Guides

When coding agents go off track.

Practical ways to diagnose unwanted edits, broken tests and repeated mistakes — using the tools you already have.

Claude Code / Codex / Cursor

Latest guides

A closer look at the failure, with examples you can reproduce.

A small fix should not become an unreviewed refactor. Define edit scope, inspect the diff and check the permissions that can enforce the boundary.

Passing AI-written tests can miss the real defect. Inspect mocks, protect the acceptance criteria and verify the user journey against the actual app.

Start with the symptom you recognise.

Why rules get skipped, and how to put the ones that matter in front of the edit.

Why a fix doesn’t stick between sessions, and how to attach the lesson to the file.

What to keep in your agent’s notes, what to cut, and what to stop writing by hand.

Understand the boundaries behind a reliable agent workflow.

Bring architectural boundaries into the run, then review the changed responsibilities.

The boundaries, evidence, and review that keep autonomous changes accountable.

Grant the smallest useful scope, and make every expansion a visible decision.

Carry known errors, guardrails, and required tests into the runs that follow.

Keep one project boundary across every MCP-capable executor you use.