This directory is the Java SDK's Maven reactor. Paths and commands below are relative to this directory, whether it is nested at src/sdk/java or exported as java. Keep API usage examples in README.md, the SDK documentation, and public Javadoc rather than duplicating them in these contributor instructions. - sdk/src/main/java/ contains the handwritten client, session, and RPC APIs. sdk/src/test/java/ contains unit and integration tests. - sdk/src/generated/java/ contains generated protocol and event types. - copilot-native/ packages platform-specific runtime artifacts separately from the…
This directory contains the independent github-copilot-sdk crate. In standalone SDK workflows, paths and Cargo commands below are relative to this directory, and this crate's own lockfile and toolchain apply. When nested at src/sdk/rust, the runtime CLI consumes the same checked-out sources through the Bazel integration described under Development. The runtime's N-API architecture does not otherwise govern this crate. - Use the existing Error and ErrorKind in src/errors.rs. Extend that public error contract instead of adding parallel per-module error types or…
[!NOTE] The modern (as of 2026-06) AI is a misnomer, - it is not conscious / not a consciousness, - it is not sentient, - it is not intelligent, - it does not think, - it does not understand the code, - it is merely a next-token guesser, ... therefore it is merely an (hyper-) advanced IDE. [!NOTE] A contribution is any externally-observable interaction with a project. [!CAUTION] Failure to follow the following rules may result in repercussions, possibly without…
Show how to compose a SequentialAgent in ADK Kotlin: two LlmAgents that run in order, where the second consumes what the first produced. The domain task — fact-check an answer, then correct it — is a vehicle for three ADK concepts that are hard to demonstrate in isolation: 1. Sequential multi-agent composition with a shared model instance. 2. A built-in tool (GoogleSearchTool) grounding one sub-agent. 3. An AfterModelCallback post-processing a sub-agent's raw output before it reaches the user. This is…
A clone-and-study ambient agent: there is no interactive chat loop. Expense reports arrive as Pub/Sub events, get pushed to an ADK trigger endpoint, and flow through an ADK 2.0 graph-based Workflow. Business rules stay in code (a $100 threshold routes low-value expenses straight to auto-approval); only high-value expenses reach an LLM review_agent, which then pauses for human-in-the-loop approval via RequestInput before logging a decision. The graph itself is compact — the interesting parts are the ambient plumbing: authenticated Pub/Sub push,…
A clone-and-study ADK agent that remembers user preferences and facts across sessions using Vertex AI Memory Bank. The agent itself is a deliberately mundane demo (weather/time tools); the interesting part is the Memory Bank wiring — how memories are written after each turn (generatememoriescallback → addsessionto_memory()) and recalled at the start of a later session (PreloadMemoryTool), a single shared memorybankconfig that declares which topics to extract, and two deploy paths (Agent Engine and Cloud Run) that both enable Memory Bank…
A clone-and-study fullstack research agent built on ADK + Gemini. Unlike the RAG recipes, the interesting part is the agent: app/agent.py builds a multi-agent graph that plans (with a Human-in-the-Loop approval step), runs an iterative search → critique → refine loop until a quality bar is met, then composes a report with inline citations back to web sources. It ships a React frontend over an ADK-powered FastAPI backend. There is no datastore, no ingestion pipeline, and no Terraform — all…
A clone-and-study full-stack, multi-agent ADK app that generates retail media — Virtual Try-On (VTO) images/videos, 360° product spins (Reference-to-Video / R2V), product fitting on a virtual model, and background swaps — by orchestrating Gemini and Veo 3.1 through multi-stage media pipelines. The interesting part is not the router agent (a thin ADK wrapper around one MCP toolset); it's the MCP tool server + workflow pipelines (framing → parallel Veo generation → automated validation/retry), the in-memory catalogue vector search, and the…
