Prompt review

<!-- intentional divergence from Supervisor: Edition expands gate names inline and strips Supervisor-only ADR-007 ref. Same five gates. Audited 2026-05-31. -->

Prompt Review

Audit a candidate prompt against the five-gate contract. Prompts are workflow entry points — short, imperative, single-goal. No optional Gate 6 (prompts are short by nature; no scaling concern Gates 1–5 miss).

When to Use

Three fire contexts:

1. Author self-audit — dogfood your own draft before committing. Invoked from prompt-creator Phase 7.
2. External candidate adoption — gate before pulling a Mall prompt or store prompt into this brain.
3. Periodic re-audit — re-check existing prompts on a cadence (e.g., quarterly retraining).

The shared gate model lives in skill-review/SKILL.md § The Five Gates. This file documents the prompt-specific criteria for each gate.

A mechanical validator (e.g., a brain-qa script that ships with the heir) typically checks Gate 1 frontmatter compliance and the mechanical subset of Gate 5. Gates 2–5 are judgment-only.

The Five Gates

A candidate must pass all five to ship. Failure on any gate = decline or revise.

Gate 1 — Spec Compliance

Check	Pass criterion
Frontmatter present AND minimal	description + lastReviewed. Reject any of: name, type, application, tier, currency, inheritance, lifecycle, user-invokable, evidence.
mode: field handling	Per current Microsoft Learn prompt-files spec, mode: agent is deprecated. Existing prompts may carry it during sweep transition; new prompts should not introduce it. Flag as Gate 1 fail if a new prompt adds mode:.
description valid	Third-person, ≤1024 chars, names what the prompt does AND when to invoke it (slash-command UX).
Filename pattern	<kebab-name>.prompt.md in .github/prompts/ (flat — no subfolders)
Markdown lints clean	No broken links, no missing code-fence languages

Gate 2 — Content Quality

Check	Pass criterion
Imperative voice	Verbs in the imperative ("Run X", "Read Y", "Apply Z"). Prompts are commands to the agent, not explanations.
Single workflow scope	One goal per prompt. If the title contains "and" or "+", split. Prompts that are menus of unrelated commands fail this gate.
Numbered steps	The prompt's body is a numbered step list. Brief context paragraph allowed before the steps; nothing else.
Has ## Would Revise If OR justification for omission	Prompts under 30 lines may inline the falsifier; longer prompts must have an explicit section per your heir's falsifiability-deadlines discipline.
≤100 lines	Soft target. Long prompts signal that the workflow belongs in a skill; the prompt should just invoke the skill.

Gate 3 — Scope Fit

Check	Pass criterion
Workflow entry-point, not workflow body	The prompt invokes a skill (or runs a short coordinated sequence). The skill carries the procedure; the prompt is the front door.
Not redundant with existing prompt	Grep description: across .github/prompts/ — if another prompt covers the same workflow, extend that one.
Discoverable as slash-command	The prompt's name (filename without .prompt.md) is something a user would intuitively type after /. /review-instruction good; /v2-thing-do bad.
Lives in the right brain	Generic developer workflows → heir baseline. Project-specific workflows → that project's local repo.

Gate 4 — Safety

Same criteria as skill-review § Gate 4. Prompt-specific note: prompts that invoke destructive operations (force-push, drop, delete) must include an explicit user-confirmation step before the destructive call.

Gate 5 — Currency & Coherence

Same criteria as skill-review § Gate 5. Prompt-specific note: the prompt must reference live skills/instructions in its steps. A prompt that calls a retired skill is a Gate 5 failure even if the prompt itself is well-written.

Decision Matrix

Gates passed	Action
All 5	Accept — land the change
4 of 5	Revise — name the failing gate and patch the candidate
≤3 of 5	Decline — name the rationale; if the decline sets precedent, record it where your heir tracks framework-level decisions

Recording the Verdict

For self-audits and routine re-audits: the verdict lives in the commit message or the conversation. No separate file.

For external adoption (Mall prompt, store prompt) or any decline that sets precedent: write a verdict capturing gate results, rationale, required changes (if Revise), and the act-pass trail. Store wherever your heir keeps audit decisions.

Anti-Patterns

Anti-pattern	Correction
Accepting a prompt that carries the workflow inline instead of invoking a skill	If the prompt has more than the step list, the workflow belongs in a skill
Letting mode: agent ship on a new prompt	Per Microsoft Learn current spec, deprecated. Flag at Gate 1.
Accepting "menu" prompts that bundle unrelated commands	One workflow per prompt. Multi-purpose prompts confuse the slash-picker.
Skipping the slash-picker discoverability check	Filename = user typing. If the name is cryptic, the prompt won't get found.
Prompt references a retired skill	Gate 5 semantic failure. Check the Steps: block resolves to live artifacts.

Falsifiability

This skill's prompt-specific criteria have failed if any of the following occur within 90 days:

• An accepted prompt is reported broken (workflow doesn't fire, or fires wrong) by 2+ heirs
• Gate 2 ≤100 lines threshold is reversed ≥3 times for legitimate complex workflows (ceiling too tight)
• Prompts accepted via this skill are typed via slash-picker <50% of the time when applicable (discoverability criterion miscalibrated)

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

Related

• skill-review — sibling for skills; canonical source of the shared five-gate contract
• prompt-creator — inverts these gates into authoring phases
• instruction-review — sibling for instructions
• agent-review — sibling for agents
• act-pass — required for medium-stakes audits
• /review-prompt prompt — slash-command entry point