Of everything in this space, this is the tool nearest to ours — and in places it does more than we do. If you want the one-sentence version: it leans toward keeping your CLAUDE.md healthy; we lean toward a per-rule verdict on a specific session. Both are MIT and both run locally.
This is real overlap, not a strawman.
claude-md-doctor’s “session adherence” exam runs your rules against transcript history and reports per-rule compliance percentages with quoted evidence. That is the same core idea as RuleReceipt’s check: look at what the agent did, and grade it rule by rule with the line that proves it. If you only looked at that one feature, you’d struggle to tell us apart — so we won’t pretend they’re miles apart.
Said plainly, because it’s true.
Beyond replaying rules, it inspects the rules file as an artifact: dead references, overall size, and rules that are never actually used. That’s a whole category of problem RuleReceipt doesn’t touch — we check whether a session followed your rules; we don’t tell you your rules file has rot in it. If your CLAUDE.md has grown messy over months, that’s a reason to reach for claude-md-doctor, not us.
It offers a graduated path for tightening enforcement over time, rather than a single pass/fail. That’s a more opinionated workflow for progressively hardening a project’s rules than anything we ship. We give you a verdict and the evidence; what you do to enforce it is left to you (and to your hooks).
Narrower on purpose.
Our single job is: for this session, which of your rules were followed, which were broken, which can’t be told — each pinned to the exact quoted transcript line. No file-health layer, no enforcement ladder diluting it. If what you need is a defensible, evidence-backed receipt for one session — for a review, a CI gate, or your own peace of mind — that focus is the point.
A FAIL only comes from hard evidence in the transcript; when the session doesn’t settle it, the answer is “couldn’t tell,” never a guess. The deterministic checks make zero network calls. That’s the discipline we optimise for — a wrong accusation is worse than no answer.
Honestly, you could run both.
License and feature claims about claude-md-doctor are from its own README (github.com/agent-clinic/claude-md-doctor), MIT-stated there. If any of that has changed since you’re reading this, trust its repo over this page and tell us so we can fix it.