Problem
My AI agents contradict each other — how do I fix it?
Your agents contradict each other because each one builds in isolation, blind to what the others already decided. CatWrangler fixes it two ways: agents read the relevant decisions before they touch code, and incompatible choices get surfaced and negotiated at intent time — before any code exists.
Contradictions aren't a discipline problem you can fix with better prompts — they happen because two agents make conflicting choices without ever seeing each other's reasoning. CatWrangler removes the blindness instead of asking you to police it.
Why your agents drift apart
Every change in CatWrangler records the decision behind it — the choice, the reasoning, the evidence, and the conditions that would change its mind — bound to the exact code it governs. Before an agent touches anything, it reads the relevant decisions. So it builds with the prior reasoning in hand, not against it.
When two agents are about to make incompatible choices, that conflict surfaces at intent time — before either has written a line — and gets negotiated then, not discovered later as two contradictory features sitting in the same codebase.
- Decisions read first: agents see the choices already made and the reasoning behind them before they build.
- Intent-time negotiation: incompatible plans between two agents are surfaced and resolved before any code exists.
- Higher-tier rules hold: foundational decisions sit at a higher tier and can't be casually overridden by a quick fix.
- Work is never silently overwritten — and two agents in different parts of the same file don't block each other.
Foundational rules don't get steamrolled
Most contradictions come from a fast fix quietly undoing a deliberate, load-bearing choice. CatWrangler ranks decisions by tier: the foundational ones can't be casually overridden by a one-off change. An agent that wants to go against them has to reckon with the reasoning first, not route around it. Two years later, you can still see why the code is the way it is.
“My script would have been strictly worse — it skipped steps I didn’t know existed. The conflict detector caught a real design mistake.”
Related questions
Doesn't this just slow my agents down?
No — agents get the relevant decisions handed to them, not a repo to crawl. Reading the reasoning before building is faster than discovering a contradiction after both versions ship and one has to be unwound.
What if an agent genuinely needs to override an earlier decision?
It can — but a foundational, higher-tier decision can't be casually overridden by a quick fix. The agent has to engage with the original claim, reasoning, and the conditions that would change its mind, so the override is deliberate, not accidental.
Keep reading
Vibe-Engineering
Many agents. One codebase. Zero collisions.
Point your agents at CatWrangler and build — the discipline runs underneath.