# Personal AI Prompt Library — Template

A single-file starter you can keep in Markdown, paste into Notion, or convert
to a spreadsheet. One prompt per entry. Keep it boring on purpose.

## Fields (recommended)

- **ID** — stable, never reused (e.g. `p-0001`).
- **Name** — short, human, describes the job to be done.
- **Category** — Client comms, Writing, Productivity, Engineering, etc.
- **Tags** — 2–5 lowercase, comma-separated.
- **Prompt** — the full text with `{{variables}}` clearly marked.
- **Variables** — the list of `{{variables}}` used, in order.
- **Model** — the model you last verified this on (e.g. `gpt-4o-mini`).
- **Version** — semantic-ish: bump minor for edits, major for behavior changes.
- **Last tested** — ISO date (`YYYY-MM-DD`).
- **Test result** — one line: what worked, what you had to fix.
- **Privacy** — Safe / Redact client data / Do not paste secrets.
- **Notes** — pairings, warnings, links to related prompts.

## Naming conventions

- Use kebab-case or Title Case, pick one and stick to it.
- Start the name with the action: "Draft…", "Summarize…", "Explain…".
- Avoid dates and model names in the name; those live in their own fields.

## Version numbers

- Start every prompt at `1.0`.
- **Minor bump (1.0 → 1.1):** wording changes, small structure tweaks.
- **Major bump (1.1 → 2.0):** new output shape, new required variables,
  swapped model class.
- Never overwrite a major version — keep the previous entry archived.

## Example entry (Markdown)

```markdown
### p-0001 — Weekly client status email
- Category: Client comms
- Tags: email, status, recurring
- Model: gpt-4o-mini
- Version: 1.2 (2026-07-20)
- Variables: {{client_name}}, {{tone}}
- Privacy: no client data in prompt body

Prompt:
> Draft a concise weekly status email for {{client_name}} covering:
> progress this week, blockers, next week's plan, and one clear ask.
> Tone: {{tone}}. Length: under 180 words.

Last test: kept structure; removed one filler sentence.
```

## Maintenance routine (15 min / week)

1. Open the library. Sort by "Last tested" ascending.
2. Re-run the top 3 oldest prompts on their current model.
3. If output still matches expectation, update `Last tested`. Done.
4. If output drifted, bump version, note the drift, fix.
5. Archive anything you have not used in 90 days.

## Privacy notes

- Never paste API keys, passwords, or full customer records into a prompt
  body you save.
- If a prompt needs sensitive input, mark it `Privacy: fill at runtime` and
  keep the placeholder in the stored copy.
- Assume anything you save in a shared tool is visible to whoever has
  workspace access. Design accordingly.

## Migration advice

- **From chat history:** export or copy the prompts you actually re-used
  in the last 60 days. Skip everything else.
- **From scattered docs:** consolidate into one file first, then split
  into a spreadsheet or Notion database once you exceed ~25 entries.
- **From spreadsheet to Notion (or reverse):** map columns to fields
  first, dry-run 3 rows, then bulk-import.