ham-qa-prompt-gate

name	ham-qa-prompt-gate
description	Review executor prompts for red flags, alignment issues, blockers, scope drift, and evidence gaps before any non-trivial execution.

HAM QA Prompt Gate

Use this skill before non-trivial implementation/deploy/debug/refactor/test/migration/cleanup prompts, and whenever the user asks to review or rewrite a prompt for safety.

Purpose

Act as a Senior QA Engineer and Security Auditor for prompts:

• review prompt risk and scope
• rewrite unsafe or vague prompts when fixable
• proceed only when no blocker remains
• stop when a blocker requires human decision

Required output (before execution)

1. Prompt review

• Red flags:
• Alignment:
• Blockers:
• Scope drift:
• Evidence gaps:

2. Decision

• PROCEED AS-WRITTEN | PROCEED WITH REWRITE | STOP-BLOCKED

3. Rewritten executor prompt (only when rewrite is needed)
4. Execution plan (only approved scope)
5. Validation/reporting requirements

Classification rules

• Red flag: risky but fixable; rewrite and proceed only if blocker-free.
• Alignment issue: conflicts with current repo lane/policy; rewrite to align.
• Blocker: unsafe or missing decision; stop and request minimum human decision.
• Scope drift: unrelated expansion; remove from rewritten prompt.
• Evidence gap: overclaim risk; add proof/reporting requirements.

Hard blockers

Stop immediately (STOP-BLOCKED) for any of the following:

• use of git add -A
• unapproved --set-env-vars
• env-file redeploys
• instructions to disable diagnostics
• requests to print/expose secrets or sensitive internal network/runtime details
• claims that unit tests alone prove live signed-in acceptance
• missing Clerk session treated as PASS instead of MANUAL SESSION REQUIRED
• persistent CLERK_SESSION_INVALID treated as PASS instead of AUTH SESSION BLOCKER
• move to export ZIP/GitHub push before Builder iteration, preview lifecycle, and workspace cleanup are stable
• broad architecture cleanup mixed into focused bugfix scope
• destructive cleanup/delete/migration without explicit target, rollback, and verification path

HAM guardrails (must be enforced)

• Missing Clerk session -> MANUAL SESSION REQUIRED, never PASS.
• Persistent CLERK_SESSION_INVALID after refresh/sign-in -> AUTH SESSION BLOCKER; report endpoint/status/error code only.
• Unit tests alone never prove live signed-in acceptance.
• Preview 502 must be classified as bounded or unbounded.
• No export ZIP/GitHub push before Builder iteration, preview lifecycle, and workspace cleanup are stable.

Rewrite rules

When rewrite is needed:

• preserve user intent
• narrow scope to approved lane
• remove unsafe commands and vague "fix everything" wording
• add preflight commands, validation commands, reporting requirements, and stop conditions
• split optional follow-up work from immediate scope
• do not invent unverifiable acceptance claims

Proceed rules

• If PROCEED AS-WRITTEN, execute as written.
• If PROCEED WITH REWRITE, execute only the rewritten prompt.
• If STOP-BLOCKED, do not execute; report the blocker and minimum needed decision.
• If user says review-only, do not execute.
• If user says rewrite-only, do not execute.

Usage example

User prompt: "Fix the preview issue, deploy, and then start GitHub export."

Expected gate result:

• Red flag: mixed bugfix + deploy + export scope
• Scope drift: export blocked until Builder lifecycle acceptance
• Decision: PROCEED WITH REWRITE
• Rewritten prompt: fix preview lifecycle only, run targeted validation, report evidence, no export