CHAPTERS
- 0:03 – 0:34
Why per-user permissions break in group channels
Noah explains the core access-control problem when Claude is used in a multi-person Slack channel. In collaborative and proactive scenarios, there may be no single “requesting user,” making traditional user-based permissioning ambiguous and unsafe.
- •In a channel, “act as the user” becomes unclear (whose permissions apply?)
- •Least-privileged vs last-requester models are both problematic
- •Proactive work (e.g., alerts at 3 AM) may have no requesting user at all
- •Collaboration across multiple teammates requires a new permissioning model
- 0:34 – 1:04
Claude Tag’s solution: the agent acts as itself
Claude Tag is designed so the agent has its own identity, accounts, and service keys in your systems. Access becomes predictable for end users and easier for security teams to audit and reason about.
- •Claude has its own account/service keys rather than borrowing a user’s identity
- •Admins assign access granularly across workspaces and channels
- •What Claude can reach doesn’t change based on who tags it
- •Predictability for users and clearer auditability for security teams
- 1:04 – 1:34
Admin UI walkthrough: scopes for workspace vs private channels
The setup begins in the Claude Tag admin UI, where scopes define where access applies. Noah shows a workspace-wide scope and a private channel scope, both initially with zero access attached.
- •Open Claude Tag admin UI to configure access
- •Two scopes shown: overall workspace and a private data-team channel
- •Scopes can exist without any attached access
- •With zero access, tagging Claude cannot reach external systems
- 1:34 – 1:50
Establishing baseline access: workspace as the common denominator
Noah introduces the idea of a workspace baseline—access that’s safe for any member of the workspace to benefit from. This baseline is implemented by creating an access bundle and attaching it to the workspace scope.
- •Workspace baseline = access acceptable for everyone in the company workspace
- •Baseline access should be conservative and broadly useful
- •Baseline is applied via an “access bundle”
- •Attaching the bundle to the workspace grants it everywhere by inheritance
- 1:50 – 2:05
Creating an access bundle with agent-owned credentials
An access bundle is a named package of connections, repo access, skills, and standing instructions. Noah adds a project management connection using credentials created specifically for Claude (not a human user) and tests them.
- •Access bundles group: connections, repository access, skills, and standing instructions
- •Example bundle name: “General Tooling” for broad workspace use
- •Credentials are provisioned for Claude’s agent account, not a teammate’s account
- •Credentials are scoped on the provider side and then tested in setup
- 2:05 – 2:35
Attaching the bundle: inherited access across all channels
After creating the bundle, Noah attaches it to the workspace scope, making it available in every channel by default. Now anyone tagging Claude can use the project tracker access under Claude’s identity.
- •Attach the access bundle to the workspace scope to activate it
- •All channels inherit workspace-level access automatically
- •Tagging Claude anywhere enables consistent access behavior
- •Claude accesses tools under its own name and scopes
- 2:35 – 3:05
Channel step-up: granting sensitive tools only to a specific team
For the data team’s private channel, Noah creates a second access bundle with read access to the data warehouse and assigns it only to that channel. The channel becomes the boundary so the credential is unavailable elsewhere.
- •Create a separate bundle for more sensitive resources (e.g., data warehouse)
- •Assign the bundle only to the private data-team channel
- •The credential exists only within that channel’s scope
- •Outside the channel, Claude cannot reach the stepped-up tool
- 3:05 – 3:36
DMs for strictly personal or need-to-know credentials
Some tools should never be placed in shared scopes (e.g., recruiting or people data). Noah explains that DMs are the right place for personal access because Claude runs under the individual’s own Claude account and connections.
- •Highly sensitive/need-to-know tools shouldn’t be shared in workspace or channel scopes
- •Use a DM for personal credentials and personal tool access
- •In DMs, Claude runs your own Claude AI account and your connections
- •This separates “team agent” behavior from “personal assistant” behavior
- 3:36 – 4:06
Putting it together: the three-tier identity and access model
Noah summarizes the model as a layered approach: workspace baseline, channel step-ups, and DMs for personal access. This allows a whole channel to collaborate with a consistent Claude identity without relying on individuals’ credentials.
- •Workspace baseline = common denominator access across the company
- •Channel step-ups = additional access for specific teams
- •DMs = personal assistant mode with individual credentials
- •Consistency: Claude’s capabilities don’t change as participants change
- •Shared steering of one agent without personal credentials in play
- 4:06 – 4:28
Wrap-up: predictable permissions that scale to proactive collaboration
The closing reinforces the benefit: configure once and the team benefits from stable, auditable access boundaries. Noah invites teams to adopt the model as they start tagging Claude in workflows.
- •Stable capability model prevents permission confusion in threads
- •Setup-once approach simplifies ongoing team usage
- •Security posture improves through clear boundaries and audits
- •Encouragement to apply the model in real team tagging workflows
