Thinking simulator action
🧠 A Knowledge action · 🔭 Discovery (#165). Challenges the target note with structured critical-thinking prompts — questions that push you to stress-test an idea instead of letting it ossify. The prompts adapt to the note's actual gaps in the knowledge model (#153). Relation- and signal-based, no text or AI inference. Deterministic, offline, and fully usable with AI disabled — AI can enrich the answers later, but is never required.
The prompts
Two kinds, always returned in a fixed order.
Universal prompts (always shown)
The four idea-agnostic challenges apply to any claim, so the simulator is never empty — even a brand-new or unindexed note gets these:
| Prompt | Purpose |
|---|---|
| What if this is false? | force the falsification frame |
| Who would disagree, and why? | surface the opposing view |
| What happens in the extreme case? | probe the boundaries |
| What hidden assumption does this rest on? | expose the unstated premise |
Gap-adaptive prompts (fired by the model)
Appended after the universals, only when the note actually shows that weakness:
| Prompt | When it fires |
|---|---|
| What evidence is missing to support this? | the note makes a claim but cites no source (#148) |
| What contradicts this? | nothing contradicts it — no contradicts relation in or out (#147) |
| What is a concrete example? | the note declares no outgoing example relation (#147) |
| How does this connect to what you already know? | nobody builds on it — it has no incoming edge |
| What open question does this raise? | the note raises no question relation (#147) |
The gap-adaptive set is the deliberate reframing of the suggest next move signals: that action gives a to-do ("add a source"); this one gives a challenge ("what evidence is missing — and what if it's false?").
Options
- Result property — where the prompts are written (default
thinkingPrompts). - Write to —
FrontmatterorContext(also exposed as{{property}}). - Target note (optional) — the note to challenge; empty ⇒ the note being built.
Result
An array of short, sentence-case questions (always ≥ 4), plus a Notice with the count. An
unknown/unindexed target still yields the four universal prompts — the simulator never returns an
empty list. If the index isn't ready, the action safely no-ops.
Capabilities
File-system read + the disclosed result-property write. No network, no AI.