Files
vercel__chat/packages/adapter-github
github-actions[bot] 7a1798bdfd chore(release): version packages (#841)
This PR was opened by the [Changesets
release](https://github.com/changesets/action) GitHub action. When
you're ready to do a release, you can merge this and the packages will
be published to npm automatically. If you're not ready to do a release
yet, that's fine, whenever you add more changesets to main, this PR will
be updated.


# Releases
## @chat-adapter/shared@4.39.0

### Minor Changes

- 153bd96: guard Messenger attachment downloads against SSRF and
oversized responses

`downloadAttachment` in `@chat-adapter/shared` accepts an optional
`hosts` allowlist that restricts downloads, including redirect targets,
to the given hosts and their subdomains. The Messenger adapter uses it
to download attachment media only from Meta's `fbsbx.com` and
`fbcdn.net` hosts, with the shared SSRF guard, 25 MB size cap, and 30
second timeout. External fallback and link-share URLs are rejected
before any network request.

- bb92688: secure anonymous attachment downloads against SSRF and
oversized responses

`@chat-adapter/shared` gains `downloadAttachment`, a guarded downloader
that refuses private and internal addresses (as URL literals, through
DNS resolution, and after redirects), decodes compressed responses, caps
the body size at 25 MB, and bounds the whole download with a 30 second
timeout. All of these are configurable, including the transport for
proxied deployments.

The Teams adapter uses it for anonymous attachment downloads. HTTPS
attachments on any public host keep working, plain-HTTP URLs are
refused, and the Bot Framework Emulator's loopback connector now uses
bot authentication so local development keeps working.

- e71bfea: Add `normalizeCodeFences` to `@chat-adapter/shared`: a
code-fence normalizer for platforms whose triple-backtick fences treat
the text after the opening fence as code rather than a CommonMark info
string. Only paired fences become code blocks (unpaired fences, fences
inside inline code, and fences on quoted lines stay literal text), text
following a closing fence cannot be promoted to a block construct, and
per-segment callbacks keep text-level rewrites out of code content.

The WhatsApp adapter now uses it when parsing incoming messages: the
first line of a code block is preserved in message text and formatted
content, and bold/strikethrough rewriting no longer corrupts fenced
code.

- b6fa24c: guard attachment downloads across the remaining adapters

Slack, Discord, and WhatsApp attachment downloads now go through the
shared guarded downloader: private and internal addresses are refused
(as URL literals, through DNS resolution, and after redirects),
responses are capped at 25 MB, and downloads time out after 30 seconds.
Slack sends the bot token only on hops to trusted Slack origins, and
WhatsApp keeps its access token on Meta's media hosts and the configured
Graph origin. Telegram enforces the same size cap and timeout with the
Web Fetch API so downloads keep working in runtimes like Cloudflare
Workers.

`downloadAttachment` in `@chat-adapter/shared` now resolves `headers`
per hop (pass a function to control what each redirect target receives),
forwards the resolved headers to custom transports, and accepts an
`onResponse` hook to reject unexpected final responses before the body
is read.

### Patch Changes

- Updated dependencies [2ce2be0]
- Updated dependencies [16ea171]
- Updated dependencies [169788b]
- Updated dependencies [eddcd7e]
- Updated dependencies [5b538f6]
- Updated dependencies [929878b]
- Updated dependencies [500b7e6]
  - chat@4.39.0
## @chat-adapter/slack@4.39.0

### Minor Changes

- 2ce2be0: Add Slack Agent Sessions lifecycle support, native stop
cancellation through `thread.signal`, automatic session titles, and
session stop/title-change events while preserving the legacy
`assistant_view` compatibility path.

### Patch Changes

- 864d922: Keep alert attachment content on normalized Slack messages.
Attachments that aren't link unfurls now contribute their pretext, title
(linked to `title_link` when present, with the URL also surfaced in
`message.links`), text, and fields instead of being dropped; `fallback`
fills in when nothing else on the attachment carries content. Matching
how Slack renders these fields, they are treated as plain text unless
listed in the attachment's `mrkdwn_in` array, so literal `*`, `_`, and
backticks in alert text survive normalization. Tables inside attachment
blocks now stay adjacent to their attachment's text.

Because attachment content is part of `message.text`, mention detection
and `onMessage` pattern handlers see it too: an attachment that quotes
the bot's mention routes to `onNewMention`, and patterns match alert
text. Handlers that should ignore other integrations' alerts can check
`message.author.isBot`.

- 7c26965: prevent attachment downloads from sending credentials to
untrusted hosts
- e71bfea: Preserve the first line of incoming Slack code blocks when
extracting message text and formatted content.

Only paired triple-backtick fences become code blocks: an unpaired
fence, a fence inside inline code or a `<…>` token, and a fence on a
quoted line all stay literal text, matching how Slack renders them. Bold
and strikethrough rewriting no longer touches fenced code content, and
text following a closing fence can no longer turn into a blockquote,
heading, or list.

