Guardrails for AI coding agents
Give an agent a narrow task and it often edits far beyond it. LLMtoMD records the approved boundary up front and catches anything outside it — checked against your stored spec — before it ever reaches a commit.
git diff tells you what changed — not what was allowed
AI coding agents are eager. Ask one to fix a single function and it may refactor neighbouring files, “tidy” unrelated code, or change behaviour you never approved — all in the name of being helpful.
The damage hides in a large diff. git diff shows every change, but not which ones were in scope for the task. Catching the unapproved edits by hand, on every commit, is exactly the review that gets skipped when you're moving fast.
And the agent forgets. Each session starts fresh with no memory of what it was — or wasn't — supposed to touch, or what the specification actually requires.
Record the boundary. Catch the drift. Stay in control.
Approve a boundary up front
Before editing, the agent records the files, folders, or globs it's allowed to change for this task — and the project it belongs to.
Spec-aware drift checks
Every changed file is checked against the boundary and your stored requirements, so a flagged edit tells you whether the spec actually justifies it.
Block it at commit time
A local pre-commit hook stops any commit that touches files outside the boundary — only file paths leave your machine, never your code.
You decide — nothing auto-deletes
Approve the wider scope in one click or revert. The task contract and every decision persist across sessions and machines.
Convert anything to AI-ready Markdown
PDFs, Office docs, images, audio, and whole websites — clean Markdown and RAG-ready exports for your LLM, in seconds.