safely-use-ai-101
The primer — flagship
The five baseline rules: no invented facts, no PII echoed back, no auto-send. Every other skill builds on this.
Email triage. Safe summaries. Meeting notes. Weekly review. Proposals. Invoice reminders. Drop into Claude, ChatGPT, or Cursor — with compliance guardrails on every output.
Plus the flagship safely-use-ai-101 primer — load it once per session and every other skill builds on it.
The five baseline rules: no invented facts, no PII echoed back, no auto-send. Every other skill builds on this.
Respond / read / archive with one-line reasoning per email. Never auto-sends or archives.
Auto-redacts SSN, DL, bank routing, full DOB before a single word of the summary leaves the model.
Decisions, action items with owners, open questions. No invented commitments — ever.
Acknowledges before answering. No pressure tactics, no manufactured urgency.
From your actual week's activity. No "delve," no em-dash floods, no invented stats.
Shipped / slipped / lesson / energy / next-three / drop-one. 20 minutes, every Friday.
From a discovery call to a structured proposal — every assumption flagged so nothing fabricated slips through.
30 / 45 / 60-day tiers. No threats, no fabricated late fees, no legal language.
Compare 2-4 vendors on price, features, risk. Returns a memo that helps you decide — doesn't decide for you.
The settings at the top of every file tell the AI the rules to follow on every reply. It's not a prompt you have to remember — it's loaded once per session and enforced for the whole conversation.
This is what stops AI from inventing statute numbers, leaking your client's SSN into a summary, or auto-sending an email you haven't reviewed.
--- name: safely-use-ai-101 jurisdiction: none version: 2026.05 guardrails: human_in_loop: true no_invented_facts: true cite_or_flag: always pii_minimization: [SSN, DL, bank routing, full DOB, credit card] refuse_to_draft: [legal advice, medical advice, tax advice] escalation_triggers: [confidential client data, accusations, regulatory matters] --- # Safely Use AI — The Primer Load this skill once at the start of any work session. It teaches the AI the baseline rules for handling sensitive information and producing outputs you can ship without rewriting...
Forward unread mail into Claude with email-triage. Get respond / read / archive with reasoning. Nothing sent without you.
Paste a 20-page doc with safe-summarize. Names, SSNs, bank info redacted before the summary ever leaves the model.
Dump the week's notes with weekly-review. Structured walk: shipped, slipped, lesson, energy, next-three.
"These skills are built from real client files and the compliance edges people actually trip on. Nothing in the pack is theoretical."
A skill is a Markdown file that teaches your AI how to handle a specific job — with guardrails, examples, and "what NOT to do" rules baked in. Unlike a one-shot prompt, it stays loaded across the whole conversation, behaves consistently, and prevents the failures (invented facts, leaked info, off-brand voice) that make AI output unshippable.
Yes. The files are plain Markdown — they work with Claude Desktop, Claude Code, ChatGPT (Custom Instructions or Projects), Cursor, Gemini, Perplexity, and any AI that reads files or accepts a paste.
The bundle ships with a step-by-step instructions PDF for every major AI platform. The short version: save the .md file, then either upload it as an attachment, paste it as a system message, or add it to your AI's Project or Custom Instructions. Under five minutes per platform.
Yes. Open in any editor, modify, save. Most buyers swap in their own example data, adjust the voice to match their writing style, or layer their own guardrails on top.
Yours to use and edit for your own work or your single business. Open the files, change anything, save your own versions. Off-limits: repackaging them for resale as your own product. Team and multi-seat licenses available — email for pricing.
One-time payment of $39. Lifetime updates.