跳到正文
Unite.AI·· 2 小時前AI 評分65

Slack Code API 正式向所有 Slack 方案開放

Slack Code API Reaches General Availability on All Slack Plans

AI 導讀

Slack 於 2026 年 10 月 28 日向所有 Slack 方案開放 Slack Code API,開發者可藉此讓智能體建立及管理任務專屬的程式碼頻道。Slack 表示,逾 150 名客戶及合作夥伴曾要求開放相關存取。智能體沿用頻道的權限及管理員控制,頻道封存後仍可檢視和搜尋其歷史紀錄。

正文

Slack made the Slack Code API generally available on all Slack plans on October 28, 2026, opening the code-channel system it launched in August 2026 to any developer’s agent. In a post on Slack’s developer blog, the company said more than 150 customers and partners had asked for access to build code-channel experiences for their own agents, a capability previously limited to a handful of launch partners.

Slack introduced Slack Code in August alongside partners including Anthropic, Cognition, GitHub, and Vercel, with the launch announcement also naming OpenAI and listing ChatGPT as available soon. A code channel is a dedicated space an agent opens for a single task, with the right people, context, and plan already in it; the channel archives itself when the work is done but remains searchable. Slack reported at launch that over 70% of code channels spin up and close within a single day, from idea to merged pull request.

Bring Your Own Agent

With the API generally available, any developer can have an agent start a code channel, add the relevant people and files, post updates as it works, and archive the channel when the work is done. Slack said developers get the same primitives its launch partners use, whether they are building a homegrown agent for an enterprise or a new agent for the Slack Marketplace.

Slack said bringing coding agents into Slack previously meant routing agent output through standard threads and direct messages, which it described as cramped once an agent needs to attach a large log file, propose a 50-line code diff, or run a multi-step execution plan. Enterprises with strict security requirements, Slack said, could not use off-the-shelf third-party agents at all.

The company describes the API as unlocking three capabilities: agents that work in the open by creating their own code channel when a request deserves its own space; agents a whole team can steer, using bookmarks and a full execution history; and agents enterprises can trust, with message compliance built in alongside the workspace controls and data boundaries admins already use. The APIs can show code diffs, HTML previews, and thinking steps while Slack handles the rest of the interface. Slack said it remains an agent-neutral hub, so teams can swap agents or models later while continuing to work in the same place.

Slack reported that GitHub Copilot opens a dedicated code channel on 30% of its agent sessions. Tim Hunter, a staff engineer at GitHub, said GitHub is building the Copilot interactions users are already familiar with into the same space where they do the rest of their work. Mael Kerichard, a software engineer at Mistral, said the API felt designed with AI agents in mind, pointing to its flexibility, simple setup, and depth of functionality.

Governance is built in, according to Slack: an agent inherits the permissions and admin controls of the channel it works in, with no separate access model for developers to build and nothing new for IT to configure. A person can approve each high-stakes change, and the record of what the agent did remains searchable after the channel is archived.

Slack illustrates the model with a production alert at 2:00 a.m., in which a custom observability agent opens a code channel, pulls in the relevant logs, and begins work on a fix while on-call engineers read along and redirect it mid-task. The company said the same pattern applies to bug triage and to DevOps tasks such as configuration changes or rollbacks.

How the API Works

Slack’s developer documentation states that apps control a code channel’s title, status, context bar, and view tabs through two API method families. The agents.sessions.* methods are unified across code channels and session threads, with a thread timestamp parameter selecting the session type, and they require the chat:write scope. The agents.conversations.* methods manage code-channel-specific tasks and require the code_channels:manage scope.

Documented agents.conversations.* methods include create, archive (which can optionally record a summary), setProperties, setCommands, setView (supporting html, diff, block_kit, or canvas views), listViews, removeView, getCanvas, and setCanvasContent.

Apps enable code channels by setting features.code_channels.enabled to true in the app manifest, with an optional slash_command_url. When a user mentions the app, it receives an app_mention event and decides whether to create a dedicated code channel or reply in place. The create call takes the origin channel and message timestamp; Slack then creates the channel, matches its privacy setting, and invites the bot automatically.

Under the documented authorization model, the caller must be the agent bot assigned to the channel or a member of the channel, with calls otherwise returning not_agent_channel or not_authorized errors. Code channels are designed to be semi-ephemeral (one channel per task, archived when the session finishes), and once archived a channel and its history remain viewable, searchable, and able to be reopened.

Upcoming Tools and How to Start

Slack said improvements to the pull-request experience are available soon: GitHub users will be able to request reviews in Slack, comments will sync bi-directionally between Slack and GitHub, and a merge prompt will appear in the code channel once a pull request is approved and tests pass. An enhanced Agents & Tools tab is also listed as available soon, offering a bird’s-eye view of agent conversations and code channels with status, channel membership, and recent context, plus the ability to browse public code channels.

Slack outlined three ways to start: read the documentation and add the API to an agent manifest, submit an agent to the Slack Marketplace, or try existing Slack Code agents in the Marketplace, which include Claude, Devin, GitHub Copilot, and Vercel.

來源:Unite.AI · unite.ai