continuum

Get a Continuum agent up and running — Python 3.13 venv, infra via `continuum up`, smallest possible BaseAgent + AgentRunner example. Invoke when the user asks "how do I start", "set up Continuum", "run my first agent", or is at the very beginning of a project.

Set up Claude Code for a Continuum project — CLAUDE.md wiring, skill imports, project settings, and hooks. Invoke when the user asks "set up Claude Code for my project", "CLAUDE.md for Continuum", "add skills to my project", or "configure Claude Code hooks".

Set up OpenAI Codex CLI for a Continuum project — AGENTS.md content, project conventions, and what Codex needs to write correct Continuum code. Invoke when the user asks "set up Codex for my project", "AGENTS.md for Continuum", or "configure Codex CLI".

Set up Cursor for a Continuum project — .cursor/rules/ MDC file, glob patterns, and what to include so Cursor generates correct Continuum code. Invoke when the user asks "set up Cursor for my project", "Cursor rules for Continuum", ".cursor/rules", or "MDC file for Continuum".

Build BaseAgent instances and run them with AgentRunner — covers fields, lifecycle hooks, structured outputs, ReAct mode, instruction modifiers, and the full execution flow. Invoke when the user asks "create an agent", "configure max_turns", "add lifecycle hooks", "structured output with Pydantic", or anything around the core agent abstraction.

Pick the right LLM provider, configure structured outputs, control context-window compression, and use the LLMClient directly. Provider routing is by model-string prefix; LiteLLM has been removed. Also covers Smart Gateway integration for multi-provider routing. Invoke when the user asks about "switch to Claude", "Gemini structured output", "context length error", "rate limiting", "JSON schema", "Smart Gateway", "gateway_mode", or "how does the LLM client work".

Configure and use Continuum's two-tier memory system — mem0+Qdrant/Milvus for long-term facts, Redis for short-term sessions, with multi-tenant scopes (USER / AGENT / SHARED / RUN / CONVERSATION). Invoke when the user asks about "remember", "user preferences", "long-term memory", "vector search over memories", "multi-tenant isolation", "PII filtering on memory writes", "Milvus", "Qdrant", or memory-related errors.

Evaluate agent quality with the `EvaluatorAgent`, build golden datasets from Langfuse traces, and run DeepEval/RAGAS metrics over conversations. Invoke when the user asks "test agent quality", "evaluate output", "RAG metrics", "DeepEval", "RAGAS", or "regression-test my agent".

Build durable agent workflows with Temporal — sequential/parallel/loop/conditional steps, human-in-the-loop approval gates, custom workflows and activities. Invoke when the user asks "long-running workflow", "approval gate", "human in the loop", "retry on failure", "workflow survives restart", or anything Temporal-related.

Build agent-to-agent transitions with Continuum's `Handoff` system — triage routing, history summarization modes (FULL/SUMMARY/RECENT_N/HYBRID), cycle detection, depth tracking, return-to-parent. Invoke when the user asks "route customer requests to specialists", "agent that can transfer to another", "summarize history before handing off", or anything multi-agent that's a *transition* (not a *workflow*).

Trace agent runs with Langfuse, decorate functions with @observe, collect latency/token/error metrics, and report errors. Invoke when the user asks about "see what the LLM was prompted with", "Langfuse traces", "track latency", "metrics dashboard", "error reporting", or "instrument my function".

Copy-pasteable Continuum patterns — RAG, plan-and-execute, ReAct, multi-tenant agents, FastAPI integration, structured output, prompt-injection scanning, custom containers. Invoke when the user asks "how do I do X with Continuum" and X is a common app pattern rather than a single API question.

Stream tokens, tool calls, handoffs, and memory events out of a Continuum agent in real time using `runner.run_stream()` and the `EventType` enum. Invoke when the user asks "stream tokens to UI", "websocket chat", "live progress", "see tool execution as it happens", or anything that needs token-by-token output.

Write tests for Continuum agents — mock LLM and memory clients via the DI Container, use fakeredis for sessions, snapshot agent responses, and run pytest-asyncio. Invoke when the user asks "test my agent", "mock the LLM", "fakeredis", "container injection", "pytest", or wants their CI to validate agent behavior without burning real API tokens.

Connect MCP servers (Stdio/SSE/StreamableHTTP) to a Continuum agent, configure tool filtering, set up tool-context capture/injection (e.g. session_id), and read run artifacts (UI widgets, structured tool data). Invoke when the user asks "connect MCP", "filesystem tool", "remote API tool", "auto-capture session_id", "agent uses too many tools", or "expose widget data".

Use Continuum's nine workflow agents — Sequential, Parallel, Loop, Reflection, Router, Planner, Debate, Scatter, SupervisedSequential — to build multi-agent pipelines. Invoke when the user asks "chain agents", "run agents in parallel", "iterate until done", "self-improving agent", "route to specialist", "decompose a goal", or anything multi-agent.