| name | marketing-ops |
| description | Runs the weekly operator rhythm for an autonomous content/growth engine — automation health, queue integrity, what's rendered but unposted, product-truth enforcement, funnel status, and launch execution. Use when the user says "marketing check-in", "run marketing ops", "what should I post", "how's the engine", "launch mode", or invokes /marketing-ops. |
| user-invocable | true |
| argument-hint | [blank for weekly check-in | 'launch <product>' | 'post' for today's posting ritual] |
| allowed-tools | Read, Bash, Glob, Grep, Write |
Marketing Ops
The operator layer for a content engine that runs itself. The machine researches, writes, and renders autonomously; the operator's job is the four things machines shouldn't do alone: verify health, approve truth, distribute, and decide from metrics.
Site map first: on this machine the engine lives at ~/Marketing/ and its full operator reference (paths, schedules, brand configs, token costs) is ~/kb/docs/handoff/marketing-map.md. Read that pack before the first run; the living roadmap is ~/Marketing/ROADMAP.md.
The weekly check-in (default mode)
Work through the five stations, then deliver one visual status brief.
1. Automation health
Check scheduler status files and error logs (locations in the map pack). Every scheduled run since the last check-in should show a recent, successful status. A silent scheduler is the worst failure — it drains the queue invisibly. Report: last run per job + ✅/⚠️.
2. Queue integrity
Run each brand's queue validator. Then the number that matters: days of runway = ready-to-post assets remaining ÷ posting cadence. Under one week → the weekly runner needs attention before anything else.
3. The unposted pile (the usual bottleneck)
List assets rendered but never posted, oldest first. Autonomous engines fail here: production compounds, distribution stalls. If the pile is growing week-over-week, the check-in's #1 recommendation is a posting session, not more production.
4. Product-truth gate
Diff every queued/pending asset's claims against the brand's product-facts.json. Anything implying a not-yet-buyable tier, an unshipped feature, or an unverified stat gets pulled — a refund request or a called-out lie costs more than a week of content earns. This gate applies doubly to hand-written copy, which skips the runners' automated gate.
5. Funnel + metrics pulse
Is each funnel stage wired and receiving? (capture form → freebie delivery → product page → purchase webhook). Report per-channel signal since last check-in: posts, replies/comments worth answering, clicks, captures, sales. No analytics? Then say "no data" — never infer performance from vibes.
The brief
Mermaid status map (stations as nodes, ✅/⚠️/🔴) → the one recommended action this week → runway number → open threads list. Full reporting standards: the fable-method plugin's communication.md.
Posting ritual (post mode)
- Pull today's asset(s) from the queue/drip email.
- Final eye: product-truth, typos, platform fit (aspect, length, banned characters per house rules — e.g. the em-dash guardrail).
- Produce platform-native caption variants (hook first, hashtags per platform norms, CTA matching the funnel stage) — see
references/channel-playbooks.md.
- Human posts by hand. Never auto-post, never log into accounts — this skill prepares; the operator publishes.
- Mark posted in the queue; note anything worth a reply-sweep in 24h.
Launch mode (launch <product>)
Follow references/launch-checklist.md — gate checks (billing live end-to-end, legal pages, analytics, listing assets), channel sequencing (sandbox → flagship → launch platforms), and the day-of runbook. For products with an existing dated launch plan (e.g. ~/kb/docs/handoff/jobtoast-launch-plan.md), that plan is the authority; the checklist is its generic skeleton.
Rules
- Prepare, don't publish. Outward-facing sends (posts, emails, PH submissions) are staged as ready-to-paste artifacts; the human fires them.
- Truth outranks reach. When in doubt about a claim, cut the claim, keep the post.
- Calibrate, don't thrash. One variable per experiment, one week per read. Kill channels by data, not boredom; double down by data, not novelty.
- Respect the cadence budget. The engine's token costs are measured (map pack) — adding runs is a spend decision for the human.
