Agents

Agents in Venue.sh let your organization define purpose-built AI agents with custom instructions, example prompts, and capability bindings. Each agent can be selected from the Venue.sh Assistant to handle specific workflows - from answering questions about your catalog to performing tool-assisted tasks across your engineering platform.

Key Capabilities

• Organization registry - Browse and search all agents configured for your organization.
• Custom instructions - Define what an agent does and how it should respond, in plain natural language.
• Capability bindings - Grant agents access to Venue tools, MCP server tools, and other agents.
• Templates - Start from a predefined template to get up and running quickly.
• Safety controls - Require human approval before an agent executes sensitive actions.
• Lifecycle management - Move agents through Draft, Ready, Disabled, and Archived states.

Browsing Agents

1. Open Agents from the sidebar (under AI).
2. Use the search box to filter by name.
3. Use the Status filter to narrow by agent state.
4. Click an agent name to open its detail page.

The list shows each agent's name, description, status, capability summary, creator, and timestamps.

Creating an Agent

Admins and Editors can create agents. Contributors, Viewers, and Demo users can browse agents but cannot add or edit them.

1. Go to Agents and click Create agent.
2. OPTIONAL: Select a template to pre-fill the name, description, instructions, and example prompts.
3. Enter a Name for the agent.
4. Enter a Description summarizing the agent's purpose.
5. Enter Instructions - write what this agent does and how it should respond. Instructions are sent to the AI model as the system prompt for every conversation that uses this agent.
6. OPTIONAL: Add Example prompts - short phrases that guide users on how to interact with the agent. Type a prompt and click Add (or press Enter) to include it.
7. OPTIONAL: Enable Require approval for sensitive actions to have the agent pause and wait for a human decision before executing actions flagged as sensitive.
8. OPTIONAL: Open Capability bindings to select the tools this agent may use (see Capability Bindings below).
9. Click Create agent. Venue.sh creates the agent in Draft status and opens its detail page.

Capability Bindings

Capability bindings control which tools and sub-agents an agent can invoke at runtime. Three types are available:

• Venue tools - Built-in Venue.sh tools such as catalog search, entity lookup, and other platform features.
• MCP tools - Individual tools from your organization's registered MCP servers. Tools are grouped by server. Use Add all to bind every tool from a server at once, or select individual tools with the checkboxes.
• Agents - Other agents in your organization. Binding an agent as a capability allows this agent to delegate specific tasks to it during a conversation.

A summary of all selected tools appears at the top of the picker and can be expanded to review or remove individual bindings.

Capability bindings can be configured at creation time or updated later by editing the agent.

Viewing Agent Details

The detail page shows:

• Name and status - The agent name with a status chip indicating its current lifecycle state.
• Metadata - Creator and last updated timestamp.
• Instructions - The system prompt the agent uses.
• Example prompts - Suggested phrases for users, if configured.
• Capability bindings - All tools and sub-agents bound to this agent, filterable by type (All, Venue tools, MCP tools, Agents). Each binding card shows the tool name, type, source server (for MCP tools), and whether the binding requires approval.

Agent Lifecycle

Agents move through the following states:

Status	Meaning
Draft	Newly created; not yet available for use in conversations.
Ready	Active and available for selection in the Assistant.
Needs attention	One or more capability bindings are degraded or disconnected.
Disabled	Temporarily unavailable for new conversations.
Archived	Permanently retired; no longer available for selection.

The following actions are available from the agent detail page:

• Mark as ready - Make a Draft or Needs attention agent available for use in the Assistant.
• Duplicate - Create a copy of the agent with the same name, instructions, and capability bindings.
• Disable agent - Prevent the agent from being selected for new conversations. Restore it later with Restore.
• Restore - Re-enable a Disabled agent by setting it back to Ready.
• Archive agent - Permanently retire the agent. Archived agents cannot be restored.

Example Agents

Here are some example agents that you could build.

Atlassian Agent

This is an example of an agent that can be created to help you interact with Atlassian products such as Jira, Confluence, and Bitbucket.

• Have a repository imported that is imported from a Bitbucket connection linked to a Jira project and Confluence space
• Add the Atlassian Rovo MCP Sever to the dashboard via the MCP Servers
  • Information about how to use the API token to access the Atlassian Rovo MCP server can be found here
  • At the time of writing (July 2026), the Atlassian Rovo MCP server exposes around 45 tools
• Create an agent with the following instructions:

  • Name and description: Atlassian Agent

  • Instructions:

    • You are an expert assistant for Atlassian products (Jira, Confluence, and Bitbucket) within this organization.
      
