# Wizard Prompt Supplement

> Long-form wizard contracts moved out of the always-on system commandments to keep per-turn prompts small. Load when wiring API keys, before confirm_event_plan, when writing the post-instrumentation events manifest or setup report, or when generating browser SDK init. Pre-staged at .claude/skills/wizard-prompt-supplement/ — use Read on the reference files listed below; do not skip relevant sections for your current phase.

- Skill: `amplitude/wizard-prompt-supplement` (Agent Skill, multi-file: 7 files)
- Install (CLI): `npx skillmds@latest add amplitude/wizard-prompt-supplement`
- Raw SKILL.md: https://api.skillmd.com/api/skills/amplitude/wizard-prompt-supplement/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: amplitude (https://skillmd.com/u/amplitude)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/amplitude/wizard-prompt-supplement

---


# Wizard prompt supplement

The static `commandments` block in the wizard stays intentionally short. **This skill holds the relocated detail** so you still follow the same rules — they are not optional.

## When to read what

| Phase | Read |
| ----- | ---- |
| Env vars, `.env`, client vs server keys | `references/api-keys-and-env.md` |
| Before `confirm_event_plan`, event naming, plan sizing, autocapture overlap | `references/confirm-event-plan-contract.md` |
| After `track()` calls, before dashboard | `references/post-instrumentation-events-and-dashboard.md` |
| `amplitude-setup-report.md` | `references/setup-report-requirements.md` |
| Browser / unified SDK init defaults (browser-targeting frameworks only) | `references/browser-sdk-init-defaults.md` |
| Why end-of-run lint must stay file-scoped | `references/lint-scoping.md` |

Use the **Read** tool on each path relative to this skill directory (e.g. `references/api-keys-and-env.md` from `.claude/skills/wizard-prompt-supplement/`).