DEDUPED Hot AI repos, turned into use cases Get the Monday 5

How to build a code-first agent loop with agent_kit with Goal Flow

Use the vendored agent_kit SDK to create a ReAct or custom agent and run it inside a graph node.

Short answer: Run git clone on the goalflow repository and cd goalflow; then run pip install -e; read docs/agent-kit.md and choose the ReAct, Deep, or custom loop you need. All steps.

Who it is forA Python developer who needs an agent that calls tools, routes between models, and uses skills, without a visual editor.

The problemPlain workflow graphs cannot express open-ended tool-using loops, and building one from scratch takes time.

The outcomeA working agent defined in code, with middleware and skills, that can be hosted by a graph node in the GoalFlow engine.

What you need first

  • Python 3.12
  • Redis (cluster or standalone) and MySQL
  • A cloned copy of the goalflow repository and a filled-in .env file copied from .env.example
  • A Dify DSL export (.yml) if you want to transpile a visual flow

Steps

  1. Run git clone on the goalflow repository and cd goalflow.
  2. Run pip install -e ..
  3. Read docs/agent-kit.md and choose the ReAct, Deep, or custom loop you need.
  4. Import the Agent class from the agent_kit package under src/agent_kit/ and define your agent with the middleware and skills you want. (inferred, check the README)
  5. Attach the agent to a graph node that uses agent_base.py in src/goalflow/node/.

Prompt for your agent

In the goalflow repo, create a ReAct agent with agent_kit under src/agent_kit/ that answers questions about <topic> using the skills in skills/. Save the code to <output-path>.py. Verify that the agent imports cleanly and report which middleware and model routing options you configured.

Caveat. The README names agent_kit and its loop types but gives no code sample, so check docs/agent-kit.md for the real class and method names before writing code.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for wanmol/goal-flow and "Build a code-first agent loop with agent_kit"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer