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

How to run an agent loop until your tests pass with Graph Code

Create a goal-based loop in Graph Code with a shell predicate, such as make test, so the agent keeps working until the check exits 0.

Short answer: Install Graph Code with brew install --cask scgopi/graphcode/graphcode and add your project; then create a loop and choose the Goal-based type; write the task, for example fix the build. All steps.

Who it is forA developer with a failing build or test suite who wants the agent to keep going without being re-prompted each time.

The problemAgents often stop early and declare victory, so you check, find failures and prompt them again by hand.

The outcomeA loop that only finishes when your own done check passes, with the live session open to attach to if it goes off track.

What you need first

  • macOS
  • At least one CLI coding agent installed

Steps

  1. Install Graph Code with brew install --cask scgopi/graphcode/graphcode and add your project
  2. Create a loop and choose the Goal-based type
  3. Write the task, for example fix the build
  4. Set the done check to a shell predicate such as make test
  5. Press Test on the done check to run it exactly as the daemon will
  6. Start the loop; attach to its terminal if you want to correct it mid-run

Prompt for your agent

Create a GraphCode goal-based loop on <agent> for this repo with the prompt <fix the failing tests in module X without changing test files> and the done check <make test>. Tell me how to test the check before starting.

Caveat. Pick a done check that can actually succeed, or the loop keeps running.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for scgopi/GraphCode and "Run an agent loop until your tests pass"

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