- b6fa24c: guard attachment downloads across the remaining adapters

Slack, Discord, and WhatsApp attachment downloads now go through the
shared guarded downloader: private and internal addresses are refused
(as URL literals, through DNS resolution, and after redirects),
responses are capped at 25 MB, and downloads time out after 30 seconds.
Slack sends the bot token only on hops to trusted Slack origins, and
WhatsApp keeps its access token on Meta's media hosts and the configured
Graph origin. Telegram enforces the same size cap and timeout with the
Web Fetch API so downloads keep working in runtimes like Cloudflare
Workers.

`downloadAttachment` in `@chat-adapter/shared` now resolves `headers`
per hop (pass a function to control what each redirect target receives),
forwards the resolved headers to custom transports, and accepts an
`onResponse` hook to reject unexpected final responses before the body
is read.

- Updated dependencies [2ce2be0]
- Updated dependencies [153bd96]
- Updated dependencies [16ea171]
- Updated dependencies [169788b]
- Updated dependencies [eddcd7e]
- Updated dependencies [bb92688]
- Updated dependencies [5b538f6]
- Updated dependencies [e71bfea]
- Updated dependencies [929878b]
- Updated dependencies [500b7e6]
- Updated dependencies [b6fa24c]
  - chat@4.39.0
  - @chat-adapter/shared@4.39.0
## @chat-adapter/telegram@4.39.0

### Minor Changes

- c4a359e: require webhook verification by default with an explicit
unverified opt-in
- 26a06ca: Add `mentionOnReply`: when enabled, a reply to one of the
bot's own messages reports `isMention`, so a bot in a group keeps the
conversation going without the handle being repeated. Off by default, so
existing mention-only bots are unaffected, and readable from
`TELEGRAM_MENTION_ON_REPLY`. Implicit forum-topic replies and the bot's
own echoed messages never count, and polling mode now retries the
startup `getMe` lazily so a transient outage cannot leave mention
detection disabled.
- d5ebec1: Implement `reply` in the Telegram adapter so `Thread.reply()`
threads the answer to its target instead of throwing
`NotImplementedError`. The reference travels as Bot API
`reply_parameters` and covers text, rich messages, documents,
attachments and media groups; `allow_sending_without_reply` keeps
delivery working when the target has been deleted. Malformed reply
target ids are rejected before anything is sent, and a rich-message
gateway that rejects `reply_parameters` falls back to a regular threaded
send.
- a18e792: Describe the message kinds Telegram sends with no text and no
file. A shared location, venue, contact, poll, dice, game, invoice or
story used to arrive as an empty message: the payload carried the
content, but a handler reading `text` saw nothing. Each now gets a short
literal description (`📍 55.75, 37.61`, `👤 Ada Lovelace +1555…`, `📊 Lunch
or dinner?`), and the structured payload stays on the raw message for
anyone who needs the numbers.
- a0ba986: Parse stickers and animations. A sticker used to arrive as an
empty message, since it carries no text, and an animation (Telegram GIF)
was dropped entirely. A sticker now reports the emoji it stands for as
its text (falling back to the sticker set name, then to "sticker") plus
an attachment matching its real format: an image for a still WebP
sticker, a video for a WebM one, a file for a Lottie (TGS) one. An
animation arrives as a single video attachment; the redundant `document`
field Telegram sets alongside it for backward compatibility is no longer
reported as a second attachment.

### Patch Changes

- eddcd7e: Return Telegram file downloads as portable ArrayBuffer data
while preserving Buffer support in the shared attachment contract.
- b6fa24c: guard attachment downloads across the remaining adapters

Slack, Discord, and WhatsApp attachment downloads now go through the
shared guarded downloader: private and internal addresses are refused
(as URL literals, through DNS resolution, and after redirects),
responses are capped at 25 MB, and downloads time out after 30 seconds.
Slack sends the bot token only on hops to trusted Slack origins, and
WhatsApp keeps its access token on Meta's media hosts and the configured
Graph origin. Telegram enforces the same size cap and timeout with the
Web Fetch API so downloads keep working in runtimes like Cloudflare
Workers.

`downloadAttachment` in `@chat-adapter/shared` now resolves `headers`
per hop (pass a function to control what each redirect target receives),
forwards the resolved headers to custom transports, and accepts an
`onResponse` hook to reject unexpected final responses before the body
is read.

- Updated dependencies [2ce2be0]
- Updated dependencies [153bd96]
- Updated dependencies [16ea171]
- Updated dependencies [169788b]
- Updated dependencies [eddcd7e]
- Updated dependencies [bb92688]
- Updated dependencies [5b538f6]
- Updated dependencies [e71bfea]
- Updated dependencies [929878b]
- Updated dependencies [500b7e6]
- Updated dependencies [b6fa24c]
  - chat@4.39.0
  - @chat-adapter/shared@4.39.0
## @chat-adapter/twilio@4.39.0

### Minor Changes

- 75cadbf: feat(twilio): add RCS support with rich cards, button
actions, and location sharing

Extends the Twilio adapter with full RCS support: inbound button tap
routing via `processAction`, location share parsing, Content API
integration for rich outbound cards with SMS fallback, and channel
metadata detection. Cards sent to RCS-capable senders (Messaging Service
or `rcs:` address) are automatically rendered as Twilio Content
templates with embedded SMS fallback variants.

