Marketing Orchestrator

marketing-orchestrator — Marketing Skills Package entry point

Single entry point for the Marketing Skills Package — 10 sibling skills (marketing-orchestrator, marketing-campaign-planner, marketing-positioning, marketing-icp-research, marketing-channel-strategist, marketing-copywriter, marketing-seo-researcher, marketing-social-content, marketing-launch-runbook, marketing-analytics-attribution) installed globally at ~/.claude/skills/marketing-* (symlinked to ~/Pi-Dev-Ops/skills/marketing-*). Available in every project, not just Pi-Dev-Ops.

Discovery brief gate (turn 1, mandatory)

Adopted from nexu-io/open-design (Apache-2.0). Before any sub-skill is dispatched, the orchestrator emits a structured brief and refuses to proceed until every required field is filled. No "I'll work it out as I go" — the brief locks scope and prevents downstream skills from inventing audience, tone, or channel mid-stream.

Required fields

Optional but recommended: topic, competitors, forbiddenAngles, linearTicket.

Hard stop conditions

• Any required field missing or vague → orchestrator returns the brief with blocked: true and a per-field reason. Founder fills in, re-runs.
• tone includes a value outside the brand's voice.tone → block.
• outcome is unfalsifiable ("be impactful") → block.

This is the single most reliable lever to keep downstream wave plans short and sub-skills focused. Do not relax it.

Invocation

The user can invoke the package by:

• Saying any of: "use the Marketing Skills Package", "marketing package", "use marketing".
• Naming any individual skill (e.g. "use marketing-copywriter for the landing page").
• Submitting a brief that names a marketing artifact: "campaign for X", "launch plan for Y", "GTM strategy for Z", "landing page for ...", "email sequence for ...".

The marketing substrate

Templates, frameworks, and per-job artifacts live at:

~/Pi-Dev-Ops/marketing-studio/
├── templates/         — campaign-brief, launch-runbook, email-sequence, landing-spec, …
├── frameworks/        — JTBD canvas, positioning canvas, ICP canvas, AIDA, PAS
├── scripts/           — UTM builder, attribution helpers
├── .research/         — per-job artifacts (campaigns/, icp/, seo/, wave-plans/)
└── outputs/           — fallback when calling project has no .marketing/ dir

Brand voice / forbidden words / audience / tagline read from the shared BrandConfig files at ~/Synthex/packages/brand-config/src/brands/{slug}.ts — single source of truth across both packs. No duplication. (Migrated from Pi-Dev-Ops/remotion-studio/src/brands/ per RA-1985 / Synthex SYN-897.)

Output

A wave plan JSON written to marketing-studio/.research/wave-plans/{job_id}.json:

{
  "jobId": "synthex-launch-2026-04-28T15-30-00",
  "brand": "synthex",
  "campaignType": "product-launch",
  "audience": "ML engineers and platform teams",
  "channels": ["linkedin", "youtube", "email"],
  "durationDays": 30,
  "deliverables": ["positioning-doc", "icp-research", "landing-spec", "email-sequence", "launch-runbook", "video-assets"],
  "linear": { "teamId": "...", "projectId": "..." },
  "outputDir": "~/Pi-CEO/Synthex/.marketing/",
  "waves": [
    { "id": 1, "parallel": [
      { "skill": "marketing-positioning" },
      { "skill": "marketing-icp-research" }
    ]},
    { "id": 2, "parallel": [
      { "skill": "marketing-channel-strategist" },
      { "skill": "marketing-seo-researcher" }
    ]},
    { "id": 3, "parallel": [
      { "skill": "marketing-copywriter", "args": { "artifact": "landing-spec" } },
      { "skill": "marketing-copywriter", "args": { "artifact": "email-sequence" } },
      { "skill": "marketing-social-content", "args": { "channels": ["linkedin", "x"] } }
    ]},
    { "id": 4, "parallel": [
      { "skill": "remotion-orchestrator", "args": { "composition": "Explainer", "channel": "linkedin", "durationSec": 60 } },
      { "skill": "marketing-launch-runbook" },
      { "skill": "marketing-analytics-attribution" }
    ]}
  ]
}

Wave-count discipline

• ≤3 waves for a single artifact (one blog post, one ad).
• ≤5 waves for a content batch (3-5 artifacts, no launch).
• ≤8 waves for a full launch (positioning → research → content → assets → runbook → attribution).

Campaign type routing (brief → campaignType)

Brand resolution

Same slug map as Remotion: dr / nrpg / ra / carsi / ccw / synthex / unite. If the brand is unknown, dispatch remotion-brand-research (cross-pack call) before any marketing work — voice and forbidden-words must exist before copy gets written.

Composition with Remotion Skills Package

When the wave plan includes any video deliverable, marketing-orchestrator does NOT author video itself. It calls remotion-orchestrator with a fully-formed sub-brief: {brand, composition, channel, durationSec, topic}. The Remotion pack handles storyboard / motion / render. Marketing pack handles strategy / copy / non-video.

Per-project usage model

The package is shared infrastructure; each calling project supplies its own runtime config and API keys.

Per-project env keys (graceful degradation when absent):

• ANTHROPIC_API_KEY / OPENAI_API_KEY — copy + research LLM calls.
• PERPLEXITY_API_KEY — ICP + SEO + competitor research (skip with warning if missing).
• RESEND_API_KEY / MAILCHIMP_API_KEY — actual email send (skip; produce drafts only).
• LINEAR_API_KEY — campaign tickets.
• GOOGLE_ANALYTICS_PROPERTY_ID / POSTHOG_API_KEY — attribution dashboards.

What this skill does NOT do

• Does not author copy — that's marketing-copywriter.
• Does not run keyword research — that's marketing-seo-researcher.
• Does not render video — that's remotion-orchestrator (cross-pack call).

It only plans + delegates.

Adding a new campaign type

1. Add a row to the routing table above.
2. Define the wave plan template for that type (which skills, in what order).
3. Document in PACKAGE.md.

5-D critique gate (mandatory before final wave emission)

Adopted from nexu-io/open-design (Apache-2.0). After the final wave's deliverables land in <calling-project>/.marketing/{job_id}/, spawn an Opus 4.7 subagent via opus-adversary to score the campaign across five dimensions, each 0–10 with cited evidence:

1. Philosophy consistency — does the campaign carry one positioning thesis through every artifact (landing page hook, email sequence opening line, social hook, video script)? Cite the thesis from marketing-positioning output and check each artifact echoes it.
2. Visual hierarchy — across artifacts, does the eye land on the same hero stat/promise first? Cite which artifact breaks the order.
3. Detail execution — UTMs present and parameterised, no broken anchors, no AI-filler words leaked through, no "We/Our/I/Us/My" lead-ins (per Pi-CEO content rule).
4. Functionality — does the channel mix actually serve the audience surfaced by marketing-icp-research? Email cadence within sustainable bounds set by marketing-channel-strategist?
5. Innovation — one campaign move that earns memory (specific stat, contrarian framing, unusual CTA), or generic-launch median?

Pass / fail

• Pass: every dimension ≥ 6 AND mean ≥ 7 → wave plan emits, Linear ticket opens.
• Fail: any ≤ 5 OR mean < 7 → loop back: assign Fix list to the responsible sub-skill (positioning / copywriter / channel-strategist / social-content) and re-run that wave only.
• Override: founder may force-pass via explicit message. Logged to marketing-studio/.research/wave-plans/{job_id}.critique.json.

Use opus-adversary for the spawn — do not write a parallel harness. The critique JSON sits beside the wave plan; the founder reviews before any external dispatch (paid ads, scheduled email blast, Linear publication).