Scripts & templates

TASK: [what you want done]
AUDIENCE: [who reads it]
CONTEXT: [facts it needs; paste them, don't make it guess]
FORMAT: [length, bullets/table, tone]
RULES: If you don't know something, write UNKNOWN. Cite a source URL for every number.
CHECK: At the end, list anything you assumed.

Covers the four things most weak prompts leave out: audience, context, format and a self-check.

Here is a claim: "[paste]".
1) Find the primary source (official site, law, filing). Give the URL.
2) Quote the exact sentence that supports or contradicts it.
3) Verdict: SUPPORTED / CONTRADICTED / NOT FOUND. Do not guess.

Forces a quote you can Ctrl+F on the source page.

Summarize the attached document for [audience] in [N] bullets.
Then list every date, dollar amount and deadline with its page number.
Do not add anything that is not in the document.

Page numbers make every fact checkable.

About me: I run a small business in Oregon. I'm not technical.
How to answer: plain English, short paragraphs, the answer first.
Always: give a source link for prices, laws and dates. Say "I'm not sure" instead of guessing.
Never: invent numbers, quotes or links. Never claim a task is done without a link or ID I can check.

Paste into ChatGPT Custom Instructions, Claude profile preferences, or a Gemini Gem.

Standing rules:
- Ask before sending email, posting, buying, or deleting anything.
- Tell me before creating any new scheduled job, with how often it runs.
- When you say something is done, include the link, ID or screenshot.
- If a tool fails twice, stop and tell me instead of retrying.

Cuts surprise actions and wasted runs.

Every Monday 8 AM: list every scheduled job/routine you run for me, its schedule, last run time, last result, and whether it produced anything I used. Recommend which to turn off. Do not change anything.

One cheap weekly run that catches jobs silently burning usage or silently dead.

Every weekday 7 AM: read my calendar for today and unread email from the last 12 hours. Send me 5 bullets: meetings, anything needing a reply, deadlines. Link each item. Do not reply to anyone.

Read-only, so no approval risk.

curl https://api.openai.com/v1/responses \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "MODEL_NAME", "input": "Write one sentence about flood safety."}'

Responses API (platform.openai.com/docs). Pick MODEL_NAME from OpenAI's models page. Keep the key in an environment variable, never in code you share.

from openai import OpenAI
client = OpenAI()  # reads OPENAI_API_KEY
r = client.responses.create(model="MODEL_NAME", input="Write one sentence about flood safety.")
print(r.output_text)

pip install openai

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model": "MODEL_NAME", "max_tokens": 300,
       "messages": [{"role": "user", "content": "Write one sentence about flood safety."}]}'

Messages API (docs.anthropic.com). max_tokens is required and caps the cost of each reply.

import anthropic
client = anthropic.Anthropic()  # reads ANTHROPIC_API_KEY
m = client.messages.create(model="MODEL_NAME", max_tokens=300,
    messages=[{"role": "user", "content": "Write one sentence about flood safety."}])
print(m.content[0].text)

pip install anthropic

curl "https://generativelanguage.googleapis.com/v1beta/models/MODEL_NAME:generateContent" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"contents": [{"parts": [{"text": "Write one sentence about flood safety."}]}]}'

Gemini API (ai.google.dev). Get a key in Google AI Studio.

curl https://api.x.ai/v1/chat/completions \
  -H "Authorization: Bearer $XAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "MODEL_NAME", "messages": [{"role": "user", "content": "Write one sentence about flood safety."}]}'

xAI API (docs.x.ai) is OpenAI-compatible, so the OpenAI SDK works with base_url https://api.x.ai/v1.

const r = await fetch("https://api.x.ai/v1/chat/completions", {
  method: "POST",
  headers: { "Authorization": `Bearer ${process.env.XAI_API_KEY}`, "Content-Type": "application/json" },
  body: JSON.stringify({ model: "MODEL_NAME", messages: [{ role: "user", content: "Hello" }] })
});
console.log((await r.json()).choices[0].message.content);

Node 18+. Run server-side only: an API key in browser code is public.

# Stop a loop before it runs away. Usage fields come back on every API response.
budget, used = 200_000, 0
for prompt in prompts:
    r = client.responses.create(model="MODEL_NAME", input=prompt)
    used += r.usage.total_tokens
    if used > budget:
        print("Budget hit, stopping at", used); break

Also set a hard monthly limit in the provider's billing settings.