What skills do you use in Claude/Codex when a decision looks straightforward at first and you know it isn’t?

I use a few different skills here, depending on the decision:

• Pre-mortem: Imagine the decision has gone badly, then work backward to what might have failed and what I can check beforehand.
• Gadfly: Press on the assumptions and strongest objections. What evidence would make me change my mind?
• Unknown unknowns: Look for people, conditions, or interpretations missing from how I’ve framed the problem, then find a small way to investigate them.
• Insight: Match roles and perspectives to the decision—a practitioner, a researcher, or the person affected—and see what each helps me notice.

@nurijanian highlights three of his skills:

• causal-tree: Trace possible causes backward and a decision’s consequences forward, making assumptions explicit.
• structure-problem: Work from principles down and evidence up, then compare where they agree or conflict.
• edge-cases: Find scenarios a specification misses and spell out what should happen.

(The descriptions are public; the skill files are part of his AI PM OS.)

Two replies offer small prompts I like. I’d call @hodisra’s a second-example check: “force a second concrete example the first phrasing would also cover.” And @KakaluoteW45042 offers a hidden-preconditions check: “list its hidden preconditions, then attack them one by one.”

This connects with some explorations I’ve been doing around the Meno problem: how do we search when we don’t yet know the answer, and how might we recognize something worth finding?