Skip to content

v1.7.0 — Wave 2: cross-runtime + eval harness + Wave 2 sub-skills

Choose a tag to compare

@AgriciDaniel AgriciDaniel released this 18 May 16:18
· 66 commits to main since this release

Release walkthrough video

claude-ads 1.7.0 release walkthrough on YouTube

📺 Watch the 1:52 release walkthrough: https://youtu.be/qUXSZHiMPK8

Chapters

  • 0:00 claude-ads 1.7 is live
  • 0:07 The gap: Amazon retail media, Meta Andromeda, Google AI Max
  • 0:19 What shipped: 22 sub-skills, 8 platforms, 209 checks, 41 tests, 6 hosts
  • 0:28 How we compare: no other skill covers more than 2 platforms
  • 0:39 Cross-host install: one script, six hosts
  • 0:47 Eval harness: 41 tests, 209 checks, deterministic scoring
  • 0:57 Google AI Max pre-flight: 16 weeks to forced migration
  • 1:07 Meta Entity-ID predictor: score creative before launch
  • 1:16 Amazon Ads: Sponsored Products, Brands, Display
  • 1:24 The full sub-skill map: 22 modules, 8 categories, 1 orchestrator
  • 1:32 Roadmap: Wave 3 (Walmart, CTV), v2.0 (MMM, incrementality)
  • 1:41 Install and community

Wave 2 of the 90-day category-leader plan.

Added

  • Cross-runtime install matrixinstall.sh / install.ps1 --target={claude,codex,cursor,windsurf,gemini,goose} with whitelist validation, path-traversal blocks, no shell injection.
  • 41-test pytest eval harness — routing snapshots, bidirectional 209-check catalog coverage, scoring math determinism, 28-case SSRF regression suite. Runs in CI on every commit.
  • 3 new sub-skills:
    • /ads amazon — Sponsored Products / Brands / Display, ACOS / TACOS, search-term harvesting
    • /ads attribution — AdAttributionKit + GA4 + Consent Mode V2 + MMP health
    • /ads server-side-tracking — sGTM, CAPI Gateway, event_id dedup, PII hashing discipline
  • /ads google AI Max era rewriteai_max_setting.enable_ai_max, AI Brief, FUE, brand exclusions
  • /ads meta Andromeda + GEM + Lattice era rewrite — Entity-ID clustering detection, creative-similarity suppression awareness
  • 10-Principle Thinking Framework — load-on-demand cognitive discipline (OBSERVE × 2 / LISTEN / THINK / CONNECT × 2 / FEEL / ACCEPT / CREATE / GROW) referenced by the orchestrator + 4 thinking-heavy sub-skills
  • Outbound Network Destinations table in SECURITY.md documenting every endpoint every script can reach + its gating mechanism

Full changelog