Lucid AI Lucid AI Operator's Toolkit

Operator's Toolkit

The reusable building blocks behind the Lucid AI certificates — ready to copy into your own work. Tune the bracketed parts to your business.

RCTF master prompt AI acceptable-use policy Lead-qualifier schema Extraction schema Production-readiness checklist

1 · The RCTF master prompt

The four dials that turn a vague request into a usable result: Role, Context, Task, Format. Paste it, fill the brackets, iterate.

ROLE: Act as my [job function, e.g. customer-support lead] with [N] years' experience in [industry].

CONTEXT: [Who this is for, the situation, any constraints, what good looks like, and what to avoid. Paste the relevant facts the model can't already know.]

TASK: [The single specific thing you want produced. One job per prompt.]

FORMAT: Return it as [exact structure — e.g. 3 short paragraphs / a table with columns X, Y, Z / a 5-bullet summary]. Tone: [plain / formal / friendly]. Length: [limit].

If anything essential is missing, ask me one clarifying question before you start.

2 · AI acceptable-use policy (5 lines that make it safe)

The minimum policy that keeps AI use accountable. Adopt it as-is or expand each line. The named owner is the most important part.

LUCID-STYLE AI USE POLICY — [Company name], v1, [date]

1. ALLOWED DATA: We may put [public info, internal drafts, anonymised data] into approved AI tools. We never paste [customer PII, card/bank data, unreleased financials, contracts, source IP, passwords].
2. APPROVED TOOLS: [Tool A] for [use], [Tool B] for [use]. Anything else needs sign-off.
3. HUMAN REVIEW: A named person reviews AI output before it ships to a customer, goes on the public record, or informs a decision with legal/financial/people impact.
4. NO-GO ZONES: No fully-automated [hiring/dismissal, legal advice, medical advice, final financial sign-off]. No feeding restricted data to public tools.
5. OWNER: [Person's name] owns this policy, answers questions, and reviews it every [quarter].

3 · Lead-qualifier schema (BANT as JSON)

Make a sales agent return a structured, scoreable judgement instead of prose — so routing and follow-up can be automated and audited.

Return ONLY this JSON for the lead, based strictly on what they said:
{
  "budget":    { "signal": "none|low|fit|strong", "evidence": "" },
  "authority": { "signal": "none|influencer|decision_maker", "evidence": "" },
  "need":      { "signal": "none|nice_to_have|urgent_pain", "evidence": "" },
  "timing":    { "signal": "none|exploring|this_quarter|now", "evidence": "" },
  "score":     0,                      // 0-100, your weighted judgement
  "route":     "nurture|sdr_followup|book_demo|disqualify",
  "confidence":"low|medium|high",
  "next_step": ""
}
If a field is unknown, use "none" and say so in evidence — never invent it.

4 · Document extraction schema

Force structured, checkable output from documents (invoices, forms, contracts) — with a confidence flag so low-certainty fields go to a human.

Extract the following from the document. Return ONLY valid JSON.
{
  "fields": {
    "[field_1]": { "value": "", "confidence": "low|medium|high", "source_quote": "" },
    "[field_2]": { "value": "", "confidence": "low|medium|high", "source_quote": "" }
  },
  "totals_checked": true,            // did the numbers reconcile? (tool/maths, not your guess)
  "needs_human_review": false,       // true if ANY field is low confidence or totals don't reconcile
  "notes": ""
}
Rules: quote the source text for every value. Never compute a total yourself — defer to the provided calculation. If unsure, set confidence "low" and needs_human_review true.

5 · Production-readiness checklist

Before any AI automation runs against real data or customers, walk this list. It's the line between a demo and something a business can actually depend on.