Skip to content

Step 1 · Connect the agents in a workflow (12 min)

Goal: Make the SRE agent run its two ADK agents one after the other, with a code step between them.

Builds on: ADK basics: 4. Workflow, 5. Runner, session and events.

The idea: an ADK workflow is a graph

The SRE agent (sre_agent/src/sre_agent/sre_workflow.py) uses the Agent Development Kit (ADK). It has two agents:

Agent Input Output
🕵️ trace_analyzer The recent failing and slow requests One trace ID
🩺 log_correlator The spans and logs of that trace The root cause and a mitigation plan

Between the two agents, a plain Python function, fetch_telemetry, gets the spans, the logs and the project topology for the trace ID. A model is not necessary for this task, so it is code.

An ADK Workflow connects these nodes. A node can be an agent or a function. A tuple in edges is a chain: each node gets the output of the node before it.

flowchart TB
    S([START]) --> TA["🕵️ trace_analyzer"]
    TA -->|trace ID| FT["fetch_telemetry (Python)"]
    FT -->|spans + logs| LC["🩺 log_correlator"]

At this time, the chain stops after trace_analyzer. The trace ID goes nowhere, and the report has no root-cause analysis.

No API key? The agents still run

Without GEMINI_API_KEY, the agents use SimulatedLlm (sre_agent/src/sre_agent/simulated_llm.py). This is an ADK model class that answers with fixed rules, not with a neural network. The workflow, the agents and the tool calls are real ADK. Only the text comes from rules. With a key, the same agents use Gemini.

Your task

  1. Find the TODO:

    git grep -n "TODO(step-1)"

  2. In _run_adk_diagnostics, change edges to the chain START → trace_analyzer → fetch_telemetry → log_correlator.

Run it

  1. Do these commands:

    uv run workshop/check.py 1
uv run simulate_incident.py --engine-only

  2. Find the new sections at the start of the report: Root Cause Analysis, Observability Metrics and Recommended Mitigation. The log_correlator wrote them.

  3. Look at Observability Metrics. It says Not checked: this agent has no query_metrics tool. Step 2 fixes this.

With a GEMINI_API_KEY, Gemini writes its own text: the sections can have other titles, and there is no Not checked line. The agent has no tools yet, so it can only guess the cause from the spans and logs.

The cascade table and the post-mortem come after the agents' text. The workflow adds them from the tools, so they are always correct.

Discuss

  • Why is fetch_telemetry a function and not an agent? A model is slow, costs money and can make mistakes. Use a model only for the decisions that need one.
  • Why does trace_analyzer return only a trace ID? A small, exact output is easy to check: the workflow makes sure that the ID is one of the candidates.

Stuck? Run uv run workshop/step.py hint 1, or ask the workshop coach in Antigravity. To apply the solution: uv run workshop/step.py solve 1.

Hint and solution

Try the step first. Open the hint when you are stuck, and the solution only after the hint.

Hint: where to change the code

File: sre_agent/src/sre_agent/sre_workflow.py, in _run_adk_diagnostics(). The TODO says:

# TODO(step-1): The workflow stops after the TraceAnalyzer. Make it a chain of 4 nodes:
#   START -> trace_analyzer -> fetch_telemetry -> log_correlator.
#   A tuple in `edges` is a chain: each node gets the output of the node before it. Thus
#   fetch_telemetry gets the trace ID, and log_correlator gets the spans and logs.
Solution 
--- a/sre_agent/src/sre_agent/sre_workflow.py
+++ b/sre_agent/src/sre_agent/sre_workflow.py
@@ -272,11 +272,9 @@

     try:
         # Define the ADK 2.0 graph workflow
-        # TODO(step-1): The workflow stops after the TraceAnalyzer. Make it a chain of 4 nodes:
-        #   START -> trace_analyzer -> fetch_telemetry -> log_correlator.
-        #   A tuple in `edges` is a chain: each node gets the output of the node before it. Thus
-        #   fetch_telemetry gets the trace ID, and log_correlator gets the spans and logs.
-        sre_diagnostics_workflow = AdkWorkflow(name="sre_diagnostics_workflow", edges=[(START, trace_analyzer)])
+        sre_diagnostics_workflow = AdkWorkflow(
+            name="sre_diagnostics_workflow", edges=[(START, trace_analyzer, fetch_telemetry, log_correlator)]
+        )

         session_service = InMemorySessionService()
         runner = Runner(node=sre_diagnostics_workflow, app_name="sre_diagnostics", session_service=session_service)

To apply it to your code: uv run workshop/step.py solve 1.