      Use the available tools to answer questions and perform actions:
      
      - Use the catalog-search, catalog-entity, repo-search, and repo-summary tools to find the repository or component the user is asking about, and to identify its linked Jira project
      - Use the jira tool for quick lookups of issues, projects, and sprints already associated with the repository in the Venue catalog
      - Use the Atlassian Rovo MCP tools for anything the jira tool cannot answer, such as issue details, comments, worklogs, sprint reports, Confluence pages, and Bitbucket pull requests
      
      When responding:
      
      - Confirm which Jira project or repository you are referencing before giving a status update
      - Summarize sprint, issue, or pull request status in plain language: what is done, what is in progress, and what is blocked
      - Include issue keys and links when available so the user can jump directly to Jira or Confluence
      - Before creating, updating, or transitioning a Jira issue, confirm the target issue and the intended change with the user
      - If you cannot find a linked Jira project or repository for a request, say so and ask the user to confirm the project key or repository name

  • Example prompts:

    • "How is the express-api project doing?"
    • "What is the status of the Sprint 12?"

  • Capability bindings:

    • Venue tools:
      • catalog-search
      • catalog-entity
      • jira
      • repo-search
      • repo-summary
    • MCP tools:
      • Look for the Atlassian Rovo MCP server that was added to the dashboard. For example:
        • atlassianUserInfo
        • getJiraIssue
        • Other tools of interest that are available to your Atlassian Rovo MCP server

DevOps Agent

This is an example of an agent that can be created to help you troubleshoot Helm-based services and their live state in Kubernetes.

• Have Helm-based repositories imported into the catalog so the agent can find each service's chart, values, and repository
• Add a Kubernetes MCP server to the dashboard via the MCP Servers, such as kubectl-mcp-server
  • kubectl-mcp-server exposes tools for pods, deployments, logs, events, and Helm releases, plus higher-level diagnostics such as diagnose_pod_crash and compare_namespaces
  • Configure the server with read-only or non-destructive mode (e.g., --read-only or --disable-destructive) if you want the agent limited to investigation rather than making changes to the cluster
• Create an agent with the following instructions:

  • Name and description: DevOps Agent

  • Instructions:

    • You are an expert in Kubernetes and Helm, helping engineers understand and troubleshoot the services running in this organization's clusters.
      
      Use the available tools to answer questions and perform actions:
      
      - Use the catalog-search, catalog-entity, repo-search, and repo-summary tools to find the repository or component the user is asking about, and to identify its Helm chart, values, and target namespace.
      - Use the Kubernetes MCP tools to inspect live cluster state, such as pods, deployments, logs, events, and Helm releases.
      - Prefer read-only tools (get_pods, get_logs, get_pod_events, helm_status, helm_get_values, kubectl_describe) before suggesting or taking any action that changes cluster state.
      
      When responding:
      
      - Confirm which service, namespace, and cluster context you are referencing before giving a status update.
      - When troubleshooting, walk through the diagnosis: what you checked, what you found, and the likely root cause, before recommending a fix.
      - Reference the Helm release and chart values when a service's configuration is relevant to the issue.
      - Before running any command that changes cluster or Helm state (_e.g.,_ scaling, rolling back, or upgrading a release), confirm the target and the intended change with the user.
      - If you cannot find a catalog entity or Helm release for a request, say so and ask the user to confirm the service name, namespace, or cluster context.

  • Example prompts:

    • "Why is the payments-api pod crashing in staging?"
    • "Show me the Helm values for the checkout-service release in production."

  • Capability bindings:

    • Venue tools:
      • catalog-search
      • catalog-entity
      • repo-search
      • repo-summary
    • MCP tools:
      • Look for the Kubernetes MCP server that was added to the dashboard. For example:
        • get_pods
        • get_logs
        • diagnose_pod_crash
        • helm_status
        • helm_get_values
        • Other tools of interest that are available to your Kubernetes MCP server

Useful Things to Ask About

The Assistant can answer questions about your organization's agents. For example:

• "What agents are available in our organization?"
• "Summarize the available agents"
• "Which agents have access to Jira tools?"

Permissions

Role	Permissions
Admin, Editor	Create, Read, Update, and Delete
Contributor, Viewer, Demo	Read only

Create, update, and delete actions on agents are recorded in the Activity Log.

Troubleshooting

Situation	What to do
Agent is not selectable in the Assistant	The agent may be in Draft, Disabled, or Archived status. Open Agents, find the agent, and click Mark as ready from the detail page.
Capability binding shows Needs attention	A bound MCP server or tool may have been removed or become unreachable. Open the agent, review the capability bindings, and remove or replace the affected binding.
Cannot see the Create agent button	Your role may not include Agents create permission. Contact your Venue administrator.
An archived agent needs to be recovered	Archived agents cannot be restored. Use Duplicate on a similar agent, or create a new agent with the same configuration.
Agent requires approval unexpectedly	The agent was created with Require approval for sensitive actions enabled. Edit the agent and toggle the setting off if approval is no longer needed.

See Also

• Assistant - Using the Assistant in the Venue.sh portal.
• MCP Servers - Register and inspect MCP servers whose tools you can bind to agents.
• Skills - Browse agent skills auto-discovered from connected repositories.
• Activity Log - Audit trail for configuration changes.