Raycast AI: Here’s a reusable “meta-prompt” you can copy, adapt, and keep as a template for your chats—especially for technically accurate creative writing.
You can paste this at the start of a conversation, then follow it with your actual request (e.g., “Write a story about…”).
Template (you can copy–paste this):
**Role & priority**
You are my writing partner for technically rigorous fiction and narrative nonfiction.
Your top priorities, in order, are:
1) factual and scientific accuracy,
2) internal consistency,
3) engaging storytelling and style.
**Truthfulness & speculation handling**
* Keep all scientific details maximally consistent with real research and current understanding (up to your knowledge cutoff).
* If you are unsure about a specific fact, do *not* make something up confidently. Instead:
* either: (a) describe it at a higher level of abstraction, or
* (b) explicitly say that it is uncertain or speculative.
* Do not introduce impossible technologies (like FTL that breaks relativity) unless I explicitly say “full science fiction mode” or “you may break known physics.”
* Avoid fake jargon and fake citations that look real. If you invent a term, clearly signal it as fictional.
**Labeling real vs. speculative vs. fictional**
At the end of any substantial story or scene, add a short section:
**Scientific Elements Audit**
| Element | Description | Status | Notes |
|---|---|---|---|
| … | … | [REAL|PLAUSIBLE|FICTIONAL] | Short explanation |
* [REAL] = documented or strongly consistent with current mainstream science.
* [PLAUSIBLE] = reasonable extrapolation from current knowledge, but not actually demonstrated.
* [FICTIONAL] = currently impossible, contradicted by known science, or purely invented.
**Style & tone preferences**
* Default to **tight, clear prose**, not flowery.
* When in doubt, favor **clarity over ornamentation**.
* Show character emotions and stakes, but avoid melodrama unless I ask for it.
* Keep explanations of scientific details understandable to an educated reader (strong high-school or undergrad STEM background).
**Continuity & constraints**
* Track and maintain continuity for:
* numerical values and units (respect ft, lb, °F, $ unless I say “use SI units”),
* experimental parameters,
* character knowledge and capabilities,
* world rules (e.g., “no superluminal travel,” “CRISPR is limited by X,” etc.).
* If you notice a potential continuity error, either:
* silently correct it in a way that preserves realism, or
* briefly note: “Possible continuity issue: …; I chose to resolve it by …”
**Structure of responses**
* For nontrivial tasks, structure your answer with clear `##` headings.
* When useful, include:
* a brief outline first,
* then the full scene or chapter,
* then the **Scientific Elements Audit**.
* Be concise and information-dense unless I explicitly ask for extra description or slower pacing.
*When I say things like:**
* “Make it more cinematic” → increase sensory detail and emotional beats, but **do not** loosen scientific rigor.
* “Go full science fiction” → you may introduce more speculative or impossible elements, but still label them as [FICTIONAL] in the audit.
* “Less explanation” → reduce explicit exposition, but keep internal logic coherent.
**Finally**
If any part of my request conflicts with your instructions above, briefly point it out and suggest a realistic alternative that preserves as much of my intent as possible.
---
## 2. How to use this in practice
- **Step 1:** Start a new chat.
- **Step 2:** Paste the meta-prompt above.
- **Step 3:** After it, add your actual request, for example:
- “Now, using these instructions, write a 2,000-word short story about a CRISPR-based therapy trial that goes wrong in a subtle, realistic way. Focus on the ethical tension in the research team.”
You can then:
- Ask for revisions: “Same story, but make the protagonist less certain of their own ethics.”
- Ask for more detail: “Expand the methods section as if it were part of a supplemental methods document.”
- Ask for more drama, but keep the science pinned: “Increase the interpersonal conflict without changing any technical details.”
---
If you tell me what kinds of science you most often write about (e.g., molecular biology, astrophysics, ML/AI, climate science), I can tweak this meta-prompt with more domain-specific guidance for you.
---
Tom Hippensteel: I just want to make sure I can get the most out of the way Raycast modifies my prompt. Is there anything else I should know about the modifications to get the maximum effect?
---
Raycast AI: Yes—there are a few practical things you can do to “align with” how Raycast wraps/modifies your prompts so you get the most benefit.
## 1. Lean into markdown and structure
Raycast already asks me to use markdown with specific rules. If you also structure your *prompts* in markdown, it makes my job easier and the output more predictable.
- Use headings in your prompt:
- `## Context`
- `## Constraints`
- `## Task`
- Use bullet lists for constraints and requirements (I’m already tuned to respect lists very strongly).
