From e65936dae456f6153c5a84d658f93d788df6c94a Mon Sep 17 00:00:00 2001
From: Thomas Hutcheson <thomashutcheson3080@gmail.com>
Date: Mon, 22 Sep 2025 20:42:31 +0100
Subject: [PATCH 1/2] feat: Add documentation for early exit mechanism to
 SequentialAgent using escalate action #2999

---
 .../workflow-agents/sequential-agents.md      | 55 ++++++++++++++++++-
 1 file changed, 53 insertions(+), 2 deletions(-)

diff --git a/docs/agents/workflow-agents/sequential-agents.md b/docs/agents/workflow-agents/sequential-agents.md
index a2f1e16a4..4f13c952e 100644
--- a/docs/agents/workflow-agents/sequential-agents.md
+++ b/docs/agents/workflow-agents/sequential-agents.md
@@ -21,6 +21,59 @@ When the `SequentialAgent`'s `Run Async` method is called, it performs the follo
 
 ![Sequential Agent](../../assets/sequential-agent.png){: width="600"}
 
+### Early Exit with `exit_sequence`
+
+The `SequentialAgent` supports early termination when a sub-agent encounters a terminal condition that makes continuing the sequence unnecessary or impossible.
+
+#### When to use early exit:
+- **Validation failures:** A validation agent detects critical errors
+- **Empty results:** A search agent finds no data to process
+- **Blocking conditions:** An agent hits an unrecoverable state
+- **Conditional logic:** Based on input analysis, remaining steps aren't needed
+
+#### Usage
+
+Sub-agents can call the `exit_sequence` tool to terminate the sequence early:
+
+```python
+from google.adk.tools import exit_sequence
+
+# In your agent's tool function
+def validate_input(tool_context: ToolContext):
+    if critical_validation_fails:
+        exit_sequence(tool_context)  # Stops the sequence here
+        return "Validation failed - terminating sequence"
+```
+
+#### Example: Search → Analysis → Report Pipeline
+
+```python
+from google.adk.agents import LlmAgent, SequentialAgent
+from google.adk.tools import exit_sequence, google_search
+
+search_agent = LlmAgent(
+    name="SearchAgent",
+    instruction="Search for results. If no useful results found, call exit_sequence.",
+    tools=[google_search, exit_sequence]
+)
+
+analysis_agent = LlmAgent(
+    name="AnalysisAgent", 
+    instruction="Analyze the search results."
+)
+
+report_agent = LlmAgent(
+    name="ReportAgent",
+    instruction="Generate final report."
+)
+
+# If search finds no results, analysis and report agents won't execute
+pipeline = SequentialAgent(sub_agents=[search_agent, analysis_agent, report_agent])
+```
+
+!!! note "Async Mode Only"
+    Early exit via `exit_sequence` is currently supported in async mode only. Live streaming mode uses the existing `task_completed()` mechanism for natural completion.
+
 ### Full Example: Code Development Pipeline
 
 Consider a simplified code development pipeline:
@@ -51,5 +104,3 @@ This ensures the code is written, *then* reviewed, and *finally* refactored, in
         ```java
         --8<-- "examples/java/snippets/src/main/java/agents/workflow/SequentialAgentExample.java:init"
         ```
-
-    

From f5b8da6b4f133623430027d157598c6fbd8ec048 Mon Sep 17 00:00:00 2001
From: Thomas Hutcheson <thomashutcheson3080@gmail.com>
Date: Sun, 19 Oct 2025 17:24:22 +0100
Subject: [PATCH 2/2] docs: Update early exit mechanism documentation for
 SequentialAgent to include live mode support

---
 docs/agents/workflow-agents/sequential-agents.md | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/docs/agents/workflow-agents/sequential-agents.md b/docs/agents/workflow-agents/sequential-agents.md
index 4f13c952e..f4099b75d 100644
--- a/docs/agents/workflow-agents/sequential-agents.md
+++ b/docs/agents/workflow-agents/sequential-agents.md
@@ -71,8 +71,11 @@ report_agent = LlmAgent(
 pipeline = SequentialAgent(sub_agents=[search_agent, analysis_agent, report_agent])
 ```
 
-!!! note "Async Mode Only"
-    Early exit via `exit_sequence` is currently supported in async mode only. Live streaming mode uses the existing `task_completed()` mechanism for natural completion.
+!!! note "Live Mode Support"
+    Early exit via `exit_sequence` is supported in both async and live streaming modes. In live mode, the escalate action provides immediate termination, whilst `task_completed()` remains available for natural completion signals.
+
+!!! info "Similar to LoopAgent"
+    The `exit_sequence` tool works similarly to [`exit_loop`](loop-agents.md) in LoopAgent, providing a consistent early-exit pattern across workflow agents.
 
 ### Full Example: Code Development Pipeline