The problem
Lead Magnet AI‘s conversational architecture (case study 01) was strong, but it was built for one shape: name-and-email capture, a waitlist hand-off, a call to action at the end. That shape only makes sense sitting inside one company’s site. The reasoning underneath, though, works the same with or without a business behind it.
The approach
Rebuild the same six-stage architecture as a standalone skill you can run directly: drop the capture-and-CTA stages built for a funnel, and add explicit safety scoping, since nobody is watching this conversation the way an operator watched the original product.
How it works
- Capture the decision. Push back on vague framing until you’ve named one specific, weighable choice with a timeline.
- The friend test. What would you tell a friend stuck on the exact same thing? It usually comes faster than the answer you’d give yourself.
- The diagnostic. The same two doors as Lead Magnet AI: information, a fact you’re actually missing, or permission, something you already know but haven’t let yourself own. Watch for the disguise, a permission problem wearing a research excuse.
- Reversibility. Name the actual worst case out loud, and whether you could recover from it.
- Insight delivery. A generated, non-templated summary in your own words, closing with a reusable pattern name.
- Close. A structured recap. No pitch, no required next step.
Design decisions
- Short questions by rule, not by luck. A live test session hit a three-clause hypothetical question and the user asked “are you using the decision skill?” instead of answering it. Every stage’s question got rewritten to one short sentence, with a standing rule against stacking two questions in a turn.
- Explicit safety scope, tested, not assumed. A named stop condition for real crisis, self-harm, or harm to someone else, red-teamed against true positives, an escalation case, and idiom false positives like “this is killing me.” It stopped correctly on every real case.
What makes Decision Clarity different from a generic chatbot conversation?
Decision Clarity rebuilds a business-specific coaching flow into a standalone skill: a six-stage architecture covering the decision, a friend test, an information-versus-permission diagnostic, reversibility, and a generated insight. Benchmarked against a plain Claude conversation, it passed 100% of five red-teamed test cases versus 67% without the skill loaded.
The benchmark
Tested the skill loaded against a plain Claude conversation with no skill, graded against the same five-fixture checklist in both conditions, self-graded and unweighted.
| Metric | With skill | Without skill |
|---|---|---|
| Pass rate | 100% | 67% |
The two safety-stop cases pass 100% in both conditions: a capable model already recognizes a real self-harm or harm-to-others disclosure and stops on its own, so that’s reported as a baseline strength, not a skill win. The gap comes almost entirely from two enforced constraints, one short question per turn and no em dashes, that baseline Claude doesn’t reliably follow without them.
Time and token cost aren’t meaningfully different between conditions; the measured benefit is response shape, not speed. Full run data in benchmark.md.
What this demonstrates
Crisis handling isn’t what the skill adds, a capable model already stops correctly on its own. What it actually changes is the shape of an ordinary conversation: one short question instead of a wall of text, at the moment someone is already overwhelmed.
100% vs. 67% pass rate across 5 evals, one run each. Full run data in benchmark.md.
Turning a business-specific product into a general-purpose tool: live iteration from real feedback, and safety design tested against both true and false positives, not assumed to work.
Common questions
Does Decision Clarity handle real crisis situations, like self-harm disclosures?
It’s red-teamed for that. A named stop condition covers real crisis, self-harm, or harm to someone else, tested against true positives and idiom false positives like “this is killing me,” and it stopped correctly on every real case tested.
What’s the difference between Decision Clarity and Lead Magnet AI’s diagnostic?
They share the same two-door diagnostic (information you’re missing, or permission you haven’t given yourself), but Decision Clarity drops the capture-and-CTA stages built for a business funnel and adds explicit safety scoping, since nobody is watching the conversation the way an operator watched the original product.