palladius/adk-sre-benjamin

記録された最新のソース活動: 2026年8月25日 08:58

SkillsMP カタログ更新: 2026年10月29日 18:30

収集済み skills: 2

GitHub スター: 2
GitHub フォーク: 0

AI アシスタントでインストール

このプロンプトをコピーして、使用中の AI アシスタントに送ってください。

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/palladius/adk-sre-benjamin.

リポジトリの README

README の元ファイル：README.md

README の元ファイルを読む

README 全文を表示

# 🏰 Project Benjamin SRE Agentic Framework (v1.2.14) Project Benjamin is a production-grade, highly secure automation harness for Site Reliability Engineering (SRE), built using Google's **Antigravity Development Kit (ADK) for Python** and strictly aligned with the **IMAG (Incident Management At Google) Incident Command System (ICS)** command chain. This framework acts as the agentic brain wrapping your **SRE Extension** repository: 👉 [https://github.com/gemini-cli-extensions/sre](https://github.com/gemini-cli-extensions/sre) --- ## 🖥️ Live Incident Command Dashboard Here is the latest screenshot of the **Project Benjamin Web Dashboard**, featuring real-time incident state monitoring, timeline tracking, and live incident simulation capabilities. ![Project Benjamin Web Dashboard](doc/web_screenshot.png) --- ## 🚀 Key SRE Core Capabilities ### 1. 🔌 Modular SRE Skills Adapter The framework leverages the modular ADK Skills architecture to dynamically import, parse, and execute playbooks directly from the SRE extension repository: - **Automatic Discovery:** The `SkillAdapter` scans configured locations to resolve skill folders (e.g. `anomaly-detection`, `cloud-logging`, `safe-sre-investigator`). - **YAML Frontmatter Parsing:** Loads the `SKILL.md` from the target folder, parsing metadata (name, description, version) and cleanly separating it from the execution instructions. - **Dynamic ADK Hydration:** Appends pre-authored playbook instructions directly to the `LlmAgent` instructions at runtime, instantly extending the agent's cognitive capabilities. - **Configurability:** Point to your cloned local repo of `https://github.com/gemini-cli-extensions/sre` via environment variables: ```bash export GEMINI_CLI_SRE_DIR="/home/riccardo/git/sre" # or export SRE_EXTENSION_DIR="/home/riccardo/git/sre" ``` ### 2. 📢 Dynamic Communications Lead Renaming By default, the Communications Lead spokesperson is named **Madhavi**. To match organizational alignments or change names, the spokesperson's name is dynamically loaded from environment variables: - **Configuration variables:** `COMMS_LEAD_NAME` or `COMMUNICATIONS_LEAD_NAME` (fallback `MADHAVI_NAME`). - **Prompt Synchronization:** When renamed (e.g., to `Lucia`), the agent's internal system instruction prompt automatically synchronizes, replacing all `"Madhavi"` references with `"Lucia"`. ```bash export COMMS_LEAD_NAME="Lucia" ``` ### 3. 💬 Interactive ADK Agent Chat / Query You can interact, prompt, and query specific ADK agents directly via the terminal evaluation harness or programmatically: - **Terminal CLI Chat Harness (`src/cli.py`):** Use `--agent` and `--message` (or pipe stdin) to talk to any specific agent persona: ```bash # Talk to the Communications Lead (Lucia/Madhavi) uv run python3 src/cli.py --agent comms --message "Hello, who are you and what is your status?" # Query the Operations Lead uv run python3 src/cli.py --agent ops --message "How do you triage metric latency?" # Talk to the Incident Commander (Benjamin) uv run python3 src/cli.py --agent commander --message "status" ``` - **Programmatic `.run()` Method:** Every lead exposes a `.run(prompt)` method that executes the underlying `google.adk.agents.LlmAgent` (or safe fallback handler), enabling seamless inter-agent communications and manual engineering prompts. ### 4. 📢 Interactive SRE Telegram Bot & Safety Gates Integrates a fully interactive SRE terminal console directly via Telegram (`@SREBenjaminBot`): - **Structured Keyboard Menus**: Triage active incident status, projects, and incidents list instantly using mobile-friendly navigation menus. - **Interactive Inline Selectors**: Features 4 core context-switching primitives (`/incidents`, `/projects`, `/incident <id>`, `/project <id>`) with dynamic inline keyboard buttons displaying incident status emojis (`🟢` active vs `⚪` resolved). - **Gemini STT Voice Commands**: Send audio or voice notes to auto-transcribe and dispatch complex SRE commands on-the-fly. - **Human-in-the-Loop Safety Gate Approvals**: Safeguard mutations with interactive validation confirmation dialogs and Telegram reply options. - **Structured Pending Mutation Queue**: Supports queuing multiple proposed mutations with safety risk level indicators (🟢, 🟡, 🟠, 🔴), risk reasoning, and action justifications. Operators can selectively approve/reject queue items via API, and provide feedback comments that are dynamically piggybacked into the Incident Commander prompt context. - **Git Version-Controlled Chronicles**: Scribe Agent automatically commits all SRE chat interactions, timelines, and metadata states to git history, appending unique Git Notes to incident check-ins. ### 5. 