (load `RINGING-FLUTTER.md` + `RINGING-FLUTTER-blueprints.md` on top of the Video pair) |
| "Build a Flutter feeds app", "create an activity feed app", "build a social feed in Flutter", "create a Twitter/Instagram clone" | **A - New app** (load `FEEDS-FLUTTER.md` + `FEEDS-FLUTTER-blueprints.md`; use Twitter-style UI unless the user explicitly specifies otherwise) |
| "Add/integrate Stream into this app", "wire Chat into my Flutter project" | **B - Existing app** |
| "Add video calling to my Flutter app", "integrate Stream Video into my existing app" | **B - Existing app** (load `VIDEO-FLUTTER.md` + `VIDEO-FLUTTER-blueprints.md`) |
| "Add a feed to my Flutter app", "integrate Stream Feeds into my existing app", "add activity feed" | **B - Existing app** (load `FEEDS-FLUTTER.md` + `FEEDS-FLUTTER-blueprints.md`; use Twitter-style UI unless the user explicitly specifies otherwise) |
| "Install Stream packages", "set up Stream in Flutter", "wire auth/token" with no broader feature request | **D - Bootstrap / setup** |
| "Migrate my Flutter app from Sendbird", "replace `sendbird_chat_sdk` / `sendbird_uikit` with Stream", "switch off Sendbird / SendbirdUIKit" | **M - Migrate from Sendbird** ([`sendbird-migration.md`](sendbird-migration.md)) |
| Bare `/stream-flutter` with no args | List the tracks briefly and wait |
### Styling-depth flag (orthogonal to Tracks A/B/C/D)
If the request carries a **target appearance** — an attached screenshot, a Figma frame, or "make it look like WhatsApp / iMessage / Telegram / Slack / <app>" — then **before** feature work (Track A or B), **first run the strategy decision below** (components vs custom):
- **Components** → run [`design-matching.md`](design-matching.md): read its **procedure half in full** — everything above the `# Reference` divider (Steps 1–5; it may span two Reads, so page past a mid-procedure truncation) — then Grep its Reference half per region while implementing.
- **Custom** (livestream / overlay / bespoke) → run [`custom-ui.md`](custom-ui.md) instead — same region-by-region rigor, different mechanism.
The rest of this section is the **components** case (the common one). A reference design is a checklist of regions (header, composer buttons, where the timestamp + read receipts sit, bubble shape/tail, date separators, attachments...), and most of them differ from Stream's defaults **structurally**, not just by color. Do **not** stop at the wallpaper and bubble color — that is the known failure mode. Decompose every region first (capturing its **dimensions**, not just colors), then route each to one of **two axes**:
1. **Theming** — `StreamTheme` (a `ThemeExtension` on `MaterialApp.theme.extensions`) owns design foundations (spacing / radius / typography / colors) **and** all fine-grained/leaf styling including the whole message row (`messageItemTheme` bubble/text/attachment/metadata, `reactionsTheme`, `reactionPickerTheme`, `avatarTheme`, `textInputTheme`, `mediaViewerTheme`, …), while `StreamChatThemeData` (passed to `StreamChat(themeData:)`) owns only the chat composite-widget slots (the three headers, message-list background/wallpaper, channel-list item, quoted message, thread tile, poll themes).
2. **Widget replacement** — the component factory, populated via **core named slots** on `StreamComponentBuilders(...)` (~48 leaf slots: `messageBubble`, `messageText`, `reactions`, `reactionPicker`, `mediaViewer`, `jumpToUnreadButton`, `textInput`, `avatar`, …) **and** **chat slots** via `streamChatComponentBuilders(...) → extensions:` (~30 composite slots: `messageItem`, `messageComposer` + sub-slots, `channelListItem`, per-attachment builders, …), passed to `StreamChat(componentBuilders: StreamComponentBuilders(...))` (global) or `StreamComponentFactory(builders: …, child: …)` (scoped); plus per-widget builders (`messageBuilder`, `itemBuilder`) for a single instance.
Recurring traps the doc guards against:
- **(a)** overriding a **composite widget** (`messageItem` builder, `messageComposer` builder, `messageBuilder` on `StreamMessageListView`) silently drops the sub-features the default rendered — the incoming-message avatar, grouping, reactions, replies, status, or the send/voice/edit/slow-mode button — unless you read the default's `build()` and reproduce them (grouping state is available inside a custom `messageItem` via `StreamMessageLayout.of(context)`);
- **(b)** **model-driven title logic** must live in one shared helper used by both the channel list `itemBuilder` and the header, so the two surfaces cannot diverge;
- **(c)** route deliberately — bubble padding / color / shape live on **`StreamTheme`** (`StreamTheme.messageItemTheme.bubble`); structural changes (send button outside the field, metadata inside the bubble, a bubble tail, a Slack-style flat row) need **widget replacement** (component factory or a per-widget builder).
The match is **not done until you run, seed data that triggers every region, compare region-by-region against the reference on the real navigation path, and iterate** ([`design-matching.md`](design-matching.md) Step 5), deleting any throwaway verification scaffold — the UI must be as close to the reference as possible, not approximately like it. **Implement every region, the composer included** — never deliver a partial match with the rest labelled "known cosmetic gaps"; a region left at the SDK default is a FAIL, not a footnote. And **work in batches**: ground the pinned SDK version + local checkout once, read the source you need in one pass, implement all regions, then verify once on hot reload — don't rebuild-and-screenshot after every small edit.
### Chat only: pick the UI strategy first (before any code)
Stream Chat ships **two layers**, and choosing between them is an **architecture decision that dwarfs any styling choice** — getting it wrong wastes a day either way. So decide deliberately, and **default to the pre-built components**:
| Strategy | Use when the design is... | Mechanism | Runbook |
|---|---|---|---|
| **Pre-built UI components** (`stream_chat_flutter`) — **the default** | A messenger: bubbles, or a channel list → conversation, or per-message avatar/timestamp/receipts/reactions/attachments. Social, marketplace, workplace, support, DMs. "Make it look like WhatsApp / iMessage / Telegram / Slack." | Customize via the two axes — theming (`StreamTheme` for foundations + the message row/leaf widgets, `StreamChatThemeData` for the chat composite widgets) + widget replacement (the component factory `StreamComponentBuilders` / `streamChatComponentBuilders` / `StreamComponentFactory`, plus per-widget builders) | [`design-matching.md`](design-matching.md) |
| **Custom UI on `stream_chat_flutter_core`** — the exception | Not a messenger: a flat bubble-less author-inline feed, an overlay/ticker on video, high-volume ephemeral livestream chat (Twitch / YouTube / Kalshi), live shopping, or anonymous/guest read-only viewers vastly outnumbering posters (→ `livestream` channel type). Every message rendered identically; bespoke app chrome around it. | Build your own widgets on the headless `stream_chat_flutter_core` controllers (+ the low-level client); **no** `stream_chat_flutter` widgets | [`custom-ui.md`](custom-ui.md) |
**Lean hard toward components.** They're built to be *customized*, and the litmus test is: *if the two axes — theming (`StreamTheme` + `StreamChatThemeData`) + a few component-builder slots — could get there, it's a components job* — even strong messenger reskins. Pick **custom only** when matching the design would mean replacing the message row, composer, header, AND list all at once — i.e. you'd be using the SDK purely as a data source, not for any of its widgets. Over-choosing custom (rebuilding a worse messenger by hand, losing avatars/grouping/reactions/threads/attachments/typing/receipts/pickers) is the common, expensive mistake; over-choosing components costs a few hours of fighting layout. **When unsure, build the components version first** — it's faster to confirm-or-reject. And treat **"livestream" / "live-shopping" as signals, not labels**: they point to custom only when the *shape* matches (flat identical rows, overlay/ticker, high volume, read-only viewers) — a livestream app that wants a normal bubble/channel-list chat panel stays on **components**. This is the full decision rubric; [`custom-ui.md`](custom-ui.md) is the build runbook you follow *after* the decision lands on custom, not a doc you open to decide.
**Workplace / Slack-style hybrid is a components job too.** A Slack/Teams/Discord surface is a channel list + message list + composer, so it stays on the pre-built components — but two things differ from a messenger and must be matched, not punted: (1) the **message row** is flat and left-aligned (avatar-top rounded-square, an author·custom-status·timestamp header line, body, **bottom reaction pills**, and a **thread-reply summary**), with **no incoming/outgoing bubble split** — reproduce it by overriding the `messageItem` component builder (a composite slot — reproduce its sub-features); (2) the **header** and chrome are custom. The full workplace archetype is in [`design-matching.md`](design-matching.md). Workplace apps are also **thread-first**, so wire the thread-reply summary and thread screen.
