Agent Skills for AG2 developers
This repository also configures its own agents. See what ag2-skills tells them →
Skill Claude CodeCodex
Expose an AG2 Agent over the Agent-to-Agent (A2A) protocol so any A2A-compliant client (or another AG2 agent) can call it across process/host boundaries. Wrap the agent with A2AServer(agent) and serve it via buildjsonrpc(...), buildrest(...), or buildgrpc(...) — each returns a ready-to-serve ASGI/gRPC object that…
Skill Claude CodeCodex
Drive external CLI coding agents (Claude Code, Codex, OpenCode) as first-class AG2 Agents over the Agent Client Protocol (ACP): point an Agent at a ClaudeCodeConfig / CodexConfig / OpenCodeConfig preset from ag2.acp and ask()/run() it like any agent, with its thinking, tool calls, plans and permission prompts…
Skill Claude CodeCodex
Add a custom Python tool to an AG2 Agent using the @tool decorator. Use when the user wants to give an Agent a new capability backed by Python code (API calls, DB queries, computations, file ops). Covers sync and async tools, parameter typing, Pydantic schema customisation, returning typed Input / ToolResult (text /…
Skill Claude CodeCodex
Expose an AG2 Agent over the AG-UI protocol so a frontend (CopilotKit, custom React/Next.js, or any AG-UI client) can stream responses (text and reasoning), render tool calls, sync shared state, and track sub-task steps. Wraps the agent with AGUIStream(agent) and mounts it in FastAPI via stream.dispatch(...) or…
Skill Claude CodeCodex
Compare AG2 agents, models, or prompts to decide which is better. runvariants scores several named agents on one suite and ranks them on a leaderboard (Variants holds a mapping of named Agent instances plus an axis label). runpairwise with pairwisejudge does head-to-head LLM comparison using a dual-order position swap…
Skill Claude CodeCodex
Evaluate, test, and track an AG2 Agent offline. Build a Suite of tasks, run the agent with runagent, and grade answers with prebuilt scorers (finalanswermatches, toolcalled, notoolerrors, tokenbudget) or a custom @scorer — including the agentjudge LLM judge. Read the RunResult scorecard (passrate, scorestats…
Skill Claude CodeCodex
Pause an AG2 Agent mid-run to collect human input via context.input(), or gate a tool call with approvalrequired() middleware. Use when the user wants the agent to ask for confirmation, request missing info (passwords, API keys, data), or have a human approve sensitive / irreversible / expensive tool calls (sending…
Skill Claude CodeCodex
Persist agent state across runs, shape what the LLM sees per turn, and cap history to fit a context window. Covers KnowledgeStore (memory / sqlite / disk / redis), KnowledgeConfig (store=, exposetool=, writeeventlog=, compact=, aggregate=, bootstrap=) and its opt-out flags, aggregation strategies…
Skill Claude CodeCodex
Build realtime voice / live audio agents with AG2's ag2.live module. Wrap a prompt + provider config in LiveAgent and open a bidirectional voice session with agent.run(), pumping mic audio in and playing synthesized speech out. Covers the two realtime providers — Gemini Live (GeminiRealTimeConfig) and OpenAI Realtime…
Skill Claude CodeCodex
Host an MCP server that exposes an AG2 Agent (plus prompts and resources) to MCP clients like Claude Desktop, Cursor, or the MCP Inspector. Wrap the agent with MCPServer(agent) — it surfaces Agent.ask() as a single conversational tool and serves over stdio (runstdio()) or streamable HTTP (it is itself an ASGI app for…
Skill Claude CodeCodex
Intercept the AG2 agent loop with BaseMiddleware — wrap full turns (onturn), each LLM call (onllmcall), each tool execution (ontoolexecution), or each human-input request (onhumaninput). Use for retry, logging, history trimming, request mutation, tool auditing, guardrails, or rate limiting. Built-ins…
Skill Claude CodeCodex
Send images, audio, video, or documents into an AG2 Agent alongside text. Pass ImageInput, AudioInput, VideoInput, or DocumentInput as positional args to agent.ask(...). Use when the user wants the agent to process non-text input — describe a photo, transcribe audio, summarise a PDF, analyse a video. Covers…
