What I tried
README says @actantdb/mastra works with Mastra, LangGraph, OpenAI Agents SDK, or hand-rolled agents. Tried to build a hand-rolled one.
What I hit
- The synthetic
model_call event carries "model":"user-provided" + "role":"planner" for agents without their own model-call surface. A fresh user has no idea what those mean or how to override.
- For real Mastra usage, you need
npm install @mastra/core separately. For LangGraph: @langchain/langgraph. For OpenAI Agents: @openai/agents. None of this is in the README.
Fix
One short subsection per supported framework: how to install, what shape the agent must expose, what the wrapper does in each case. Could live in packages/actant-mastra/README.md or in the top-level README's "Install" section.
Bonus
If we're going to claim cross-framework support, add at least one integration test per framework to keep the claim honest.
What I tried
README says
@actantdb/mastraworks with Mastra, LangGraph, OpenAI Agents SDK, or hand-rolled agents. Tried to build a hand-rolled one.What I hit
model_callevent carries"model":"user-provided"+"role":"planner"for agents without their own model-call surface. A fresh user has no idea what those mean or how to override.npm install @mastra/coreseparately. For LangGraph:@langchain/langgraph. For OpenAI Agents:@openai/agents. None of this is in the README.Fix
One short subsection per supported framework: how to install, what shape the agent must expose, what the wrapper does in each case. Could live in
packages/actant-mastra/README.mdor in the top-level README's "Install" section.Bonus
If we're going to claim cross-framework support, add at least one integration test per framework to keep the claim honest.