The strategy also picks the **channel type and permission model** (e.g. `messaging` membership-gated for social/marketplace vs `livestream` public + anonymous viewers) — see [`RULES.md`](RULES.md) → "Surface permission prerequisites proactively", and Step 0.5 → "Permissions awareness". Decide both axes together.
If it's genuinely unclear, ask one question:
> Does this chat look like a standard messenger (channel list + bubbles), or a bespoke surface like livestream/overlay chat? It decides whether we customize the pre-built components or build custom UI on `stream_chat_flutter_core`.
### Disambiguation flow
If the request is ambiguous between **build/integrate** and **reference lookup**, ask one short question and wait:
> Do you want me to wire this into the project, or just map the Flutter SDK pattern and widgets?
### After classification
- **Tracks A, B, D** -> run **Project signals** once per session, then continue in [`builder.md`](builder.md) and [`sdk.md`](sdk.md). If the styling-depth flag was raised, run the design-match phase alongside per the UI-strategy decision — [`design-matching.md`](design-matching.md) for the pre-built components (the common case; read its procedure half in full — everything above the `# Reference` divider, which may span two Reads — then Grep its Reference half per region) or [`custom-ui.md`](custom-ui.md) for a bespoke `stream_chat_flutter_core` surface (design-match rigor is a mandatory phase, not an optional add-on).
- **Track C** -> skip the probe if the product + package are explicit. Only run it on demand if the SDK layer is ambiguous.
- **Track M** -> run [`sendbird-migration.md`](sendbird-migration.md): detect the existing Sendbird integration shape, swap packages/init/views/channels/messages, re-apply theming via the two axes, and verify design + functional fidelity per screen against the Sendbird original. Reuses Step 0.5 credentials and [`design-matching.md`](design-matching.md)/[`custom-ui.md`](custom-ui.md) rigor; hands off to the shared data-migration runbook when the code migration is done.
---
## Step 0.5: Credentials, token, and seed data (tracks A, B, D — and Track M at phase M2)
Run this once per session, right after intent classification, before the Project signals probe.
### Goal
Collect the Stream **API key**, a **user token**, and optionally seed channels or calls - all before touching code - so the app has real data to show from the first run.
### Single upfront question (ask exactly once, then act immediately)
Post **one message** asking all relevant things together. Do not split into multiple rounds.
**For Chat projects:**
> To wire everything up with real data, I need a few quick answers:
>
> 1. **Credentials** - Should I fetch your API key from the dashboard and generate a token via the Stream CLI, or will you paste them yourself?
> 2. **Token expiry** - If I'm generating the token: should it expire? (e.g. `1h`, `1d`, `30m`) or never expire?
> 3. **Seed channels** - Should I pre-create a few channels with random usernames so the app has something to show immediately?
>
> If you want to handle everything yourself, just paste your API key and token and tell me whether to seed channels.
**For Video projects** (calls are ephemeral - no seeding needed):
> To wire everything up, I need a couple of quick answers:
>
> 1. **Credentials** - Should I fetch your API key from the dashboard and generate a token via the Stream CLI, or will you paste them yourself?
> 2. **Token expiry** - If I'm generating the token: should it expire? (e.g. `1h`, `1d`, `30m`) or never expire?
>
> If you want to handle everything yourself, just paste your API key and token.
> **Guest-viewer requirement — surface this BEFORE building any app that signs viewers in as guests**. A guest connects with the `guest` role, which by default has minimal capabilities. On the `livestream` call type guests **cannot even read or join a call** until the integrator grants those capabilities to the `guest` role. When a guest-based viewer flow is in scope, tell the integrator it is a **prerequisite** to grant the `guest` role `read-call` and `join-call` (plus `create-call` if a viewer may open the call before the host) on the relevant call type via Stream Dashboard → Video & Audio → Call Types → <type> → Roles & Permissions (or the API), or to use authenticated `User.regular` viewers instead. Details: [`references/VIDEO-FLUTTER.md`](references/VIDEO-FLUTTER.md) → Guest users, and [`references/LIVESTREAM-FLUTTER.md`](references/LIVESTREAM-FLUTTER.md) → Roles, permissions, and backstage security.
**For Feeds projects** (no pre-built UI; feed groups required):
Ask **one message** with all setup questions together — do not split into rounds:
> To wire everything up, I need a few quick answers:
>
> 1. **Credentials** - Should I fetch your API key from the dashboard and generate a token via the Stream CLI, or will you paste them yourself?
> 2. **Token expiry** - If I'm generating the token: should it expire? (e.g. `1h`, `1d`, `30m`) or never expire?
> 3. **Feed groups** - I need to create 3 feed groups in your Stream project (user, timeline, notification). Should I set these up automatically, or have you already created them?
> 4. **Seed posts** - Should I add a few sample posts so the feed has content from the first run?
>
> If you want to handle credentials yourself, just paste your API key and token.
Once the user replies, execute all steps without pausing. For feed groups, if the user said "set up automatically":
```bash
getstream api CreateFeedGroup --request '{"id": "user", "type": "flat"}'
getstream api CreateFeedGroup --request '{"id": "timeline", "type": "flat"}'
getstream api CreateFeedGroup --request '{"id": "notification", "type": "notification"}'
```
If the CLI commands fail (the Feeds API may use different endpoints than Chat), tell the user once:
> Please create these in Stream Dashboard → Activity Feeds → Feed Groups: `user` (Flat), `timeline` (Flat), `notification` (Notification).
For Feeds projects, always generate two separate helpers in `main` after `connect()`:
1. `_setupFollows(client)` — **always called, unconditionally.** Makes `timeline` follow `user` so the user's own posts appear there. Do not merge this into seed logic — once seed data exists the guard returns early and the follow call never runs.
2. `_seedPosts(client)` — only if the user said yes to seeding. Adds sample activities and exits early if data already exists.
See [`references/FEEDS-FLUTTER.md`](references/FEEDS-FLUTTER.md) for both implementations.
The package is `stream_feeds: ^0.5.1` — not the deprecated `stream_feed` or `stream_feed_flutter_core`.
### After the user replies - act without further prompting
Once the user answers, execute all CLI steps in sequence **without pausing for confirmation between them**. Narrate each step briefly as you go (one line per action), but do not stop to ask "shall I continue?".
#### Step A - API key
```bash
getstream env --target flutter
```
This writes the public API key to `dart_defines.json`; the app reads it via `String.fromEnvironment('STREAM_API_KEY')` and is run with `flutter run --dart-define-from-file=dart_defines.json`. If the command returns a 401 error, the CLI session has expired - run `getstream login` to re-authenticate, then retry.
**If `getstream` is not installed** (`command not found`): ask the user to install it from https://getstream.io and wait. Or, if the user prefers, skip the CLI entirely and have them paste the API key + a token per user (Dashboard -> Explorer has a token generator). Decide based on the user's answer to the upfront credentials question; don't stall.
#### Step B - Token
```bash
# Never-expiring
getstream token <user_id>
# Expiring
getstream token <user_id> --ttl <duration>
```
Hold the token in context. Use it (and the API key) in every code snippet - no placeholder strings.
#### Step C - Seed channels (only if the user said yes)
Create 3-5 channels with random realistic usernames. Use `messaging` as the default channel type.
**Sub-step C1 — upsert all users** (seed users + the token user):
```bash
getstream api UpdateUsers --request '{
"users": {
"<token_user_id>": {"id": "<token_user_id>", "name": "<Display Name>"},
"alice": {"id": "alice", "name": "Alice"},
"bob": {"id": "bob", "name": "Bob"},
"carol": {"id": "carol", "name": "Carol"},
"dave": {"id": "dave", "name": "Dave"}
}
}'
```
**Sub-step C2 — create each channel** (no members in the body; members are added in C3):
```bash
getstream api GetOrCreateChannel --type messaging --id <channel-id> \
--request '{"data": {"created_by_id": "<token_user_id>", "custom": {"name": "<Channel Name>"}}}'
```
Repeat for each channel (e.g. `general`, `random`, `team-alpha`).
**Sub-step C3 — add members to each channel** using `add_members`. The token user **must** be in every channel so the `Filter.in_('members', [userId])` query in the app returns results.
```bash
getstream api UpdateChannel --type messaging --id <channel-id> \
--request '{
"add_members": [
{"user_id": "<token_user_id>"},
{"user_id": "alice"},
{"user_id": "bob"}
],
"user_id": "<token_user_id>"
}'
```
Generate short memorable channel IDs (e.g. `general`, `random`, `team-alpha`) and use a small set of random usernames (e.g. `alice`, `bob`, `carol`, `dave`). The token user must be added to every channel — the channel list filter is `Filter.in_('members', [tokenUserId])` and will return nothing if the user is absent.
