🛸 Autonomous Cloud SRE Agent (ADK + Antigravity with `uv`)

Welcome to the production-grade template for building, testing, and deploying an autonomous Site Reliability Engineering (SRE) Agent in Google Cloud. This system integrates the Google Agent Development Kit (ADK) for multi-agent diagnostic graphs and the Google Antigravity SDK for the agent runtime, deny-by-default safety policies, and safe execution.

The agent monitors microservices, queries distributed traces, correlates logs, diagnoses cascade bottlenecks, auto-generates comprehensive incident post-mortems, and lets you download them straight from the chat UI. The whole stack runs locally with zero GCP credentials thanks to a mock-telemetry mode.

📦 Source code: github.com/xSAVIKx/sre-agent

Google Cloud credits are provided for this project.

#AgenticArchitect #GoogleAntigravity

⚡ Featured Capabilities

1. ⛓️ Multi-Service Cascade Latency & Bottleneck Analyzer

When a request spikes in latency, the SRE Agent dissects the distributed trace. It calculates the inclusive vs. exclusive (self) execution time for every span, rendering a contribution table that pins down the exact bottleneck — e.g. a gateway request that looks 10 s slow, where 99.3% of the time is actually trapped in one database span three levels down.

2. 📄 Automated Incident Post-Mortem Generator

Following a diagnosis, the agent compiles a complete Incident Post-Mortem (RCA) including:

Incident Overview: Date/time, root service, Trace ID, impact duration, and status.
Timeline: Trace timestamps for the gateway alert, cascading failure, and mitigation.
Root Cause Analysis: Connection states, active chaos injections, or infrastructure timeouts.
Prevention Plan: Immediate remediation, short-term workarounds, and long-term preventions.

3. 📥 Interactive Chat Downloader

The web chat parses diagnostic reports server-side; when it detects a post-mortem it renders a premium, styled Download Button that exports the report to markdown (post_mortem.md) entirely client-side via the Blob API.

🏗️ Architecture

Four FastAPI services collaborate over HTTP (Agent-to-Agent / A2A) with results streamed back as Server-Sent Events (SSE). The user-facing Orchestrator is locked to a deny-by-default policy — its only capability is to delegate to the read-only SRE diagnostics agent.

flowchart LR
    User(["👤 On-call engineer"]) -->|"/chat (SSE)"| ORCH

    subgraph Safe["🛡️ Orchestrator · service: sre-agent"]
        ORCH["Antigravity runtime<br/>policy = deny('*'), allow('diagnose_sre')"]
    end

    ORCH -->|"diagnose_sre — A2A/SSE"| SRE["🔬 SRE diagnostics<br/>service: sre-sub-agent<br/>ADK: TraceAnalyzer ➜ LogCorrelator"]
    SRE -->|"topology"| INV["📚 Inventory agent<br/>service: inventory-agent"]
    INV --> FS[("Firestore")]
    SRE -->|"read-only · or MOCK_GCP"| OBS[("☁️ Trace · Logging · Monitoring")]
    APP["🐒 Target app<br/>service: sre-chaos-monkey"] -->|"write-only telemetry"| OBS
    SRE -->|"report + 📥 post-mortem"| ORCH --> User

Service	Package	Role
Orchestrator	`agent/`	User-facing agent + web chat UI; delegates via the `diagnose_sre` tool.
SRE diagnostics	`sre_agent/`	The engine: observability tools + the ADK multi-agent workflow.
Inventory	`inventory_agent/`	Discovers & caches the project topology (Cloud Run services + databases).
Target app	`app/`	OpenTelemetry-instrumented "chaos monkey" that generates synthetic incidents.
Shared lib	`sre_common/`	`otel_trace`, `retry_async`, `setup_logging`, trace-context middleware.

