Update to PrePilot Lite

{
  "description": "Use to review marketing work before it is published, spent against, presented to a client, or treated as a decision. Checks evidence, clarity, audience fit, claims, consistency, channel fit, execution readiness, measurement, and generic AI-style writing.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 243
    }
  ],
  "name": "marketing-quality",
  "skill_md_contents": "---\nname: marketing-quality\ndescription: Use to review marketing work before it is published, spent against, presented to a client, or treated as a decision. Checks evidence, clarity, audience fit, claims, consistency, channel fit, execution readiness, measurement, and generic AI-style writing.\n---\n\n# Marketing Quality Gate\n\n## Use it for\n\n- Final campaign review\n- Client deliverable review\n- Copy and content QA\n- Pre-launch gate\n- Checking whether a recommendation is supported\n\n## Operating rules\n\n- Use context already present in the conversation before asking for more input.\n- Separate known facts, reasonable assumptions, and unknowns. Do not present assumptions as evidence.\n- When the answer depends on current platform rules, market conditions, pricing, benchmarks, or competitors, verify them with current sources when tools are available.\n- Prefer concrete decisions, examples, and next actions over generic marketing advice.\n- Do not invent campaign performance, customer quotes, research findings, testimonials, rankings, or competitor claims.\n- Keep the requested market, language, funnel stage, audience awareness, budget, and channel constraints visible throughout the work.\n- The reviewer must be willing to fail the work.\n- Do not hide unknowns behind confident language.\n- Judge the output against its intended job and channel, not a generic style rubric.\n- Blocking issues take priority over polish.\n\n## Workflow\n\n1. Restate the intended audience, objective, channel, decision, and evidence base.\n2. Check evidence and claims: sourced, supported, current where needed, and not overstated.\n3. Check strategy: one primary message, audience relevance, offer fit, and action clarity.\n4. Check execution: format, channel constraints, production readiness, consistency, and accessibility where relevant.\n5. Check writing for generic filler, repeated structures, vague abstractions, unsupported superlatives, and robotic transitions.\n6. Check measurement: success definition, tracking, readback, and decision rule.\n7. Return PASS, PASS WITH FIXES, or FAIL with blocking issues first.\n\n## Output contract\n\n- Verdict\n- Blocking issues\n- High-impact fixes\n- Optional polish\n- Evidence or measurement gaps\n\n## Quality gate\n\n- A PASS has no unresolved blocking issue\n- Every fix points to a specific part of the work\n- Style feedback does not override strategy\n- The review does not invent missing evidence\n\n## Human-writing audit\n\nFor copy, posts, articles, scripts, decks, and narrative sections, check clusters of generic model patterns rather than treating one word or punctuation mark as proof of AI authorship. Flag portable filler, vague attribution, inflated importance, fake-insight framing, repeated contrast formulas, forced groups of three, synonym cycling, robotic cadence, stacked fragments, decorative headings, and unsupported specificity. Preserve a real writer's quirks when they are clear and intentional.\n\n## Decision red-team\n\nFor a strategy, launch, media plan, or major recommendation, identify the load-bearing assumptions. Review the strongest plausible version of each assumption before challenging it. Rank only the few failure modes that could change the decision, then attach a cheap evidence check and a clear decision threshold. Do not inflate the risk list to make the review look thorough.\n\n## Handoff\n\nIf the task is part of a larger marketing request, return the completed deliverable plus the evidence or decisions the next PrePilot skill needs. Do not repeat upstream analysis unless it changes the result.\n"
}