Self-improving long-horizon agent on Google ADK + Vertex AI. This file has two halves. Studying the recipe is the entry point: scan the recipe table, jump to the one pattern you want, and read the real function that implements it. Maintaining the code is the exhaustive view — conventions, callback order, state keys, env vars — for when you are changing the repo rather than lifting from it. The code is the documentation. Each pattern below points at the actual…
A clone-and-study ADK agent that reads a user's Google Drive files on their behalf, gated by an OAuth 2.0 user-consent flow. The agent is deployed to Agent Runtime (Vertex AI Agent Engine) and registered with Gemini Enterprise. The interesting part is not the agent (a thin wrapper around one Drive tool); it's the OAuth consent plumbing — the negotiate_creds() three-stage credential resolution that makes the same code work in both local ADK Web UI dev and production Gemini Enterprise, plus…
A clone-and-study RAG agent grounded on Agent Platform Search (Discovery Engine). Drop documents in a GCS bucket and a managed Data Connector ingests, chunks, embeds, and indexes them — the agent answers over that data store with almost no ingestion code. The interesting part is not the agent (a thin wrapper around one search tool); it's the Terraform + data-connector plumbing. - The user wants a RAG agent without building/maintaining an ingestion pipeline (the managed connector handles chunking + embeddings).…
A clone-and-study RAG agent grounded on Vertex AI Vector Search 2.0, with a Kubeflow Pipelines (KFP) pipeline that loads, chunks, and ingests documents into a Collection that auto-generates embeddings server-side. The hard/interesting parts are Terraform (Collection + auto-embedding config + IAM) and data handling (the ingestion pipeline); the agent itself is a thin ADK wrapper around a semantic-search tool. - The user wants explicit control over chunking/ingestion (a custom KFP pipeline) instead of a fully-managed connector. - The user needs…
A clone-and-study recipe for global safety guardrails implemented as ADK BasePlugins attached to the Runner. Two interchangeable plugins are provided — an LLM-as-a-judge and a Model Armor filter — that hook the same lifecycle callbacks to classify and block unsafe content. The interesting part is not the agents (deliberately mundane sum/fibonacci demos); it's the plugins: because they're wired at the Runner they wrap every agent and sub-agent under it, and they defend against session poisoning by never persisting harmful content…
Guidance for AI agents (Claude Code, Copilot, Cursor, etc.) working in this repository. See CONTRIBUTING.md for the full contributor workflow. Run the pre-commit hook before committing (pre-commit run --all-files, or let it run on staged files via git commit). Fix any issues it reports so the commit lands clean — CI runs the same checks. Use just for common tasks; run just --list for grouped recipes. When you open a pull request, fill in the repo's PR template at .github/pullrequesttemplate.md…
This project uses Declarative Automation Bundles for deployment. Install the Databricks CLI 0.292.0 or newer and verify with databricks -v. Read the databricks-core skill for CLI, authentication, and deployment workflow. Read the databricks-jobs skill for job-specific guidance.
Agent-facing rules for working on Claurst. Mirrors and extends src-rust/.claude/CLAUDE.md; when the two disagree, the rule closer to the code wins. Run from src-rust/ unless noted. The ratatui frontend is…
Claurst has a named-agent system that lets you select a pre-configured persona with its own tool permissions, model, system prompt, and turn budget. For larger tasks it also supports a…
This is the official MongoDB Node.js driver (mongodb npm package). It provides a TypeScript/JavaScript interface for applications to interact with MongoDB deployments. The driver implements the cross-driver MongoDB specifications. Do not hand-edit: lib/ (build output), mongodb.d.ts (generated), HISTORY.md (release-please managed), test/spec/ (vendored from specifications repo). Integration tests require a running MongoDB instance (unit tests do not). To start one locally: Tests use Mocha with 60-second timeout. Integration tests use a custom metadata UI that supports test filtering by topology, MongoDB…
This project is building a platform for "vibe coded" personal applications and AI agents that run inside a strong sandbox. The following files are commonly important to reference: The project structure is:
GitHubAGENTS.md
