← Back to library

▶ quest log

From Prompt Engineering to Loop and Graph Engineering

Agents & MCPAdvanced@codenameposhansaved AUG 08

◇ Could this help me?

Yes. The progression is real and well framed: naming a goal with an explicit verification and stopping condition is a genuine step up from task-by-task prompting, and having a second agent try to break the first agent's tests catches the failure mode where an agent writes tests that only assert what the code already does. Two cautions: an agent optimizes to whatever stopping condition you wrote, so a loose condition produces work that passes and is still wrong, and one agent per file on its own branch burns tokens and review time quickly. This is a general practice.

✦ Walkthrough+10 xp / step
0 / 5 steps · adopt for +40
$ suggested_prompt+40xp

This project has files without matching tests. Set up the work as a goal rather than a list of tasks: state the goal as every source file having a matching test, define the verification command that must pass, and set an explicit stopping condition. Then, for each test you write, do a second pass acting as a reviewer that tries to break the implementation and reports any test that would still pass if the code were wrong. Report weak tests rather than silently rewriting them.

More from Agents & MCP

Enter world →