📂 Repository Layout

.
├── README.md · AGENTS.md · BLOGPOST.md · CODELAB.md
├── pyproject.toml          # Root uv workspace (5 members)
├── uv.lock
├── docker-compose.yaml     # Full local multi-service stack + Firestore emulator
├── cloudbuild.yaml         # Root Cloud Build config
├── bootstrap.sh            # Interactive GCP project setup (writes .env)
├── deploy.sh               # Least-privilege Cloud Run deploy
├── cleanup.sh              # GCP resource teardown
├── simulate_incident.py    # Local standalone simulation (no GCP needed)
│
├── app/                    # 🐒 Target FastAPI app (OpenTelemetry-instrumented)
│   ├── main.py             # Gateway → Backend → Database incident generator
│   ├── Dockerfile · cloudbuild.yaml · pyproject.toml
│
├── agent/                  # 🛡️ Orchestrator service (user-facing + web UI)
│   ├── src/agent/
│   │   ├── config.py           # Antigravity safety policies & runtime loader
│   │   ├── routes.py           # FastAPI endpoints (/chat UI + SSE, /diagnose)
│   │   ├── main.py             # FastAPI app wiring
│   │   ├── a2ui_translator.py  # Markdown → rich A2UI schema (download button)
│   │   ├── firestore_strategy.py
│   │   └── index.html          # Premium web chat interface
│   ├── test/
│   └── Dockerfile · cloudbuild.yaml · pyproject.toml
│
├── sre_agent/              # 🔬 SRE diagnostics engine
│   ├── src/sre_agent/
│   │   ├── gcp_tools.py     # Trace/log/metric tools + cascade & post-mortem
│   │   ├── sre_workflow.py  # ADK multi-agent orchestration (two tiers)
│   │   ├── routes.py        # A2A SSE endpoint /v1/agents/sre/messages
│   │   ├── registry.py      # @register_tool decorator
│   │   ├── itinerary.py · config.py · firestore_strategy.py · main.py
│   ├── test/
│   └── Dockerfile · cloudbuild.yaml · pyproject.toml
│
├── inventory_agent/        # 📚 Infrastructure topology discovery
│   ├── src/inventory_agent/{main,routes,discovery,config,firestore_strategy}.py
│   └── Dockerfile · cloudbuild.yaml · pyproject.toml
│
├── sre_common/             # 🧰 Shared library
│   └── src/sre_common/{otel,retry,logging,middleware}.py
│
└── skills/                 # 🧩 Portable Antigravity skill (mirror of the engine)
    └── sre_incident_solver/{SKILL.md, sre_workflow.py, gcp_tools.py, registry.py}

📖 Key Deliverables

Step-by-step Tutorial: Build the agent from scratch in CODELAB.md.
Editorial Technical Post: The engineering architecture & design rationale in BLOGPOST.md.
Contributor Guide: Conventions for AI and human collaborators in AGENTS.md.
Follow-Up Exercises: Turn the codelab into your own project with hands-on extension challenges in EXERCISES.md.

🚀 Quickstart: Local Standalone Simulation

Run the entire diagnostic workflow locally in seconds with uv. No GCP account, project, or credentials required.

1. Clone the Repository

git clone https://github.com/xSAVIKx/sre-agent.git
cd sre-agent

2. Synchronize Dependencies

pip install uv
uv sync --all-packages

3. Run the Incident Simulation

uv run simulate_incident.py

This single command:

Triggers the mock target app gateway to generate a synthetic database-timeout incident.
Writes mock traces and logs to mock_telemetry_data/ (gitignored).
Boots the Orchestrator in mock mode, which calls diagnose_sre and runs the workflow in-process.
Prints the structured telemetry plus the full diagnosis — the /api/database 99.3% bottleneck table and the complete # 🚨 Incident Post-Mortem — straight to your terminal.

Want the full multi-service experience (Orchestrator + SRE + Inventory + Firestore emulator + target app) with the web chat UI?

docker-compose up --build   # then open the chat at /chat

☁️ Production Deployment: Google Cloud Run

Deploy to Cloud Run following least-privilege best practices — each service gets its own minimally-scoped service account.

1. Bootstrap GCP Settings

./bootstrap.sh   # gcloud auth login, set project + region, write .env

2. Deploy the Services

./deploy.sh

This enables the required APIs (Run, Cloud Build, Trace, Logging, Monitoring, Artifact Registry, Firestore, Secret Manager), provisions the service accounts, grants least-privilege roles, then builds and deploys four Cloud Run services:

Service account	Used by	Roles
`sre-chaos-monkey-sa`	target app (`sre-chaos-monkey`)	`cloudtrace.agent`, `logging.logWriter` (write-only telemetry)
`sre-agent-sa`	SRE diagnostics (`sre-sub-agent`)	`cloudtrace.user`, `logging.viewer`, `monitoring.viewer`, `datastore.user` (read-only)
`inventory-agent-sa`	inventory agent (`inventory-agent`)	`datastore.user`, `run.developer`, `logging.logWriter`
`sre-build-sa`	Cloud Build	`run.admin`, `storage.admin`, `artifactregistry.writer`, `logging.logWriter`

The split is the point: the app that generates chaos can only write telemetry; the agent that investigates it can only read.

