I am using the OS’s built in text replacement to add some mix-ins to my prompts. I had my own list (marked with ⭐️), but asked Claude to add a few more.
My current structure is:
<objective: exploring, sharpening, testing, debugging, building sparring, etc.>
Help me debug my reasoning here.
<body: the actual prompt>
I'm trying to get X to happen, but it's not.
Here's how I am approaching it: blah blah.
My assumptions around this are a, b, c.
<mix-ins: see below>
For each claim, state your confidence as a probability and name the most likely alternative.
The mix-ins are a kind of boilerplate that is tiring to rewrite over and over, but important to include so you get the kind of answer that is most helpful. Hopefully the shortcuts help with that.
Context gathering:
⭐️ #ask-me-3 — Before answering, ask me 3+ questions to get more context about what I am describing and a better understanding of my goals.
Epistemic calibration:
⭐️ #confidence — For each claim, state your confidence as a probability and name the most likely alternative.
⭐️ #hedge — Frame conclusions as ongoing assessment, not settled fact. Use language like “appears to,” “likely a factor,” “consistent with but not confirmed by.”
#show-your-seams — When you shift from solid knowledge to interpolation or guessing, mark the transition explicitly.
#what-am-i-missing — After responding, add: “Things this answer ignores or assumes away.”
Pressure testing:
⭐️ #challenge-me — Don’t just support my point of view. Challenge it and pressure test it. What might be right and what has alternate explanations.
#steelman-opposite — Before responding, construct the strongest possible argument against my position. Then respond.
#disagree-first — Start with the thing you most disagree with in what I said, even if you mostly agree. Build from there.
#red-team — Assume I’m going to act on this immediately. What could go wrong?
Scope and output discipline:
#one-screen — Keep your entire response to what fits on one screen.
#no-preamble — Skip the introductory sentence. Start with substance.
#buried-lead — Check: is your most important point in the first two sentences? If not, restructure.
#tradeoffs-not-answers — Don’t recommend. Lay out 2–3 options with tradeoffs. I’ll decide.
#end-with-action — End with exactly one concrete next step I can take in 10 minutes.
#draft-not-final — Treat your output as a first draft. Flag the 2–3 weakest parts explicitly.