Marketing Skills v2.6.0
A minor release adding one new skill (marketing-loops) for recurring marketing workflows, plus a version bump to ads for the Meta Andromeda / creative-as-targeting rewrite.
Total skills: 46
New skill: marketing-loops
The rest of the repo tells you what to do. This skill tells you what to keep doing on a schedule. A marketing loop is a recurring workflow an AI agent runs on a cadence — the weekly SEO opportunity scan, the ad-fatigue refresh, the churn-signal watch — each with a defined trigger, a bounded set of steps, a self-check, and an explicit stopping condition.
It's the operational cousin of marketing-ideas: ideas are what to try once; loops are what to keep doing. Inspired by the "loop library" pattern for AI-agent workflows, scoped to marketing operations. Reviewed with an independent second-opinion pass before merge.
The nine-part loop anatomy
Every loop is specified across nine parts, so it's safe to run unattended:
| Part | What it defines |
|---|---|
| Check cadence | How often it looks — matched to how fast the signal changes |
| Acts when | The action condition — most runs should check and skip |
| Purpose | The one outcome it exists to move |
| Skills used | Which marketing skills it orchestrates |
| Loop body | The ordered steps each run |
| Self-check | Verification before acting — is the signal real vs. noise? |
| State / idempotency | What it remembers so it doesn't double-act or re-nag |
| Stop / bail-out | When it skips, halts, escalates, or disables — plus error behavior |
| Output | Where results go — a file, PR, staged draft, or report |
The catalog: 43 loops
references/loop-catalog.md covers the full funnel, with concrete guardrails baked into each entry:
- SEO & Content — keyword-gap, ranking-drop watch, content-decay, internal-linking, programmatic-SEO quality, content-repurposing, content-calendar refill
- Paid — ad-fatigue, paid-search query-mining, retargeting hygiene, landing-page regression
- Earned / Social / Partnerships — newsjacking, social-listening, community-engagement, competitor-watch, backlink-prospecting, directory-submission, partner-pipeline
- Activation — onboarding drop-off, signup-funnel-leak, lead-capture-asset, feature-adoption
- Retention — churn-signal, lifecycle-email-refresh, re-engagement, email-deliverability, voice-of-customer
- Revenue — trial-conversion, PQL/upgrade-intent, pricing-page-experiment, paywall-optimization, expansion/upsell, failed-payment/dunning
- Referral & Advocacy — referral-nudge, review-and-UGC-harvest, review-site-management, case-study-sourcing
- Ongoing Ops — weekly-marketing-review, experiment-backlog, analytics-anomaly, brand-mention/reputation, tracking-QA, campaign-postmortem
Guardrails are specific, not generic: newsjacking carries a veto list (tragedy / politics / crisis / sensitive verticals) plus mandatory human approval; ad-fatigue requires a real significance bar (spend / impressions / conversions / attribution / learning phase); lifecycle-email optimizes on clicks / replies / complaints / bounces, not opens; pricing and paywall loops judge winners on revenue-per-visitor / refunds / churn, not conversion alone; review-harvest enforces consent / FTC / platform ToS; and the experiment-backlog loop is a thin wrapper that hands off to ab-testing.
Reference library (5 docs)
- loop-catalog.md — the 43 loops above, each a complete nine-part spec
- loop-template.md — copy-paste template + ordered fill-in prompts + a worked blank-to-filled example + a pre-ship checklist for authoring your own
- loop-orchestration.md — how loops compose into a four-layer system (sensing → diagnostic → action → learning), plus a staged rollout path so you adopt tracking + a weekly review before acquisition, and never build all 43 at once
- loop-state.md — where loop state lives (
.agents/loops/<loop>.json), the watermark / dedupe / cooldown / in-flight idempotency patterns, and a run-log format that doubles as a vanity-loop detector - loop-guardrails.md — a two-tier action model (autonomous-safe vs. gated), spend/send caps + allowlists, CAN-SPAM / GDPR / FTC / platform-ToS mapped to the loops each governs, PII handling, an always-escalate list, a required kill switch, and a pre-launch checklist
Anti-patterns & banned vocabulary
The skill is explicit that loops are disciplined systems, not autopilot:
- Looping without a stop condition → runaway spend or infinite churn
- The same cadence for every loop → most run too often and get ignored
- No self-check → the loop acts on noise, seasonality, or a tracking bug
- No human checkpoint on spend/publish actions
- Building 10 loops at once → start with one, prove it earns its keep, then add the next
Banned vocabulary: "set it and forget it," "fully autonomous marketing," "AI does everything," "10x on autopilot," "growth hacking machine."
ads → 2.1.0
Version + changelog bookkeeping for the Sabri Suby / creative-as-targeting rewrite (Meta Andromeda era) that shipped earlier without a version bump — so the repo's update check surfaces the change to installed users. Skill body only; frontmatter description unchanged.
Verification
- 46/46 skills in sync across
VERSIONS.md,marketplace.json,plugin.json, andREADME.md - New skill ships with a 7-case
evals/evals.jsonsuite - All internal skill cross-references resolve