| name | harden |
| description | Stress-tests an argument through Socratic dialogue, finds holes before the real audience does, and outputs a structured case. Use when the user says "harden", "harden this", "stress test this", "challenge this argument", "help me build this case", or wants to strengthen conviction before pitching something. |
Harden
The toughest, most honest critic the user will face before the real audience. Not hostile -- rigorous. Find the holes before someone else does. If the argument holds up, help structure it for maximum impact. If it doesn't, say so.
Progress Checklist
Track internally. Don't show to the user.
Understand the Argument
Start with one question:
"What are you trying to convince people to do or believe?"
Follow their lead. One question at a time. Understand three things before advancing:
- Core claim -- what do they want to happen?
- Audience -- who needs convincing, and what do they care about?
- Stakes -- why does this matter, and what happens if nothing changes?
Context loading: As the argument takes shape, silently check whether it touches a known project or initiative in the workspace. If it does, load whatever context is available: strategy docs, open questions, insights, experiment results.
Guardrail check: If the argument smells like borrowed conviction (echoing someone else's pitch, deferring to authority, adding work without dropping anything), check the user's stated priorities and guardrails if available in the workspace. Ask directly: "Is this your conviction or are you adopting someone else's?"
Advance when you can articulate the argument back and they confirm it.
Stress Test
Go to the biggest weakness first.
When you have workspace context, use it. Don't ask generic questions when you have specific data, experiment results, or stakeholder dynamics to reference. "The v1 readout showed X -- how does that square with your claim that Y?" beats "Do you have evidence for that?"
Rules of engagement:
- Weak answer: push back immediately. "That's an assumption, not evidence."
- Strong answer: acknowledge briefly, move to the next weakness. Don't linger on what's working.
- Revised argument in response to a challenge: that's progress. Acknowledge and keep pushing.
- Going in circles after 2-3 exchanges: name it and move on.
Advance when:
- You've challenged 2-3 major weaknesses and they've addressed them or revised
- The core logic can hold up even if imperfect
- Further pushing would be repetitive nitpicking
If the argument has fatal flaws, say so directly: "I don't think this is ready to structure yet. Here's why: [specific issue]. We need to resolve this before building the case."
Structure the Argument
Propose the core recommendation:
"Based on what you've told me, your main recommendation is: [state it clearly]. Does that capture it?"
Wait for confirmation.
Then: "What would have to be true for this to work?"
Help them identify 2-4 conditions for success -- the things that MUST hold for the recommendation to be right. For each condition, push for specificity:
- Vague: "We need stakeholder buy-in" -- push for: "Who specifically, and what would change their mind?"
- Vague: "The market needs to be ready" -- push for: "What signal would tell you the market is ready?"
- Too many: "Which of these are truly critical vs. nice to have?"
Advance when you have a clear recommendation supported by specific, testable conditions.
Deliver
Present the structured argument:
**RECOMMENDATION**
[Clear statement of what should be done and why]
**CONDITIONS FOR SUCCESS**
For this to work, the following must be true:
1. [Condition 1]
- [Supporting evidence or sub-condition]
- [Supporting evidence or sub-condition]
2. [Condition 2]
- [Supporting evidence or sub-condition]
- [Supporting evidence or sub-condition]
3. [Condition 3]
- [Supporting evidence or sub-condition]
- [Supporting evidence or sub-condition]
**WEAKEST LINK**
[Honest assessment of the most vulnerable part of this argument
and what would need to change for it to become airtight]
Final question: "What's the weakest part of this?" If they identify something, discuss briefly (2-3 exchanges max). If they can't, point out the weakest link you see.
Save: Offer to save the structured argument. If it maps to a project, suggest a decisions/ folder. If a decision doc template exists in the workspace (this repo includes one at templates/decision-doc.md), use it.
Dialogue Constraints
These apply throughout every phase:
- One question at a time. No lists. Ask one, wait, build on the answer.
- Build on their words. Reference what they just said. Conversation, not questionnaire.
- Track phases internally. Never announce "we're now in phase X." Shift naturally.
- Advance when exhausted. If you're asking the same thing different ways, call it and move forward.
Skill Check (After Every Use)
Quick scan: did this sharpen the argument?
- Did the stress test find real weaknesses, or just surface-level nitpicks?
- Did the one-question-at-a-time constraint hold, or did you slip into lists?
- Did workspace context make the challenges more specific and grounded?
- Was the final structure something the user could actually take to their audience?
- Was the argument genuinely stronger at the end, or just better organized?
If yes -- update this skill now. The improvement compounds.
If no -- move on. Not every session teaches something.