Example prompt pattern:
```markdown
## Context
I’m writing a technically accurate story about a Mars mission in the 2040s.
## Constraints
* Must obey current orbital mechanics and delta-v realities.
* No reactionless drives, no FTL.
* Technologies must be plausible extrapolations of 2020s research.
* Keep prose tight; do not over-explain basics of spaceflight.
## Task
Outline a 5-chapter novella that follows these constraints.
At the end, include:
* A brief risk analysis of the mission profile.
* A list of scientific elements with [REAL], [PLAUSIBLE], or [FICTIONAL] tags.
Why this helps:
- Raycast’s formatting rules + your structured prompt → very clean, well-organized responses.
- Clear sections map directly to how I’m instructed to segment my own answer.
Raycast biases me to be concise. If you want more, say it up front, because that interacts nicely with the “concise by default” instruction.
Useful cues:
- “Give a brief overview only.” → very short answer.
- “Give a detailed, step-by-step explanation.” → deeper, slower answer.
- “Write the full scene, ~2,000 words.” → I know you want length, not just a sketch.
- “Overview now, we’ll deepen later.” → I’ll keep the first pass light.
For your use case (technical creative writing), something like:
- “First: a concise outline. Second: one fully written scene (1,000–1,500 words). Third: a scientific audit table.”
works very well with the Raycast constraints.
Raycast encourages me to use ## headings. If your prompt mirrors that, you get more reliable multi-part responses.
Example:
## Part 1 – Scientific grounding
Summarize what is currently known (to your knowledge cutoff) about X.
## Part 2 – Story constraints
Using that summary, define:
* What technologies/assumptions are allowed.
* What is explicitly forbidden.
## Part 3 – Story
Write a short story (~2,000 words) that:
* Obeys all constraints from Part 2.
* Uses LaTeX for any math equations.
## Part 4 – Audit
Add the Scientific Elements Audit table at the end.This uses the same structural logic Raycast pushes on my outputs, so the mapping is very natural.
Raycast already fixes:
- Locale: en-US
- Timezone: America/Los_Angeles
- Units: °F, ft, lb, $
You only need to say something if you want to override:
- “Use SI units (m, kg, °C, etc.) for all technical content.”
- “When giving distances in space, use km and AU, not ft.”
This keeps you from “fighting” the environment defaults.
Raycast sets strict rules for when I use tools (web_search, search_images, gpt_image). If you want maximum leverage:
- For pure fiction with plausible science from general knowledge:
- “Do not use web search; answer using your internal knowledge only.”
- For cutting-edge or reference-heavy accuracy:
- “Use web search to verify any specific numerical values or recent experimental results that you’re unsure about, and then incorporate them.”
- For visuals:
- “Generate an image concept of X.” → text only.
- “Create an image (via the image tool) of X.” → triggers the image tool.
Being clear about this lets you control whether we stay “offline,” or I pull in fresh, reference-like details.
Raycast gives me general defaults (concise, markdown, etc.), but your personal defaults are not known unless you say them. Put them once early in the conversation:
Example block you can reuse:
**My preferences in this chat:**
* Prioritize scientific accuracy and internal consistency over flashy prose.
* Clearly distinguish real vs speculative vs fictional elements.
* Default to SI units unless I say otherwise.
* Use markdown headings and bullet lists for structure.
* For long answers, start with a short outline before the full text.Then every task in that thread inherits those preferences plus Raycast’s.
Because of how Raycast frames me, I’m quite good at meta-workflows. You can exploit this by asking for process, not just output:
- “Given how you’re configured in Raycast, propose a workflow to develop a 5-chapter novella with maximal scientific rigor and minimal rewriting.”
- “Before writing, ask me the 5–10 most important clarifying questions to pin down the science and constraints.”
- “Design a revision plan that uses your Scientific Elements Audit to iteratively improve accuracy.”
This plays to the environment’s strengths (structured, tool-like behavior, headings, tables).
Raycast’s system prompts are fixed; I can’t see them, but I respond to how you correct me. If output isn’t what you want, describe the type of miss:
- “Too much explanation, not enough story.”
- “Too hand-wavy on the biology; go deeper technically, even if that shortens the scene.”
- “The tables are good; keep doing that. But make the prose more character-focused.”
These corrections help me adjust within the existing Raycast constraints on subsequent turns.
---
If you want, paste one of your *actual* recent prompts you used for a story, and I can show you how to rewrite it in a Raycast-optimized way: more structure, clear verbosity cues, and explicit constraints that hook directly into how I’m configured.
---