⚙️ The Antigravity Ecosystem

This codebase shows three ways to work with the SRE agent:

1. The Antigravity SDK

Used programmatically in agent/src/agent/config.py to configure the Orchestrator: system instructions, registered tools, and a strict deny-by-default safety policy — [deny("*"), allow("diagnose_sre")]. The Orchestrator literally cannot read files, run commands, or hit arbitrary URLs; its only move is to delegate to the read-only SRE sub-agent.

2. The Antigravity CLI (`agy`)

Developers can drive and inspect the workspace from the terminal. The most direct way to exercise the full diagnostic loop is the local simulation above (uv run simulate_incident.py); the CLI also discovers the reusable skill under skills/sre_incident_solver/.

3. Antigravity 2.0 (Visual Workspace)

The desktop application auto-discovers skills placed in the skills/ directory. Opening this repository surfaces the sre_incident_solver skill via SKILL.md, letting you run and audit SRE tasks from a graphical canvas.

🧹 Tearing Down the Stack

To prevent ongoing billing charges, remove all deployed Cloud Run services, IAM bindings, and service accounts:

./cleanup.sh

Navigation Menu

Repository files navigation

🛸 Autonomous Cloud SRE Agent (ADK + Antigravity with `uv`)

⚡ Featured Capabilities

1. ⛓️ Multi-Service Cascade Latency & Bottleneck Analyzer

2. 📄 Automated Incident Post-Mortem Generator

3. 📥 Interactive Chat Downloader

🏗️ Architecture

📂 Repository Layout

📖 Key Deliverables

🚀 Quickstart: Local Standalone Simulation

1. Clone the Repository

2. Synchronize Dependencies

3. Run the Incident Simulation

☁️ Production Deployment: Google Cloud Run

1. Bootstrap GCP Settings

2. Deploy the Services

⚙️ The Antigravity Ecosystem

1. The Antigravity SDK

2. The Antigravity CLI (`agy`)

3. Antigravity 2.0 (Visual Workspace)

🧹 Tearing Down the Stack

About

Contributors

Languages

Name		Name	Last commit message	Last commit date
agent		agent
app		app
blogpost_assets		blogpost_assets
inventory_agent		inventory_agent
skills/sre_incident_solver		skills/sre_incident_solver
sre_agent		sre_agent
sre_common		sre_common
.dockerignore		.dockerignore
.gcloudignore		.gcloudignore
.gitignore		.gitignore
AGENTS.md		AGENTS.md
BLOGPOST.md		BLOGPOST.md
CODELAB.md		CODELAB.md
EXERCISES.md		EXERCISES.md
LICENSE		LICENSE
README.md		README.md
bootstrap.sh		bootstrap.sh
cleanup.sh		cleanup.sh
cloudbuild.yaml		cloudbuild.yaml
deploy.sh		deploy.sh
docker-compose.yaml		docker-compose.yaml
pyproject.toml		pyproject.toml
simulate_incident.py		simulate_incident.py
uv.lock		uv.lock

Latest commit

History

Folders and files

Repository files navigation

🛸 Autonomous Cloud SRE Agent (ADK + Antigravity with uv)

⚡ Featured Capabilities

1. ⛓️ Multi-Service Cascade Latency & Bottleneck Analyzer

2. 📄 Automated Incident Post-Mortem Generator

3. 📥 Interactive Chat Downloader

🏗️ Architecture

📂 Repository Layout

📖 Key Deliverables

🚀 Quickstart: Local Standalone Simulation

1. Clone the Repository

2. Synchronize Dependencies

3. Run the Incident Simulation

☁️ Production Deployment: Google Cloud Run

1. Bootstrap GCP Settings

2. Deploy the Services

⚙️ The Antigravity Ecosystem

1. The Antigravity SDK

2. The Antigravity CLI (agy)

3. Antigravity 2.0 (Visual Workspace)

🧹 Tearing Down the Stack

About

Topics

Resources

Stars

Watchers

Forks

Contributors

Languages

🛸 Autonomous Cloud SRE Agent (ADK + Antigravity with `uv`)

2. The Antigravity CLI (`agy`)