djangosdk

Scaffolds a Pydantic v2 output schema and attaches it to an Agent for structured JSON output. Covers provider-specific paths (OpenAI json_schema, Anthropic tool-use, Gemini response_schema, fallback parsing). Invoke when the user says "create a structured output schema", "make my agent return JSON", "define a Pydantic output model", "get structured data from the agent", "add output schema", or "validate agent output".

Diagnoses failing or misbehaving django-ai-sdk agents using a structured debug checklist and extended thinking. Invoke when the user says "my agent is not working", "debug my agent", "agent returns empty response", "tool is not being called", "agent loop not terminating", "agent is ignoring the system prompt", "why is my agent failing", or "trace an agent run".

Scaffolds calls to the embeddings, audio (transcription/synthesis), and image generation modules. Invoke when the user says "generate an image", "transcribe audio", "generate speech", "use the images module", "add image generation to my agent", "process audio with the SDK", or "embed a document".

Migrates existing AI code (raw OpenAI/Anthropic SDK calls, LangChain chains, LlamaIndex pipelines) to django-ai-sdk patterns. Produces side-by-side before/after code with a migration checklist. Invoke when the user says "migrate from OpenAI SDK", "migrate from LangChain", "convert my existing agent", "replace openai.ChatCompletion", or "port to django-ai-sdk".

Scaffolds EpisodicMemory (DB-backed ordered recall) or SemanticMemory (pgvector embedding search) backends and attaches them to an Agent. Invoke when the user says "add memory to my agent", "set up episodic memory", "set up semantic memory", "give my agent long-term memory", "use BaseMemory", "store and retrieve memories", or "make my agent remember things".

Scaffolds multi-agent orchestration patterns (pipeline, parallel, handoff/router) from the djangosdk.orchestration module. Invoke when the user says "create a multi-agent pipeline", "set up agent orchestration", "build a supervisor agent", "create an evaluator agent", "chain agents together", "fan out to parallel agents", or "build a router agent".

Scaffolds an MCP (Model Context Protocol) server and client using djangosdk.mcp. Invoke when the user says "set up MCP", "create an MCP server", "connect to an MCP client", "expose tools via MCP", "use MCP decorators", "integrate MCP", or "add MCP support".

Configures observability backends (LangSmith, Langfuse, OpenTelemetry) by generating the correct AI_SDK settings block, environment variables, and Django signal receivers. Invoke when the user says "add observability", "set up LangSmith", "configure Langfuse", "add OpenTelemetry tracing", "how do I trace my agents", "enable tracing", or "monitor my agent".

Configures token-based rate limiting and API cost analytics using djangosdk.ratelimit and djangosdk.analytics. Invoke when the user says "add rate limiting", "set up cost tracking", "track token usage", "add analytics", "limit API spend", "configure budget limits", or "track per-user costs".

Scaffolds the full DRF integration to expose an Agent over HTTP using ChatAPIView (sync) or StreamingChatAPIView (SSE streaming). Invoke when the user says "wire up a chat endpoint", "expose my agent over HTTP", "add a chat API view", "set up ChatAPIView", "add a streaming endpoint", "hook up the DRF view", or "create a Django view for my agent".

Adds a new AI provider configuration to the django-ai-sdk settings. Invoke when the user says "add a provider", "configure Anthropic", "configure OpenAI", "configure Gemini", "add Ollama", "add DeepSeek", "add xAI", "add Grok", "add Groq", "add Mistral", "add Azure OpenAI", "add Perplexity", "add OpenRouter", "set up a new provider", "how do I add [provider name]", or "configure the [provider] provider in settings". Also triggers on "add provider config".

Creates a new Agent subclass with correct mixin composition. Invoke when the user says "create an agent", "make a new agent class", "write an agent for [task]", "add a [name]Agent", "build a [tool-calling / conversational / reasoning] agent", or "I need an agent that does [X]". Also triggers on "implement an agent" or "scaffold a new Agent subclass".

Creates a @tool decorated function with proper type hints and docstring so the JSON schema is generated automatically. Invoke when the user says "create a tool", "add a tool function", "write a @tool for [action]", "make a tool that [does X]", "define a tool called [name]", "I need a tool for [task]", or "implement a tool-calling function". Also triggers on "add a tool to the agent".

Implements a django-ai-sdk module from the PRD spec. Invoke when the user says "implement the agents/base.py module", "write the streaming/sse.py file", "build the tools/decorator.py", "implement the providers/litellm_provider.py", or any phrase like "implement [module path] from the spec" or "scaffold [module] per the PRD". Also triggers on "write the [module] module", "create the [file path] file for the SDK".

Writes pytest tests for django-ai-sdk modules using FakeProvider and the SDK's assertion helpers. Invoke when the user says "write tests for", "add tests for", "test this agent", "test this module", "write unit tests", "write integration tests for the agent", "assert that the agent calls the tool", "verify the prompt is sent", or "add test coverage for [module]". Also triggers on "how do I test this agent" or "how do I test tool calling".

Reviews an implemented module or agent against the PRD spec to identify gaps, deviations, or missing requirements. Invoke when the user says "review this against the PRD", "check this implementation against the spec", "does this match the PRD", "audit this module", "is this compliant with the spec", "what am I missing from the spec", "review my implementation", "PRD compliance check", or "spec review". Also triggers on "does this match CLAUDE.md".