Existing deployments keep their thread ids: plain SMS threads stay keyed
by phone number even when the number belongs to a Messaging Service, and
`openDM` still prefers `phoneNumber` over `messagingServiceSid`. Only
taps of buttons rendered by Chat SDK become actions; foreign button taps
that carry a body keep arriving as messages.

### Patch Changes

- 28bc776: isolate Twilio message processing locks by conversation
- Updated dependencies [2ce2be0]
- Updated dependencies [153bd96]
- Updated dependencies [16ea171]
- Updated dependencies [169788b]
- Updated dependencies [eddcd7e]
- Updated dependencies [bb92688]
- Updated dependencies [5b538f6]
- Updated dependencies [e71bfea]
- Updated dependencies [929878b]
- Updated dependencies [500b7e6]
- Updated dependencies [b6fa24c]
  - chat@4.39.0
  - @chat-adapter/shared@4.39.0
## chat@4.39.0

### Minor Changes

- 2ce2be0: Add Slack Agent Sessions lifecycle support, native stop
cancellation through `thread.signal`, automatic session titles, and
session stop/title-change events while preserving the legacy
`assistant_view` compatibility path.
- 169788b: Introduce the unified History API (`bot.history`) with user,
thread, and channel scopes.

`bot.history.user` replaces `bot.transcripts` for cross-platform
per-user message persistence. The API surface is identical — migrate by
changing the `transcripts` config key to `history.user` and updating
call sites from `bot.transcripts.*` to `bot.history.user.*`.
`bot.transcripts` remains available as a deprecated alias.

`bot.history.thread` and `bot.history.channel` expose promise-based
helpers for per-thread and per-channel message access, aligned with the
existing `thread.messages` and `channel.threads()` iterators. Reads
delegate to the adapter; adapters that persist history in the SDK-side
store (`persistThreadHistory: true`) are served from that cache. An
unregistered adapter prefix or an unsupported capability throws instead
of returning an empty result.

The new `toPromptEntries` helper converts `history.user.list()` entries
into `{ role, content }` messages ready for LLM prompts.

The `TranscriptEntry` type is deprecated in favour of `HistoryEntry`.
Both are exported from `chat`.

- 5b538f6: Keep thread locks alive while message handlers run so queue,
burst, and debounce strategies remain serialized beyond the lock TTL.
Renewal is capped by the new `concurrency.maxLockLifetimeMs` option
(default 10 minutes) so a hung handler cannot block a thread forever.
When the heartbeat detects that lock ownership was lost, the queue drain
and debounce loops stop instead of competing with the new lock holder,
and the debounce loop now keeps draining messages that arrive while a
handler is running instead of stranding them until the next webhook.

### Patch Changes

- 16ea171: preserve adapter-returned thread ids when editing channel
messages
- eddcd7e: Return Telegram file downloads as portable ArrayBuffer data
while preserving Buffer support in the shared attachment contract.
- 929878b: Allow JSX link buttons to include an explicit action ID.
- 500b7e6: enforce the conversation scope on write tools and stop
trusting client-supplied message history in the web adapter

`createChatTools` now runs the same scope guard on write tools that read
tools already used, so a thread or channel id the model supplies that
resolves outside the scoped conversation is rejected before the write
executes. `sendDirectMessage` targets a user id rather than a
conversation and stays gated by approval alone.

The web adapter no longer treats the request body's `messages` array as
a source of conversation state. Only the latest user message is
consumed, and tool parts are stripped from it so a browser cannot inject
forged tool-call or approval state. Text, file, and custom data parts
pass through unchanged; a message left with no parts after stripping is
rejected with HTTP 400. Prior turns come from the state adapter when
`persistMessageHistory` is enabled.
## @chat-adapter/discord@4.39.0

### Patch Changes

- c4f709f: fix edits, deletes, and reactions on Discord thread starter
messages

Operations on a thread's starter message now try the thread first and
fall back to the parent channel when Discord reports the message as
unknown. Threads on a text channel keep their starter message in the
parent channel, so those operations used to fail; forum and media posts
keep theirs in the thread and are unaffected.

Note that deleting a text-channel thread's starter message now deletes
the message, which Discord cascades into deleting the thread.

- b6fa24c: guard attachment downloads across the remaining adapters

Slack, Discord, and WhatsApp attachment downloads now go through the
shared guarded downloader: private and internal addresses are refused
(as URL literals, through DNS resolution, and after redirects),
responses are capped at 25 MB, and downloads time out after 30 seconds.
Slack sends the bot token only on hops to trusted Slack origins, and
WhatsApp keeps its access token on Meta's media hosts and the configured
Graph origin. Telegram enforces the same size cap and timeout with the
Web Fetch API so downloads keep working in runtimes like Cloudflare
Workers.

`downloadAttachment` in `@chat-adapter/shared` now resolves `headers`
per hop (pass a function to control what each redirect target receives),
forwards the resolved headers to custom transports, and accepts an
`onResponse` hook to reject unexpected final responses before the body
is read.

- Updated dependencies [2ce2be0]
- Updated dependencies [153bd96]
- Updated dependencies [16ea171]
- Updated dependencies [169788b]
- Updated dependencies [eddcd7e]
- Updated dependencies [bb92688]
- Updated dependencies [5b538f6]
- Updated dependencies [e71bfea]
- Updated dependencies [929878b]
- Updated dependencies [500b7e6]
- Updated dependencies [b6fa24c]
  - chat@4.39.0
  - @chat-adapter/shared@4.39.0
## @chat-adapter/gchat@4.39.0

### Patch Changes

- Updated dependencies [2ce2be0]
- Updated dependencies [153bd96]
- Updated dependencies [16ea171]
- Updated dependencies [169788b]
- Updated dependencies [eddcd7e]
- Updated dependencies [bb92688]
- Updated dependencies [5b538f6]
- Updated dependencies [e71bfea]
- Updated dependencies [929878b]
- Updated dependencies [500b7e6]
- Updated dependencies [b6fa24c]
  - chat@4.39.0
  - @chat-adapter/shared@4.39.0
## @chat-adapter/github@4.39.0

### Patch Changes

- Updated dependencies [2ce2be0]
- Updated dependencies [153bd96]
- Updated dependencies [16ea171]
- Updated dependencies [169788b]
- Updated dependencies [eddcd7e]
- Updated dependencies [bb92688]
- Updated dependencies [5b538f6]
- Updated dependencies [e71bfea]
- Updated dependencies [929878b]
- Updated dependencies [500b7e6]
- Updated dependencies [b6fa24c]
  - chat@4.39.0
  - @chat-adapter/shared@4.39.0
## @chat-adapter/instagram@4.39.0

### Patch Changes

- Updated dependencies [2ce2be0]
- Updated dependencies [153bd96]
- Updated dependencies [16ea171]
- Updated dependencies [169788b]
- Updated dependencies [eddcd7e]
- Updated dependencies [bb92688]
- Updated dependencies [5b538f6]
- Updated dependencies [e71bfea]
- Updated dependencies [929878b]
- Updated dependencies [500b7e6]
- Updated dependencies [b6fa24c]
  - chat@4.39.0
  - @chat-adapter/shared@4.39.0
## @chat-adapter/linear@4.39.0

### Patch Changes

- Updated dependencies [2ce2be0]
- Updated dependencies [153bd96]
- Updated dependencies [16ea171]
- Updated dependencies [169788b]
- Updated dependencies [eddcd7e]
- Updated dependencies [bb92688]
- Updated dependencies [5b538f6]
- Updated dependencies [e71bfea]
- Updated dependencies [929878b]
- Updated dependencies [500b7e6]
- Updated dependencies [b6fa24c]
  - chat@4.39.0
  - @chat-adapter/shared@4.39.0
## @chat-adapter/messenger@4.39.0

### Patch Changes

- 153bd96: guard Messenger attachment downloads against SSRF and
oversized responses

`downloadAttachment` in `@chat-adapter/shared` accepts an optional
`hosts` allowlist that restricts downloads, including redirect targets,
to the given hosts and their subdomains. The Messenger adapter uses it
to download attachment media only from Meta's `fbsbx.com` and
`fbcdn.net` hosts, with the shared SSRF guard, 25 MB size cap, and 30
second timeout. External fallback and link-share URLs are rejected
before any network request.

- Updated dependencies [2ce2be0]
- Updated dependencies [153bd96]
- Updated dependencies [16ea171]
- Updated dependencies [169788b]
- Updated dependencies [eddcd7e]
- Updated dependencies [bb92688]
- Updated dependencies [5b538f6]
- Updated dependencies [e71bfea]
- Updated dependencies [929878b]
- Updated dependencies [500b7e6]
- Updated dependencies [b6fa24c]
  - chat@4.39.0
  - @chat-adapter/shared@4.39.0
## @chat-adapter/notion@4.39.0

### Patch Changes

- Updated dependencies [2ce2be0]
- Updated dependencies [153bd96]
- Updated dependencies [16ea171]
- Updated dependencies [169788b]
- Updated dependencies [eddcd7e]
- Updated dependencies [bb92688]
- Updated dependencies [5b538f6]
- Updated dependencies [e71bfea]
- Updated dependencies [929878b]
- Updated dependencies [500b7e6]
- Updated dependencies [b6fa24c]
  - chat@4.39.0
  - @chat-adapter/shared@4.39.0
## @chat-adapter/teams@4.39.0

### Patch Changes

- 63997ac: hydrate live Teams sender email without requiring Microsoft
Graph permissions
- bb92688: secure anonymous attachment downloads against SSRF and
oversized responses

`@chat-adapter/shared` gains `downloadAttachment`, a guarded downloader
that refuses private and internal addresses (as URL literals, through
DNS resolution, and after redirects), decodes compressed responses, caps
the body size at 25 MB, and bounds the whole download with a 30 second
timeout. All of these are configurable, including the transport for
proxied deployments.

The Teams adapter uses it for anonymous attachment downloads. HTTPS
attachments on any public host keep working, plain-HTTP URLs are
refused, and the Bot Framework Emulator's loopback connector now uses
bot authentication so local development keeps working.

- Updated dependencies [2ce2be0]
- Updated dependencies [153bd96]
- Updated dependencies [16ea171]
- Updated dependencies [169788b]
- Updated dependencies [eddcd7e]
- Updated dependencies [bb92688]
- Updated dependencies [5b538f6]
- Updated dependencies [e71bfea]
- Updated dependencies [929878b]
- Updated dependencies [500b7e6]
- Updated dependencies [b6fa24c]
  - chat@4.39.0
  - @chat-adapter/shared@4.39.0
## @chat-adapter/web@4.39.0

### Patch Changes

- 500b7e6: enforce the conversation scope on write tools and stop
trusting client-supplied message history in the web adapter

`createChatTools` now runs the same scope guard on write tools that read
tools already used, so a thread or channel id the model supplies that
resolves outside the scoped conversation is rejected before the write
executes. `sendDirectMessage` targets a user id rather than a
conversation and stays gated by approval alone.

The web adapter no longer treats the request body's `messages` array as
a source of conversation state. Only the latest user message is
consumed, and tool parts are stripped from it so a browser cannot inject
forged tool-call or approval state. Text, file, and custom data parts
pass through unchanged; a message left with no parts after stripping is
rejected with HTTP 400. Prior turns come from the state adapter when
`persistMessageHistory` is enabled.

- Updated dependencies [2ce2be0]
- Updated dependencies [153bd96]
- Updated dependencies [16ea171]
- Updated dependencies [169788b]
- Updated dependencies [eddcd7e]
- Updated dependencies [bb92688]
- Updated dependencies [5b538f6]
- Updated dependencies [e71bfea]
- Updated dependencies [929878b]
- Updated dependencies [500b7e6]
- Updated dependencies [b6fa24c]
  - chat@4.39.0
  - @chat-adapter/shared@4.39.0
## @chat-adapter/whatsapp@4.39.0

### Patch Changes

- 3e6e866: support business-scoped user IDs for inbound and outbound
WhatsApp messages
- e71bfea: Add `normalizeCodeFences` to `@chat-adapter/shared`: a
code-fence normalizer for platforms whose triple-backtick fences treat
the text after the opening fence as code rather than a CommonMark info
string. Only paired fences become code blocks (unpaired fences, fences
inside inline code, and fences on quoted lines stay literal text), text
following a closing fence cannot be promoted to a block construct, and
per-segment callbacks keep text-level rewrites out of code content.

The WhatsApp adapter now uses it when parsing incoming messages: the
first line of a code block is preserved in message text and formatted
content, and bold/strikethrough rewriting no longer corrupts fenced
code.

- 7c26965: prevent attachment downloads from sending credentials to
untrusted hosts
- b6fa24c: guard attachment downloads across the remaining adapters

Slack, Discord, and WhatsApp attachment downloads now go through the
shared guarded downloader: private and internal addresses are refused
(as URL literals, through DNS resolution, and after redirects),
responses are capped at 25 MB, and downloads time out after 30 seconds.
Slack sends the bot token only on hops to trusted Slack origins, and
WhatsApp keeps its access token on Meta's media hosts and the configured
Graph origin. Telegram enforces the same size cap and timeout with the
Web Fetch API so downloads keep working in runtimes like Cloudflare
Workers.

`downloadAttachment` in `@chat-adapter/shared` now resolves `headers`
per hop (pass a function to control what each redirect target receives),
forwards the resolved headers to custom transports, and accepts an
`onResponse` hook to reject unexpected final responses before the body
is read.

