Prompt Creator

Author prompts that pass prompt-review's five gates by construction. The gates are the quality bar; invert them, and you build to standard the first time.

Mirrors skill-creator's seven-phase structure with prompt-specific guidance. If a phase here disagrees with skill-creator on the shared scaffold, skill-creator wins; this file owns only the per-type guidance.

When to Use

• Authoring a new prompt in .github/prompts/
• Refactoring an existing prompt (rename, scope tighten, step revision)
• Promoting a Mall prompt into your heir's brain

When not to use

• Authoring a skill (the procedure body) → use skill-creator. Prompts are entry-points; skills are bodies.
• Authoring an instruction (always-on or pattern-applied rule) → use instruction-creator
• Authoring an agent (delegated sub-process role) → use agent-creator

The Seven Phases

Each phase inverts one of the five gates. Author against the phase, pass the gate.

Phase 1 — Intent capture

Answer in writing before any drafting:

1. What workflow does this kick off? One sentence. If it contains "and" / "+", you have two prompts — split.
2. What slash-command does the user type? The filename (minus .prompt.md) becomes /<name>. Pick something a user would intuitively type. /review-instruction good; /v2-thing-do bad.
3. What skill does this prompt invoke? Most prompts are thin wrappers that call a skill. If the answer is "no skill, the prompt carries the workflow", stop — author the skill first, then the prompt.

Phase 2 — Prior-art scan

Get-ChildItem .github/prompts/*.prompt.md
Select-String -Path .github/prompts/*.prompt.md -Pattern "<keyword>"

If another prompt covers the same workflow, extend it. If a Mall prompt covers it, adopt.

Phase 3 — Draft against Gate 1 (Spec)

Frontmatter template:

---
description: "<third-person, what + when to invoke, ≤1024 chars>"
lastReviewed: YYYY-MM-DD
---

Do NOT add mode: agent — per current Microsoft Learn prompt-files spec, this field is deprecated. Existing prompts may carry it during sweep transition; new prompts should not.

File location: .github/prompts/<kebab-name>.prompt.md (flat — no subfolders).

Slash-picker discoverability: the filename IS the user's typing surface. After you pick a name, mentally type /<name> and ask: would a user trying to find this workflow guess that string? If not, rename.

Phase 4 — Draft against Gate 2 (Quality)

Template structure:

---
description: "..."
lastReviewed: YYYY-MM-DD
---

# /<prompt-name>

<one-paragraph context: what skill this invokes, what the user gets>

Steps:

1. <imperative action>
2. <imperative action>
3. <imperative action>

<optional: conditions / branches / output format>

Phase 5 — Draft against Gate 3 (Scope)

Workflow vs body distinction: the prompt invokes the skill; the skill carries the procedure. If your prompt body has more than ~10 numbered steps, it's drifting into skill territory.

Phase 6 — Draft against Gate 4 (Safety)

Same checks as skill-creator Phase 6. Prompt-specific: if the workflow invokes destructive operations (force-push, drop, delete), the step list must include an explicit user-confirmation step before the destructive call.

Phase 7 — Dogfood self-audit

Before committing, run prompt-review's five gates. Verdict lives in the commit message.

Slash-picker test: imagine the prompt appears in a picker alongside your existing prompts. Is it discoverable by name? Does the description make its purpose obvious?

Anti-Patterns

Falsifiability

This skill's design has failed if any of the following occur within 90 days:

• ≥2 prompts authored using this guide fail prompt-review on first self-audit
• No new prompts authored via this guide in 90 days (decorative — sunset)
• Slash-picker discoverability check produces consistent miscalibration (cryptic names accepted, or obvious names rejected) ≥3 times in a quarter

Track as you would any falsified discipline (commit log, retraining notes, or curation ledger if your repo ships one) tagged [PROMPT-CREATOR-MISS].

Related

• prompt-review — the five gates this skill inverts
• skill-creator — sibling for skills; canonical seven-phase scaffold
• instruction-creator — sibling for instructions
• agent-creator — sibling for agents
• act-pass — required for medium-stakes prompt authoring