Skill Claude CodeCodex
Open an AG2 network discussion channel — N-party (2+) round-robin where each participant speaks in fixed order, cycling until explicit close or TTL. Use when the user wants a brainstorm with a fixed cast, a panel discussion, or round-robin reviewers. Covers agentclient.open(type="discussion", target=[...]…
Skill Claude CodeCodex
Govern an AG2 multi-agent network — identity (Passport, Resume), per-agent Rule with two top-level blocks access (AccessBlock) and limits (LimitsBlock, which nests RateBlock + InboxBlock), the swappable HubArbiter / RuleBasedArbiter access-&-routing seam, AuthAdapter / AuthRegistry registration, channel-level…
Skill Claude CodeCodex
Build a multi-agent AG2 network — load this whenever two or more Agents need to interact. The network is the standard multi-agent pattern in ag2. Covers Hub.open, LocalLink, HubClient.register, Passport, Resume, channel lifecycle (PENDING → ACTIVE → CLOSING → CLOSED), the two 2-party channel adapters (consulting for…
Skill Claude CodeCodex
Shape what an AG2 network agent perceives and which actions its LLM can take. Covers the five auto-injected LLM-facing tools that ship via NetworkPlugin (delegate, peers, channels, tasks, context) plus the adapter-owned say; replacing the default handler with agentclient.onenvelope(callback) (gateways, headless…
Skill Claude CodeCodex
Build an AG2 network workflow channel — the orchestrated N-party adapter driven by a declarative TransitionGraph. Use when the user needs conditional handoffs, multi-step pipelines, context-aware routing, feedback loops, or is migrating from the classic GroupChat + Agent.handoffs pattern. Covers TransitionGraph…
Skill Claude CodeCodex
Monitor an AG2 agent's stream — log events, detect repeated tool calls, track token spend, build trigger-driven observers, route observer alerts to the model, and halt on FATAL conditions. Covers @observer(...) (stateless), BaseObserver (stateful), built-ins (TokenMonitor, LoopDetector), Watch primitives (EventWatch…
Skill Claude CodeCodex
Map of AG2 capabilities and which sibling skill to reach for. Load first when the user mentions building with AG2 (ag2) but the specific feature isn't yet clear — agents, tools, model config, delegation, memory, observers, structured output, HITL, AG-UI, MCP server hosting, A2A protocol, realtime voice, telemetry…
Skill Claude CodeCodex
Build a minimal AG2 Agent end to end — pick a model provider, set a prompt, call agent.ask(), then continue the conversation with reply.ask() (multi-turn). Use when the user is starting a new AG2 project, has no working Agent yet, or needs the multi-turn chaining pattern. Covers OpenAIConfig, AnthropicConfig…
Skill Claude CodeCodex
Give an AG2 Agent the ability to run shell commands. Covers SandboxShellTool (client-side subprocess via LocalEnvironment, works with any provider) and the provider-native ShellTool (OpenAI Responses execution). Use when the user wants the Agent to execute commands, build/test code, manage files, or operate on a…
Skill Claude CodeCodex
Get a typed Python value back from an AG2 Agent instead of free text. Pass responseschema= (a Pydantic model, dataclass, primitive, union, ResponseSchema, or @responseschema validator) and read the parsed result via await reply.content(). Use when the user wants validated structured output, classification, extraction…
Skill Claude CodeCodex
Single-agent recursion and parallel fan-out within one AG2 Agent — auto-injected runsubtask / runsubtasks(parallel=True) (opt in via tasks=TaskConfig(...)) for self-delegation, and Agent.astool() as a lightweight no-hub way to call one named agent from inside another. Use when a single coordinator wants to break work…
Skill Claude CodeCodex
Add OpenTelemetry traces to an AG2 Agent via TelemetryMiddleware (ag2.middleware.builtin). Emits spans for the full turn, each LLM call, each tool execution, and each human-input request, following the OpenTelemetry GenAI semantic conventions. Compatible with any OTLP backend — Jaeger, Grafana Tempo, Datadog…
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: