# Changelog: Boring Marketing Skills ## 4.1.1 (2026-11-16) ### Fixed - `creative`: the visual check now requires the brand name to be readable in every ad, an offer or next step in every non-native ad, no unsupported claims about competitors, and no AI-generated person beside a real review. Found in a blind visual head-to-head, where our coffee ads won on truth but lost on brand clarity. ## 4.1.0 (2026-11-16) Tested against another popular open-source marketing library in a blind head-to-head on 32 jobs across six kinds of business (a plumber, a dental group, a software company, a coffee subscription, a law firm and an agency). The changes below come from what lost and why. ### New - **Ad creative rebuilt from real ads.** We pulled 336 live Meta ads from 48 of the heaviest DTC advertisers and looked at every one, including frame-by-frame sheets of the videos. `creative` now carries 21 formats with how common each is, how long they run, real examples and how to make each for a local business, a SaaS company, a practice or a brand; a hook bank of the openings that actually appear; a batch rule (at least 4 formats, 3 angles, one native piece, one proof piece, one UGC video with 3 hooks); and `scripts/ad_library.py`, which pulls any category's live ads so every run looks first. - **Three ways to make each ad:** real footage with a shot list, image and video models (GPT Image 2.6, Seedance 2.5, Higgsfield Marketing Studio for presenter video), and code (exact text set over photos, captions and cuts with HyperFrames or Remotion). Video now means cutting real or generated footage, not animating text. AI people can present and demonstrate, never pose as customers. - **The taste loop.** When you edit a skill's work, it saves a gold case and one to three yes/no checks in your brand folder, and every later run answers those checks before handing work over. `/performance-review` saves clear winners as gold cases too. - `site-audit`: a free, local technical SEO and AI-readiness audit. A polite crawl (robots.txt honored, about 4 requests at a time, 200 pages by default and 1,000 at most) checks indexing blockers, noindex and canonical mistakes, AI crawlers blocked in robots.txt or at the CDN (the homepage fetched as a browser and as each AI crawler), sitemaps parsed for real, broken links, redirect chains, duplicate and near-copy pages, pages built by JavaScript, Core Web Vitals from Google's PageSpeed and CrUX APIs, and evidence-backed content signals (statistics, quotations, dates, authors, short answers under question headings). The result is 5 to 10 ranked fixes in plain words, with the full findings saved, a quick mode and a re-audit that shows what changed. Python standard library only; Firecrawl re-reads JavaScript-built pages when you have a key. - `start-here` routes technical requests ("crawl my site", "are AI crawlers blocked", "page speed") to `/site-audit`; its marketing review keeps "review my site" and "audit". - `email-sequences` gains a cold outbound mode: signal-based openers, emails under 80 words, a 4 to 5 touch sequence, send and stop rules, compliance. ### Upgraded - Every skill hands over finished work first: no script paths, self-scores or bookkeeping in what the client sees, one short Sources and assumptions note at the end, and length budgets by job type. - `positioning-angles` and `direct-response-copy` add a claim gate: shipped copy uses only facts from the brief, the site, reviews or research, and stronger ideas go to a "To make this true" list. Positioning now decides and ships the first screen, the next block, objection answers and fit. - `local-seo-plan` always delivers the full page list (URL, title, week), a field-by-field profile checklist, a 13-week calendar and ready-to-use pieces (homepage first screen, profile description, review replies, review ask). - `keyword-research`, `ai-search`, `seo-content`, `programmatic-seo`, `service-area-pages` and `local-seo-plan` fold in topic plays, coverage maps and gap-aware receipts. - `social-content` writes every post finished, and ties a service seller's posting to an offer and a daily routine. - `paid-ads` follows the new test order and batch rule, with a small-budget exception for clean angle tests. ### Tests - `tests/test_site_audit.py` covers robots rules for AI crawlers, sitemap parsing, page extraction, the noindex-in-sitemap finding, content signals and a crawl of a local fixture site. ## 4.0.0 (2026-11-16) Free marketing skills for your AI agent: research, positioning, copy, pages, SEO, AI search, ads, email and content, plus a local-business pack. 31 skills (19 carried from 3.1.1, 12 new). MIT licensed. ### Renamed - The library is now **Boring Marketing Skills**: plugin `boring-marketing`, frontmatter key `metadata.boring`. Install with `npx skills add boringmarketer/boring-marketing-skills` (any agent) or `/plugin install boring-marketing@boring-marketing-skills` (Claude Code). ### New: the local-business pack (6 skills) - `google-business-profile`: checks the Google profile against the top 3 in the map pack and writes a month of posts. - `local-seo-plan`: decides which services and towns get a page, a section, or nothing. - `service-area-pages`: town and service pages that aren't the same page with the city swapped. - `reviews`: where reviews stall, the owner replies, and the ask after every job. - `first-screen-check`: the site on a phone the way a customer sees it (number, promise and proof on the first screen). - `speed-to-lead`: missed calls and slow replies, then the text-back and follow-up. ### New: ads, agency and long-form (6 skills) - `google-ads-tracking`: Google Ads counting calls and form fills, so it bids on real leads. - `ads-incrementality-check`: a pause test against the phone log, to see whether ads add calls or only take credit. - `meta-ads`: Meta API setup, campaign launch and proof the tracking fires. - `find-local-clients` and `pitch-the-fix`: the agency path. Pick a trade and a town, rank the businesses losing the most customers, then write a short email about one broken thing with a fix you can sell. - `long-form-article`: a long guide with real numbers and named sources, under any byline. ### Upgraded - `start-here`: a **lite first run** is now the default. It reads the site and what customers say (Google reviews for a local business, review sites for software and shops) in about 3 minutes, writes the core brand files and names one first move. The four-agent foundation (about 15 minutes) is opt-in: say "deep". Routing covers all 31 skills in plain words, with the local pack and the agency path ("new client", "find clients") as branches. It still works for any business. - `positioning-angles`, `competitive-intel`, `ai-search`, `keyword-research`, `landing-page`, `direct-response-copy`, `lead-magnet`, `paid-ads`, `email-sequences`, `performance-review`, `social-content`, `programmatic-seo`: sharper methods and local worked examples (a sample business, Hartline Plumbing in Dallas), with the free path first in every one. - `creative`: the model registry and routing re-verified on 2026-11-13. Sora 2 is gone (its API ended 2026-09-24), so UGC video routes to Seedance 2.0 with Veo 3.1 as the alternate; GPT Image 2 got cheaper and lost transparent backgrounds; Seedance 2.0 is cheaper without video inputs; newer models are listed for the next refresh. ### Works in more agents - Skills installed with `npx skills add` (Claude Code, Codex, Cursor and others) now load the brand files. The CLI copies skill folders only (each skill to `.agents/skills//`, with `.claude/skills/` linked to it), so `_system/` never arrived and the brand-context loader silently failed outside the Claude Code plugin. The shared runtime (`brand_context.py`, `capability.py`, `route.py`) now ships inside `start-here/scripts/`, and every skill calls it as a sibling. If python3 or the script is missing, a skill reads the brand folder directly. - Every skill now sees the business facts in `brand.yaml` (category, services, and for a local business the phone, town, service area and rating). - `route.py` gains local goals (`found`, `chosen`, `calls`, `clients`) and `--request ""`, which ranks skills by their trigger phrases. ### Checks - New tests: trigger routing for all 31 skills (no model calls), portability across the three install layouts (plus an opt-in run of the real `npx skills` CLI), the banned-terms rules, the 31-skill catalog against the README, and strict-YAML-safe frontmatter (the `npx skills` CLI skips a skill whose description holds an unquoted `: `). - `check_banned.sh` bans private names, author paths and real-looking ad account or conversion IDs, and allows the author's own tool (boringfunnels) only on a line that says it is made by the author of this library. - `doctor.sh` checks the shared runtime copies and flags frontmatter that strict YAML parsers reject. ### Release cleanup - Shared files ship inside `start-here/references/` (output format, schemas, brand memory), so skills installed with `npx skills add` can read them. - Examples from real campaigns keep the lesson and drop the figures. Illustrative examples are labeled as such, and sample quotes are no longer attributed to real people. - Plain words in every skill: no em dashes, and no product plugs beyond the one disclosed house tool. ## Earlier versions - **3.x (2026):** brand memory shared across skills, a creative engine with a model registry, and research that never comes back empty. - **2.0 (2026) and 0.1 (2025):** the first skill sets.