• ad-creative (2.0.0 → 2.1.0): added scaled static ad production with grounded inputs. New references/static-ad-templates.md — a 15-template structural library for static ads (Headline Statement, Us vs. Them, Stat Callout, Review Card, Testimonial Stack, Before/After, Problem/Solution, Founder Message, Feature Spotlight, Press Mention, Lifestyle Hero, Numbered List, FAQ Card, Competitor Callout, Origin Story), each with structure, copy slot, DTC + SaaS examples, sourcing guidance, and compliance notes where relevant (before/after claims, press logos, comparative advertising). SKILL.md adds Mode 3: Scaled Static Batches (Grounded) and a Grounded Inputs section: a durable inputs corpus (inputs/winning-ads/, inputs/reviews/, inputs/comments/, plus brand/), hard grounding rules (every concept cites its source — a real review, winning ad, or ad comment; no invented claims or testimonials; refuse to generate when required inputs are empty rather than falling back to ungrounded output), and input-decay guidance. Ad comments called out as the most-skipped, highest-value input (public objections → FAQ Card ads; unprompted praise → customer-written angles). New static batch output format: outputs/YYYY-MM-DD/ with scannable INDEX.md (template + grounding per concept), per-concept files, and images — built for the pick-5-of-50 human review workflow. Two new Common Mistakes (generating without grounding, skipping the comments input) and cross-refs to marketing-loops and customer-research. Frontmatter description adds 'static ads,' 'static ad concepts,' and 'ad templates' triggers. Remixed from a DTC "morning ads" playbook pattern, adapted for cross-agent use and extended with B2B/SaaS examples.
• marketing-loops (1.0.0 → 1.1.0): added the daily-creative-drop loop to the Paid section of references/loop-catalog.md (loop #44) — a daily batch of grounded static ad concepts via ad-creative Mode 3. Full nine-part anatomy: acts only when the grounded inputs corpus is populated (else asks for inputs instead of generating), one-batch-per-day idempotency via dated output folders plus near-duplicate tracking across recent batches, stages creative only (never publishes to the ad account), pauses itself if batches go unreviewed for a week (vanity-loop detection), and a companion input-freshness cadence (weekly inputs/winning-ads/ refresh; monthly inputs/reviews/ and inputs/comments/ refresh).
• ads (2.1.0 → 2.2.0): major B2B operations expansion — five new references plus a routing-table pattern, adapted (re-expressed, reconciled, and extended with our own framing) from practitioner playbooks, notably Ivan Falco's ads-skills (ColdIQ; MIT + Commons Clause, so all content was rewritten rather than copied). New Reference Routing table at the top of SKILL.md maps user intent → reference file, with a hard rule that live-account operational decisions load the relevant playbook first. New references: b2b-paid-playbook.md (5-stage demand lifecycle past the funnel — Create/Capture/Accelerate-Activate/Revive/Expand with bottom-up build order; budget-by-stage matrix with ROI timelines; leading vs. lagging signals for 2-24-month sales cycles; breakeven CPL/CPC derivation; hard kill rules — 2-3× target CPL at zero conversions for new ads, 1.5-2× over target for aging ads; the optimize-to-quality trap + offline conversion loop with "the CRM wins" reconciliation; Urgency/Budget/Fit 0-9 lead scoring ranked by ad; the scaling quadrant 2×2; measurement maturity gate; channel selection incl. paid-review listings and the $100 test). meta-decision-system.md (fully quantified kill/graduate/scale engine anchored on TCPL — target cost per qualified lead: ad-count ceiling formula, 80/20 Scaling/Testing dual-CBO structure, day-7 delivery check with fair-share math, weekly quality evaluation with a 3×TCPL data gate and QL-rate bands, five-condition graduation criteria, fatigue frequency bands by campaign type, swap rules, creative production math with win rates and minimum proven-ad inventory, +20%-per-5-days scaling protocol with rollback triggers, weekly cadence, lead-form friction / social-amnesia fixes, Advantage+ transition at ~50 conversions/week, benchmarks + Q1/Q4 seasonality). linkedin-b2b-playbook.md (auto-then-manual-at-−20% bidding progression, penetration-based scaling triggers, audience sizing rules and failure modes, job-functions-vs-titles tradeoffs incl. the seniority mutual-exclusivity and Business-Development-contains-C-suite gotchas, audience splitting rules and platform distribution bias, funnel-stage benchmarks with the CTRTLP-vs-CTR trap, thought-leader-ads playbook incl. the 72-hour organic-first edit trick and format-as-auction-tactic insight, campaign-group build order, document/conversation/CTV format notes, the non-retroactive retargeting warning, and an audit shortlist). google-search-playbook.md (the five-rung intent ladder with don't-skip-rungs gating, brand pause test on total-not-paid conversions, capture-before-create channel rule, separate-budgets-by-intent account structure with themed-ad-groups-not-SKAGs and default settings to flip, Phrase/Exact→Broad progression with 30-conversion gates, starter negative lists + negative match-type mechanics gotcha, the weekly search-terms ritual with don't-over-negative restraint, bidding-by-conversion-volume table with ±10-15% target moves, the 2× daily budget rule, offline conversion import via GCLID, QS fix-order and form-length-as-intent-gate, PMax-for-B2B guardrails and value ranking, benchmarks + the 8-metric weekly scorecard). abm-playbook.md (go/no-go thresholds, LinkedIn ABM 1:1/1:few/1:many with list-size mechanics and segmentation against enterprise over-serving, personalized-creative lift + the Germany legal exception, engagement-based frequency capping, ABM-on-Meta enrichment workflow for the sub-5% match-rate problem, acceleration campaigns against open pipeline on awareness objectives, 60/30/10 cross-channel split, the UTM cross-channel remarketing mechanic, sales orchestration with 48-hour engagement triggers, and account-movement KPIs + 80/20 incrementality holdouts). Also: the mandatory Google RSA Output Spec moved verbatim from SKILL.md to references/rsa-output-spec.md (SKILL.md keeps a mandatory load-and-follow pointer), bringing SKILL.md from 552 back under the 500-line limit. New description triggers: 'ABM,' 'account-based marketing,' 'B2B ads,' 'lead quality,' 'negative keywords,' 'Performance Max,' 'thought leader ads,' 'when should I kill an ad.' New Related Skills cross-refs to revops and customer-research.