🌐 Premium Project Explorer Workspace Provides a high-fidelity visual workspace for auditing discovered GCP project assets: - **Sticky Inline Navigation**: Navigation tabs (`Discovery Audit`, `Project Wiki`, `Logical Graph`, `Network Graph`) are integrated directly inline inside the sticky top header for seamless view switching. - **Toggleable Edit/View Modes**: Project Wiki notes and Graphviz DOT dependency configurations default to read-only rendering, with top-right `Edit` buttons to toggle side-by-side editing textareas. - **Interactive SVG Pan & Zoom**: Graphviz topology diagrams (Logical and VPC Physical networks) support scroll-wheel zooming and click-and-drag panning, complete with visual grab/grabbing cursor states. - **Markdown Tables Compiler**: Compiles tabular layouts (`|` pipes) in Project Wiki pages into beautiful glassmorphic, responsive HTML tables. ### 6. 📟 Collapsible SRE Chat Widget - **Pulsing Golden Pill State**: The floating SRE chat widget collapses into an ultra-compact, glowing golden pill badge featuring the text `⚡ SRE Co-Pilot (Click to Expand)` to optimize workspace real estate, expanding instantly on click. - **Flexible Resizing**: Includes maximize button to expand viewport to 90vh height and 20-30% wider to easily scroll through long diagnostic lists. ### 7. 🐳 Production Dockerization & Google Cloud Run Deploy - **Production-Ready Dockerfile**: Packages the SRE dashboard using `python:3.12-slim` base image, running as a non-root `appuser`. - **Serverless Cloud Run Pipeline**: Added the `just deploy` pipeline target running `bin/deploy-to-cloudrun.sh` to package, push to Artifact Registry, and launch on Cloud Run with IAP authentication. ### 8. 📊 OpenTelemetry (OTEL) Observability - **Trace Instrumentations**: Dynamic agent execution wrapping in `src/observability.py` logs full audit traces of inter-agent messages and tool execution paths to Cloud Trace or OTLP backends. --- ## 🧪 TDD Validation (Behavior-Driven Development) All features are implemented following a strict **Test-Driven Development (TDD)** and **Behavior-Driven Development (BDD)** workflow. Full specifications can be reviewed in [doc/bdd.md](file:///home/riccardo/git/adk-sre-benjamin/doc/bdd.md). The automated test deck consists of **93 high-fidelity unit and integration tests** verifying 100% of the dynamic loading, renaming, prompt loading, telemetry servers, and safety risk gate behaviors. ### Run tests with `uv`: ```bash uv run pytest ``` ### Run tests with coverage: ```bash # Using justfile just test # or just coverage ``` --- ## 📊 Agent Quality Flywheel & Evaluation Suite Project Benjamin implements the **5-Stage Agent Quality Flywheel** (Prepare Data → Run Inference → Grade → Analyze Failures → Optimize & Iterate) as described in the [Google Developers Blog](https://developers.googleblog.com/driving-the-agent-quality-flywheel-from-your-coding-agent/). For complete details on evaluation architecture, single-agent isolation benchmarks, and E2E incident scenarios, see: 👉 [Agent Quality Flywheel & Evaluation Guide (doc/eval.md)](doc/eval.md) ### Run all agent evaluations: ```bash just eval ``` --- ## 🚀 Getting Started Ensure you have [uv](https://github.com/astral-sh/uv) installed, then execute the following tasks: ### 1. Install Dependencies ```bash just install ``` ### 2. Verify Your Configuration and Environment Verify that your dependencies, `.env` file, credentials, and active gcloud profiles are set up correctly: ```bash just check ``` ### 3. Launch the Web Dashboard ```bash just web ``` Open your browser and navigate to: [http://localhost:10042/](http://localhost:10042/) ### 4. Run E2E Incident Simulation Click the **Trigger Live Simulation** button in the dashboard, or run it directly via the terminal: ```bash just simulate ``` --- ## 🏰 IMAG ICS Delegation Workflow 1. **Incident Declaration:** Incident Commander **Benjamin** detects alert SLO thresholds and activates the incident state. 2. **Channel Broadcast:** **Comms Lead (Lucia/Madhavi)** alerts engineering Telegram channels and creates GitHub issue tickets. 3. **Logistics Audit:** **Logistics Lead** scans environment variables, billing quotas, and evaluates proposed command risks. 4. **Operations Triage:** **Operations Lead** executes diagnostic metrics scripts, aggregates logs, and executes whitelisted system mutations once approved by Logistics. 5. **Memory Chronicle:** **Planning Lead** archives incident timelines, fetches historical runbooks via **Memento**, and commits scribe audit files using safe Git Annotations (`git notes`).

リポジトリの README を読み込んでいます…

このリポジトリの skills

ソフトウェア品質保証アナリスト・テスター

1 件の職業カテゴリ · 100% 分類済み

skill

職業分類

説明

更新

agent-platform-eval-flywheel

ソフトウェア品質保証アナリスト・テスター

Framework-agnostic Python SDK for Gemini Enterprise Agent Platform Evaluation API flywheel loops.

原文の言語: 英語

2026-08-25

google-agents-cli-eval

ソフトウェア品質保証アナリスト・テスター

5-stage Agent Quality Flywheel evaluation skill for ADK agents using the agents-cli toolchain.

原文の言語: 英語

2026-08-25

収集済み skill 2 件中 2 件を表示しています。