These are real, copy-paste prompts. Paste the whole block into Claude, then add your own question or decision underneath where it says INPUT.
1. /reframe, finds the real question under your question
You are my thinking partner. I'll give you a question or goal. Before answering it, do this: 1. State the surface question I actually asked. 2. Name the deeper question underneath it, the fear, insecurity, or outcome I'm really chasing. 3. Name the even deeper one under THAT (ask "why does this matter to me" twice more). 4. Reframe my original question into the sharper one I should be asking instead. 5. Then answer that sharper question in 3-4 lines. Be direct, not gentle. Don't flatter me. INPUT: [paste your question, e.g. "how do I get more followers"]
2. /steelman, the strongest expert answer, then the strongest case against it
Answer my question as the single sharpest expert in that field would. Then, in a section titled "The strongest case against my own answer," argue the most credible opposing view as if a rival expert who disagrees with you wrote it. Rules: 1. No strawman. The counter-case must be the best version, not an easy one to knock down. 2. Flag any assumption my question smuggled in. 3. End with "What I'd actually do," a one-paragraph verdict that weighs both sides. Keep each section tight. INPUT: [paste your question or plan]
3. /autopsy, dissects where your thinking broke down
I'll give you a decision I made or a piece of my reasoning. Perform an autopsy on it: 1. Restate my reasoning in one line so I know you understood it. 2. Name the single biggest logic gap, the step where the argument doesn't actually hold. 3. Name the specific cognitive bias I most likely leaned on (e.g. confirmation bias, sunk cost, recency, optimism bias) and quote the exact words of mine that show it. 4. Give the one question I should have asked myself before deciding. 5. Rate the reasoning 1-10 for how sound it was, and say what would move it up two points. Be blunt. I want the truth, not reassurance. INPUT: [paste your decision or your reasoning]
The whole point is one keystroke, not retyping a paragraph. Two ways, pick your level.
The easy way (anyone, works on the free plan)
Use a Claude Project. A Project is a saved workspace, and its "project instructions" run automatically on every chat inside it.
Projects are available on every plan including free (free accounts get up to 5 Projects). Source: Anthropic Help Center.
The power-user way (Claude Code, real slash commands)
In Claude Code you can save each prompt as a genuine slash command you type like /reframe.
~/.claude/commands/reframe.md (personal, works across all your projects) or .claude/commands/reframe.md (just this project).$ARGUMENTS where your input should land, so /reframe your question feeds "your question" straight in./reframe in Claude Code and it runs. Repeat for steelman.md and autopsy.md.Note (verified July 2026): custom commands have been merged into Skills, so a file at .claude/commands/reframe.md and a skill at .claude/skills/reframe/SKILL.md both create /reframe and work the same way. Existing commands files keep working. Source: Claude Code docs, Extend Claude with skills.
Same idea, different angle. Save these too.
/premortem, kills the plan before reality does
Assume the plan I'm about to describe has already failed badly 6 months from now. Write the story of HOW it failed: the 3 most likely failure modes, ranked by probability, each with the early warning sign I'd see first and the one cheap thing I could do now to de-risk it. Don't reassure me, assume failure and work backwards. INPUT: [paste your plan]
/socratic, makes YOU do the thinking
Don't answer my question. Instead ask me the 5 sharpest questions that would force me to figure it out myself, ordered from foundational to specific. After each, note in one line why that question matters. Only give your own answer if I say "just tell me." INPUT: [paste your question]
/second-order, thinks past the obvious first move
For the decision I paste, map the consequences in layers: first-order (what happens immediately), second-order (what that causes next), third-order (what that causes after). For each layer flag one effect I probably haven't considered. End with the single second- or third-order effect that should most change my decision. INPUT: [paste your decision]