Repository navigation
Closed as not planned
Labels
staleThis issue or PR has become staleThis issue or PR has become stale
Description
📋 Prerequisites
- I have searched the existing issues to avoid creating a duplicate
- By submitting this issue, you agree to follow our Code of Conduct
📝 Feature Summary
Declarative workflow patterns (Sequential, Parallel, Loop) from ADK's workflow agents
❓ Problem Statement / Motivation
This feature adds declarative workflow patterns (Sequential, Parallel, Loop) to kagent using ADK's workflow agents. Implementation follows Test-Driven Development with parallel execution where possible.
💡 Proposed Solution
ADK Workflow Example:
Reference: https://github.com/google/adk-docs/blob/main/docs/agents/workflow-agents/loop-agents.md
# Writer Agent
apiVersion: kagent.dev/v1alpha2
kind: Agent
metadata:
name: document-writer
namespace: kagent
spec:
type: Declarative
description: "Generates or refines document drafts"
declarative:
systemMessage: |
You are a skilled technical writer.
First iteration: Generate an initial draft based on the user's topic.
Subsequent iterations: Read previous draft (temp:document_draft) and
feedback (temp:critique), then improve the document.
Focus on:
- Clear structure
- Accurate technical content
- Engaging writing style
- Proper formatting
Store your draft in output_key: "document_draft"
modelConfig: default-model-config
tools:
- type: McpServer
mcpServer:
name: web-search-mcp
toolNames: ["search", "get_webpage_content"]
---
# Critic Agent
apiVersion: kagent.dev/v1alpha2
kind: Agent
metadata:
name: document-critic
namespace: kagent
spec:
type: Declarative
description: "Critiques documents and suggests improvements"
declarative:
systemMessage: |
You are an experienced editor and critic.
Read the current draft from temp:document_draft.
Evaluate:
- Structure and organization
- Clarity and readability
- Technical accuracy
- Completeness
- Grammar and style
Provide constructive criticism with specific suggestions for improvement.
If the document is excellent (no major issues), you may recommend stopping.
Store your critique in output_key: "critique"
modelConfig: default-model-config
---
# Iterative Document Improver with Loop
apiVersion: kagent.dev/v1alpha2
kind: Agent
metadata:
name: iterative-document-improver
namespace: kagent
spec:
type: Declarative
description: "Iteratively improves documents through write-critique-refine cycles"
declarative:
systemMessage: |
You orchestrate an iterative document improvement process.
The workflow loops through:
1. Writer creates/refines the document
2. Critic evaluates and provides feedback
This continues for up to 5 iterations or until the document quality is satisfactory.
After the loop completes, present the final polished document to the user.
modelConfig: default-model-config
subagents:
- role: "Iterative Document Improver"
description: "Loop workflow for document refinement"
delegationMode: Supervised
workflow:
type: Loop
agents:
- name: document-writer
description: "Writes or refines document drafts"
- name: document-critic
description: "Critiques and suggests improvements"
# the critic agent use exit_loop to stop the loop
loopConfig:
maxIterations: 5🔄 Alternatives Considered
No response
🎯 Affected Service(s)
None
📚 Additional Context
No response
🙋 Are you willing to contribute?
- I am willing to submit a PR for this feature
Reacted by Brian Fox
Metadata
Metadata
Assignees
Labels
staleThis issue or PR has become staleThis issue or PR has become stale
Type
Projects
- StatusShow more project fieldsBacklog
Activity
github-actions commentedon Feb 27, 2026
This issue has been marked as stale because of no activity in the last 90 days. It will be closed in the next 90 days unless it is tagged "no stalebot" or other activity occurs.
feat: add declarative workflow subagents (Sequential, Parallel, Loop)
marcellodesales commentedon Oct 30, 2026
@dimetron I'm looking for a workflow level for Kagent/ADK agents that can work similar to https://voltagent.dev/docs/quick-start/ where a set of namespaced agents are still not enough... If I want to provide the abstraction to multiple Engineering teams that they can self-organize, users are still looking at the various Agentic design-patterns raised in the PR #3344 ... The flat structure of the namespace does feel it it is missing that workflow abstraction where it can give users the proper way to describe how a multi-agent system is structured and how those agents communication within our system... or else we are still relying on the integration layer from the gateway to have them chat via A2A with yet other tools...
Maybe a group of us interested in similar capabilities could do something for the community?