Agent & Pipeline Builder

Create an agent called “Code Reviewer” that reviews Python code for PEP 8 compliance, security issues, and suggests improvements

Create a pipeline called “CI Pipeline” that runs code analysis, then executes tests, and finally generates a report

• Ask User enabled: The conversation pauses and asks 1-4 focused questions using structured prompts with single-select, multi-select, or free-text options.
• Ask User disabled: The conversation requests missing information in a normal chat response or proceeds with reasonable defaults.

• After successful creation, the agent or pipeline automatically opens in Canvas mode for review and editing.
• An agent chip also appears in the chat, which you can click later to reopen the agent.
• The created agent is added to your project and can be accessed later via the Agents menu.

• My GitHub Toolkit — Available via MCP on; selected tools include list_pull_requests
• Code Review Agent — mcp tag applied

List all open pull requests from the GitHub toolkit and ask the code review agent to review each one.

1. Agent calls My_GitHub_Toolkit_list_pull_requests (normalized from "My GitHub Toolkit" + list_pull_requests) to fetch open PRs
2. For each PR, calls Code_Review_Agent with a self-contained task that includes the PR title and diff inline
3. The code review agent runs its full prediction pipeline and returns its findings as text
4. The calling agent collects all reviews and presents a consolidated summary to the user

• Agent & Pipeline Builder: Enabled on the coordinator agent
• Reporting Agent and QA Agent both have the mcp tag applied

Summarize the last sprint and send the report through the reporting agent, then run the regression tests using the QA agent.

1. Coordinator discovers available agent tools from the MCP server
2. Calls Reporting_Agent with task: "Generate a sprint summary for the last sprint. Include completed items, blockers, and carry-overs."
3. Receives the sprint report as text
4. Calls QA_Agent with task: "Run the regression test suite and return the pass/fail summary."
5. Receives the test results
6. Combines both outputs and returns a single status update to the user

• Agent & Pipeline Builder: Enabled
• Dev GitHub toolkit — Available via MCP on; selected tools include list_pull_requests, merge_pull_request
• Project Jira toolkit — Available via MCP on; selected tools include search_issues, update_issue

• Dev_GitHub_list_pull_requests, Dev_GitHub_merge_pull_request
• Project_Jira_search_issues, Project_Jira_update_issue

Find all Jira tickets marked "Ready for Release" and merge the associated GitHub pull requests.

1. Agent calls Project_Jira_search_issues with JQL status = "Ready for Release"
2. For each ticket, extracts the linked PR number from the ticket fields
3. Calls Dev_GitHub_merge_pull_request for each PR
4. Calls Project_Jira_update_issue to transition each ticket to “Released”
5. Returns a summary: merged PRs and updated Jira tickets

• Agent & Pipeline Builder: Enabled on the coordinator agent
• Test Runner Agent — mcp tag applied; runs the test suite and returns results
• Failure Analysis Agent — mcp tag applied; classifies failures as known regressions or new bugs
• QA Jira toolkit — Available via MCP on; selected tools include create_issue

Run today's regression suite, analyze any failures, and log new bugs.

1. Coordinator calls Test_Runner_Agent with task: "Run the full regression suite and return pass/fail per test case."
2. Agent receives structured results with failure details
3. Coordinator calls Failure_Analysis_Agent with task: "Classify the following failures as known regressions or new bugs: {failures_list}"
4. Analysis agent returns a categorized list
5. For each new bug, coordinator calls QA_Jira_create_issue with the failure details
6. Coordinator sends a final summary: tests run, failures found, bugs filed

• With Ask User enabled: The conversation can ask structured clarifying questions (1-4 questions) with interactive single-select, multi-select, or free-text options when required information for agent or pipeline creation is missing. Questions display with progress indicators (“Question X of Y”) and navigation controls.
• Without Ask User: The conversation can only request missing information through normal chat responses, which is less structured and may require multiple back-and-forth exchanges.

• In Project Settings: Toggle both Agent & Pipeline Builder and Ask User under Default Modules
• In Agent Configuration: Enable both modules in the MODULES section
• In Conversations: Enable both from the Modules flyout panel

• Provide clear, detailed instructions that explain what the agent or pipeline should accomplish
• Use descriptive names (max 32 characters) that indicate the agent purpose
• Include specific requirements, constraints, and expected behavior in instructions
• Specify model preferences (model name, temperature) when you have specific needs
• For pipelines: provide complete node structure and edge connections in YAML format, including node types (function, toolkit, mcp, tool, loop, loop_from_tool, indexer, code, state_modifier)
• Test the created agent or pipeline immediately to verify it works as expected
• Add relevant tags to improve discoverability across your project

• Apply the mcp tag to agents you want callable via Agent & Pipeline Builder
• Use descriptive agent names that will normalize clearly (avoid special characters)
• Keep agent descriptions clear and focused so the calling agent understands when to use each tool
• Ensure agent instructions are self-contained since called agents receive only the task parameter
• Test agent-to-agent calls to verify the task parameter provides sufficient context

• Enable Available via MCP only for toolkits you want exposed via Agent & Pipeline Builder
• Carefully select which tools within each toolkit should be exposed
• Keep toolkit names descriptive and unique to avoid name collisions after normalization
• Verify toolkit credentials are valid and not expired before relying on toolkit tools
• Document which toolkits are MCP-enabled for your team

• Enable Agent & Pipeline Builder by default in Project Settings if your team frequently creates agents or uses coordinator patterns
• Enable Agent & Pipeline Builder for agents that need to delegate to other agents or use platform toolkits dynamically
• Consider enabling Smart Tools Selection when many agents and toolkits are available to reduce token usage
• Review and organize created agents periodically to maintain project quality
• Use appropriate permissions to control who can create and modify agents

• Provide complete information (name, description, instructions, model) upfront to minimize clarification rounds
• Be specific when creating agents or pipelines. Include all required fields and behavior details
• Review created agents in Canvas mode before using them in production
• Search for existing agents before creating duplicates to avoid redundancy
• If agent creation fails, verify all required fields meet validation requirements

1. Open the agent in the Agents menu
2. Go to the agent’s settings and add the mcp tag
3. Save the agent

• Ensure the target agent’s LLM is healthy and its configuration is complete
• Check if the target agent has its own toolkits that may be failing to initialize
• Simplify the task argument — the called agent must be able to complete it independently

• The user has access to the project containing the toolkit
• The toolkit’s stored credentials (API keys, tokens) are valid and not expired
• The toolkit’s Available via MCP option is still enabled

• Create a new ELITEA personal access token
• Renew the existing token if it has expired
• Retry the action after updating the token