Skip to content

CHANGELOG.md

Latest commit

 

History

History
79 lines (65 loc) · 5.75 KB

CHANGELOG.md

File metadata and controls

79 lines (65 loc) · 5.75 KB

Changelog

All notable changes to claude-ads are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[1.2.0] - 2026-03-12

Added

  • