AI 程式開發平台
Slack Code Brings Coding Agents Into Temporary Collaboration Channels, but Initial Integrations and Permission Boundaries Remain Limited
Slack Code centralizes agent conversations, plans, code diffs, inline comments, and live HTML previews in project-specific channels. It inherits Slack’s channel governance and auditing capabilities, but repository permissions and real-world deployment side effects remain controlled by each agent integration.

Slack has officially launched Slack Code, transforming coding agents that previously ran mostly in private terminals or browser tabs into “code channels” where teams can collaborate. After a user mentions a supported agent in a regular channel or direct message, the agent can automatically create a temporary channel. Members can add follow-up prompts, share files, review work status, and jointly approve changes. Channels can be public or private and can be closed or archived when work is complete. After seven days of inactivity, they are removed from the sidebar, while their messages remain searchable.
The interface separates chat from agent-generated outputs. The Artifacts section can display code diffs, Canvases, file links, and live HTML provided by the agent. Users can comment directly on specific lines of code and submit multiple pieces of feedback to the agent at once. Agents can also publish statuses such as `Working`, `Needs attention`, and `Done`, while users can stop responses that take too long. Quick actions, provided by integration developers, offer shortcuts for tasks such as creating a pull request or opening an external build tool or documentation. Slack Code therefore functions more like a shared control plane for agent sessions than a place to paste generated code into chat messages.
On security, Slack says code channels inherit the visibility settings, Enterprise Key Management, DLP, and Discovery API capabilities of standard channels, and agents must be installed through the Slack Marketplace. This allows prompts, comments, and artifacts to fall under existing retention and audit policies. It does not, however, automatically narrow the GitHub, cloud, or deployment permissions granted to an agent. Commits, merges, and environment changes performed through external tools still depend on the integration’s permissions, branch protection rules, and CI gates.
There are also discrepancies in the initial scope of support. Salesforce’s announcement lists Claude Code, Devin, GitHub Copilot, ChatGPT, and Vercel agents, while Slack’s help documentation currently names only Claude, Devin, GitHub Copilot, and Vercel as approved agents. Slack says the code-channel API will eventually be made available to platform developers, so it should not yet be treated as open to arbitrary agents. Engineering teams should next verify which agents are actually available in their tenant, the retention rules for artifacts, and whether “team approval” can be mapped to repository and deployment controls that cannot be bypassed.