**Sub-step C4 — seed messages.** Attribute the sender server-side with `message.user_id`. Send a couple per channel so the list shows previews and the channel screen isn't empty; set an explicit `message.id` on anything you want to reply to, so a threaded reply can reference it without parsing a response.
```bash
getstream api SendMessage --type messaging --id <channel-id> \
--request '{"message":{"id":"<channel-id>-m1","user_id":"alice","text":"Morning - anyone up for a sync today?"}}'
# Threaded reply into that message
getstream api SendMessage --type messaging --id <channel-id> \
--request '{"message":{"user_id":"bob","text":"I will bring the numbers","parent_id":"<channel-id>-m1"}}'
```
After seeding, print a brief summary:
> Created channels: `general` (token_user, alice, bob), `random` (token_user, carol, dave), `team-alpha` (token_user, alice, carol)
#### Step D - Proceed automatically
After all CLI steps succeed, continue without pausing - no additional prompt needed: tracks A/B/D move straight to **Project signals** and then into `builder.md`; Track M returns to [`sendbird-migration.md`](sendbird-migration.md) (phase M2). If any CLI step fails, explain the error briefly and ask the user to paste the missing value manually before continuing.
### What NOT to do
- Never put the API **secret** in app code - the CLI uses it server-side only.
- Never invent or fabricate credentials.
- Never ask "should I continue?" between Step A, B, C, and D - execute the whole sequence once the user's upfront answers are in.
### Permissions awareness (Chat - surface proactively)
Stream Chat checks permissions **per role, per scope on every client-side call** — but **server-side calls (the CLI and your backend, using the API _secret_) bypass all checks**. That asymmetry is the #1 source of "it worked when you seeded it, but the app 403s": seeding channels via the CLI succeeds regardless of grants, then the same query/join from the app hits the connected user's role and fails.
**When the app you're about to build does anything beyond chatting inside channels the user is already a member of, tell the integrator about the relevant grants _before_ writing the feature — don't wait for a runtime 403.** Map the scenario to the grant:
| App behaviour you're building | Grant the connecting role needs on the channel type | Default `messaging` for `user`/`guest` |
| ------------------------------------------------------------------------------------------------ | -------------------------------------------------------------- | -------------------------------------- |
| **Discover / browse groups** the user didn't create (`queryChannels` without a `members` filter) | `Read Channel` (`ReadChannel`) | often **off** |
| **Join an existing group** (`channel.addMembers([myId])`) | `Add Own Channel Membership` (`AddOwnChannelMembership`) | often **off** |
| **Leave a group** (`channel.removeMembers([myId])`) | `Remove Own Channel Membership` (`RemoveOwnChannelMembership`) | varies |
| **Create a group** | `Create Channel` (`CreateChannel`) | usually on |
`guest` users (name-only / no-backend sign-in) are stricter than `user` — if the app uses guest auth, the same grants must be added to the **`guest`** role too. Point the integrator to **Dashboard → Chat → Roles & Permissions** (permissions v2) for the role + `messaging` type, or `UpdateChannelType` via API/CLI. Full detail and the exact error string: [`references/CHAT-FLUTTER.md`](references/CHAT-FLUTTER.md) → Channel permissions & roles (custom-UI builds: [`references/CHAT-CORE.md`](references/CHAT-CORE.md)).
This is a **prompt, not a blocker** — build the feature as requested, but call out the prerequisite in the same turn so discover/join/create don't silently fail on first run.
---
## Project signals (tracks A/B/D/M - once per session; Track C on demand only)
Read-only local probe. Use it to detect whether the user is in a Flutter project or an empty directory.
```bash
bash -c 'echo "=== FLUTTER ==="; find . -maxdepth 2 -name "pubspec.yaml" -print 2>/dev/null; echo "=== STREAM ==="; grep -rE "stream_chat|stream_video|stream_feed" . --include="pubspec.yaml" -l 2>/dev/null; echo "=== EMPTY ==="; test -z "$(ls -A 2>/dev/null)" && echo "EMPTY_CWD" || echo "NON_EMPTY"'
```
Hold the result in conversation context. Don't re-run it unless the user changes directory or the project shape clearly changed.
Use the result to produce a **one-line status**, for example:
- `Flutter app detected - stream_chat_flutter already in pubspec.yaml`
- `Flutter app detected - stream_video_flutter already in pubspec.yaml`
- `Flutter app detected - stream_feed already in pubspec.yaml`...
© GetStream · Apache-2.0 · GetStream/agent-skills