- Updated dependencies [2ce2be0]
- Updated dependencies [153bd96]
- Updated dependencies [16ea171]
- Updated dependencies [169788b]
- Updated dependencies [eddcd7e]
- Updated dependencies [bb92688]
- Updated dependencies [5b538f6]
- Updated dependencies [e71bfea]
- Updated dependencies [929878b]
- Updated dependencies [500b7e6]
- Updated dependencies [b6fa24c]
  - chat@4.39.0
  - @chat-adapter/shared@4.39.0
## @chat-adapter/x@4.39.0

### Patch Changes

- Updated dependencies [2ce2be0]
- Updated dependencies [153bd96]
- Updated dependencies [16ea171]
- Updated dependencies [169788b]
- Updated dependencies [eddcd7e]
- Updated dependencies [bb92688]
- Updated dependencies [5b538f6]
- Updated dependencies [e71bfea]
- Updated dependencies [929878b]
- Updated dependencies [500b7e6]
- Updated dependencies [b6fa24c]
  - chat@4.39.0
  - @chat-adapter/shared@4.39.0
## @chat-adapter/state-ioredis@4.39.0

### Patch Changes

- Updated dependencies [2ce2be0]
- Updated dependencies [16ea171]
- Updated dependencies [169788b]
- Updated dependencies [eddcd7e]
- Updated dependencies [5b538f6]
- Updated dependencies [929878b]
- Updated dependencies [500b7e6]
  - chat@4.39.0
## @chat-adapter/state-memory@4.39.0

### Patch Changes

- Updated dependencies [2ce2be0]
- Updated dependencies [16ea171]
- Updated dependencies [169788b]
- Updated dependencies [eddcd7e]
- Updated dependencies [5b538f6]
- Updated dependencies [929878b]
- Updated dependencies [500b7e6]
  - chat@4.39.0
## @chat-adapter/state-pg@4.39.0

### Patch Changes

- Updated dependencies [2ce2be0]
- Updated dependencies [16ea171]
- Updated dependencies [169788b]
- Updated dependencies [eddcd7e]
- Updated dependencies [5b538f6]
- Updated dependencies [929878b]
- Updated dependencies [500b7e6]
  - chat@4.39.0
## @chat-adapter/state-redis@4.39.0

### Patch Changes

- Updated dependencies [2ce2be0]
- Updated dependencies [16ea171]
- Updated dependencies [169788b]
- Updated dependencies [eddcd7e]
- Updated dependencies [5b538f6]
- Updated dependencies [929878b]
- Updated dependencies [500b7e6]
  - chat@4.39.0
## @chat-adapter/tests@4.39.0

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
2026-08-28 20:37:34 +10:00
..

GitHub adapter for Chat SDK

@chat-adapter/github

npm package: @chat-adapter/github

Agent Stack MIT License

GitHub adapter for Chat SDK. Respond to @mentions in PR and issue comment threads.

The GitHub adapter treats issue and pull request comments as messages, and issues/PRs as threads.

Documentation: chat-sdk.dev/adapters/official/github · Guides: vercel.com/kb/chat-sdk

Installation

pnpm add @chat-adapter/github

Scaffold with the CLI

To scaffold a new GitHub bot with this adapter preselected:

npx create-chat-sdk@latest my-bot --adapter github memory

Visit the adapters directory to see other available official and vendor-official adapters.

Usage

The adapter auto-detects credentials from GITHUB_TOKEN (or GITHUB_APP_ID/GITHUB_PRIVATE_KEY), GITHUB_WEBHOOK_SECRET, and GITHUB_BOT_USERNAME environment variables:

import { Chat } from "chat";
import { createGitHubAdapter } from "@chat-adapter/github";

const bot = new Chat({
  userName: "my-bot",
  adapters: {
    github: createGitHubAdapter(),
  },
});

bot.onNewMention(async (thread, message) => {
  await thread.post("Hello from GitHub!");
});

Authentication

Option A: Personal Access Token

Best for personal projects, testing, or single-repo bots.

  1. Go to Settings > Developer settings > Personal access tokens
  2. Create a new token with repo scope
  3. Set GITHUB_TOKEN environment variable
createGitHubAdapter({
  token: process.env.GITHUB_TOKEN!,
});

Better rate limits, security, and supports multiple installations.

1. Create the app:

  1. Go to Settings > Developer settings > GitHub Apps > New GitHub App
  2. Set Webhook URL to https://your-domain.com/api/webhooks/github
  3. Generate and set a Webhook secret
  4. Set permissions:
    • Repository > Issues: Read & write
    • Repository > Pull requests: Read & write
    • Repository > Metadata: Read-only
  5. Subscribe to events: Issue comment, Pull request review comment
  6. Click Create GitHub App
  7. Note the App ID and click Generate a private key

2. Install the app:

  1. Go to your app's settings then Install App
  2. Click Install and choose repositories
  3. Note the Installation ID from the URL:
    https://github.com/settings/installations/12345678
                                               ^^^^^^^^
    

Single-tenant:

createGitHubAdapter({
  appId: process.env.GITHUB_APP_ID!,
  privateKey: process.env.GITHUB_PRIVATE_KEY!,
  installationId: parseInt(process.env.GITHUB_INSTALLATION_ID!),
});

Multi-tenant (omit installationId):

createGitHubAdapter({
  appId: process.env.GITHUB_APP_ID!,
  privateKey: process.env.GITHUB_PRIVATE_KEY!,
});

The adapter automatically extracts installation IDs from webhooks and caches API clients per-installation.

Option C: Vercel Connect

Use Vercel Connect to source installation access tokens at runtime instead of storing a GitHub App private key. The installationToken resolver supplies the token the adapter sends directly (skipping the App JWT exchange), and webhookVerifier verifies Connect trigger-forwarded webhooks via a Vercel OIDC token instead of a webhook secret.

The simplest path is the connectGitHubAdapter() helper from @vercel/connect/chat:

import { createGitHubAdapter } from "@chat-adapter/github";
import { connectGitHubAdapter } from "@vercel/connect/chat";

createGitHubAdapter({
  ...connectGitHubAdapter("github/acme-github"),
  userName: "my-bot[bot]",
});

Or wire the fields yourself:

import { getToken } from "@vercel/connect";

createGitHubAdapter({
  installationToken: () =>
    getToken("github/acme-github", { subject: { type: "app" } }),
  webhookVerifier: myConnectWebhookVerifier,
  userName: "my-bot[bot]",
});

installationToken accepts a string or () => string | Promise<string> resolver invoked per API call, so it composes with Connect's short-lived tokens. When webhookVerifier is set it takes precedence over webhookSecret and GITHUB_WEBHOOK_SECRET.

Freshness: OIDC verification replaces GitHub's signature check, so request freshness relies on the short-lived OIDC token's expiry rather than a signed timestamp, and there is no built-in nonce/delivery-id de-duplication. Keep your webhook handlers idempotent (GitHub may also redeliver events).

Set botUserId for self-message detection. In Connect mode the adapter only holds an installation token, so it can't auto-detect its own bot user id (the /app lookup needs the App's JWT). Without it, the adapter can't tell its own comments apart from users' and will reply to itself in a loop. The adapter learns the id from the first comment it posts, but that lives in memory — on serverless (where each webhook may hit a fresh instance) that isn't enough. Pass botUserId (the numeric id of your …[bot] user) so every instance knows it up front:

createGitHubAdapter({
  ...connectGitHubAdapter("github/acme-github"),
  botUserId: 12345678, // id of your-app[bot]
});

Or set the GITHUB_BOT_USER_ID environment variable, which the adapter auto-detects. Find the id (no auth needed) with curl -s 'https://api.github.com/users/your-app%5Bbot%5D'.

Installation lookup

You can resolve the GitHub App installation ID associated with a Thread or Message:

import { Chat } from "chat";
import { createGitHubAdapter } from "@chat-adapter/github";

const github = createGitHubAdapter({
  appId: process.env.GITHUB_APP_ID!,
  privateKey: process.env.GITHUB_PRIVATE_KEY!,
  webhookSecret: process.env.GITHUB_WEBHOOK_SECRET!,
});

const bot = new Chat({
  adapters: { github },
});

bot.onNewMention(async (thread, message) => {
  const installationIdFromThread = await github.getInstallationId(thread);
  const installationIdFromMessage = await github.getInstallationId(message.threadId);

  await thread.post(
    `Thread install: ${installationIdFromThread}, message install: ${installationIdFromMessage}`
  );
});
  • Single-tenant GitHub App mode returns the fixed configured installation ID.
  • PAT mode returns undefined.
  • Multi-tenant mode only succeeds after the adapter has received a webhook for that repository and cached the installation mapping. Use a persistent state adapter so the mapping survives restarts.

Direct API client

For anything beyond the unified SDK, access the underlying Octokit instance via .octokit:

const github = bot.getAdapter("github").octokit;

const { data: pulls } = await github.rest.pulls.list({
  owner: "vercel",
  repo: "chat",
  state: "open",
});

PAT and single-tenant GitHub App modes (with a fixed installationId) return the same client anywhere. Multi-tenant mode requires webhook handler context to resolve the right installation — calling .octokit outside a handler throws.

The previous .client getter still works as a deprecated alias for .octokit.

Webhook setup

For repository or organization webhooks:

  1. Go to repository/org Settings then Webhooks then Add webhook
  2. Set Payload URL to https://your-domain.com/api/webhooks/github
  3. Set Content type to application/json (required — the default application/x-www-form-urlencoded does not work)
  4. Set Secret to match your webhookSecret
  5. Select events: Issue comments, Pull request review comments

Warning: GitHub App webhooks are configured during app creation. Make sure to select application/json as the content type.

Thread model

GitHub has three types of comment threads:

Type Context Thread ID format
PR-level PR Conversation tab github:{owner}/{repo}:{prNumber}
Review comments PR Files Changed tab github:{owner}/{repo}:{prNumber}:rc:{commentId}
Issue comments Issue thread github:{owner}/{repo}:issue:{issueNumber}

Reactions

Supports GitHub's reaction emoji:

SDK emoji GitHub reaction
thumbs_up +1
thumbs_down -1
laugh laugh
confused confused
heart heart
hooray hooray
rocket rocket
eyes eyes

Configuration

All options are auto-detected from environment variables when not provided.

Option Required Description
token No* Personal Access Token. Auto-detected from GITHUB_TOKEN
appId No* GitHub App ID. Auto-detected from GITHUB_APP_ID
privateKey No GitHub App private key (PEM). Auto-detected from GITHUB_PRIVATE_KEY
installationId No Installation ID (omit for multi-tenant). Auto-detected from GITHUB_INSTALLATION_ID
installationToken No* Vercel Connect mode: installation access token, or a () => string | Promise<string> resolver invoked per API call. Skips the App JWT exchange.
webhookSecret No** Webhook secret. Auto-detected from GITHUB_WEBHOOK_SECRET
webhookVerifier No** Custom verifier (request, body) => unknown | Promise<unknown> used in place of webhookSecret. Takes precedence over webhookSecret/GITHUB_WEBHOOK_SECRET. Required in Connect mode
userName No Bot username for @mention detection. Auto-detected from GITHUB_BOT_USERNAME (default: "github-bot")
botUserId No Bot's numeric user ID (auto-detected if not provided)
apiUrl No Override the GitHub API base URL (e.g. for GitHub Enterprise Server). Auto-detected from GITHUB_API_URL
logger No Logger instance (defaults to ConsoleLogger("info"))

*One of token/GITHUB_TOKEN, appId+privateKey/GITHUB_APP_ID+GITHUB_PRIVATE_KEY, or installationToken (Vercel Connect) is required.

**Either webhookSecret (via config or GITHUB_WEBHOOK_SECRET) or a webhookVerifier is required. When webhookVerifier is set it takes precedence and the secret is ignored.

Environment variables

# Personal Access Token auth
GITHUB_TOKEN=ghp_xxxxxxxxxxxx

# OR GitHub App auth
GITHUB_APP_ID=123456
GITHUB_PRIVATE_KEY="-----BEGIN RSA PRIVATE KEY-----..."
GITHUB_INSTALLATION_ID=12345678  # Optional for multi-tenant

# Required
GITHUB_WEBHOOK_SECRET=your-webhook-secret

# Optional: GitHub Enterprise Server
GITHUB_API_URL=https://github.example.com/api/v3

Features

Messaging

Feature Supported
Post message Yes
Edit message Yes
Delete message Yes
File uploads No
Streaming Buffered (accumulates then sends)

Rich content

Feature Supported
Card format GFM Markdown
Buttons No
Link buttons No
Select menus No
Tables GFM
Fields Yes
Images in cards Yes
Modals No

Conversations

Feature Supported
Slash commands No
Mentions Yes
Add reactions Yes
Remove reactions Partial
Typing indicator No
DMs No
Ephemeral messages No

Message history

Feature Supported
Fetch messages Yes
Fetch single message No
Fetch thread info Yes
Fetch channel messages Yes
List threads Yes
Fetch channel info Yes
Post channel message No

Platform-specific

Feature Supported
Multi-tenant Yes (GitHub App)

Limitations

  • No typing indicators — GitHub doesn't support typing indicators
  • No streaming — Messages posted in full (editing supported for updates)
  • No DMs — GitHub doesn't have direct messages
  • No modals — GitHub doesn't support interactive modals
  • Action buttons — Rendered as text; use link buttons for clickable actions

Troubleshooting

"Invalid signature" error

  • Verify GITHUB_WEBHOOK_SECRET matches your webhook configuration
  • Ensure the request body isn't modified before verification

"Invalid JSON" error

  • Change webhook Content type to application/json

Bot not responding to mentions

  • Verify webhook events are configured (issue_comment, pull_request_review_comment)
  • Check the webhook URL is correct and accessible
  • Ensure the userName config matches your bot's GitHub username

"Installation ID required" error

  • This occurs when making API calls outside webhook context in multi-tenant mode
  • Use a persistent state adapter (Redis) to store installation mappings
  • The first interaction must come from a webhook to establish the mapping

Rate limiting

  • PATs have lower rate limits than GitHub Apps
  • Consider switching to a GitHub App for production use

Resources

  • Ship a GitHub code review bot with Hono and Redis — Walks through building a GitHub bot that reviews pull requests on demand. When a user @mentions the bot on a PR, Chat SDK picks up the mention, spins up a Vercel Sandbox with the repo cloned, and uses AI SDK to analyze the diff.

See all guides and templates at chat-sdk.dev/resources.

AI Coding Agents

If you use an AI coding agent such as OpenAI Codex, Claude Code, or Cursor, install the Chat SDK skill so it knows the SDK APIs, adapter patterns, and project conventions before writing code.

npx skills add vercel/chat

The skill references bundled documentation in node_modules/chat/docs, plus adapter guides and starter templates in the published package.

You can also install the Vercel Plugin for a broader agent toolkit — it includes the Chat SDK skill alongside specialist agents, agent slash commands, and more:

npx plugins add vercel/vercel-plugin

The plugin is optional; the skill alone is enough to build with Chat SDK.

For agent-readable documentation, see chat-sdk.dev/llms.txt (page index) or chat-sdk.dev/llms-full.txt (full text).

License

MIT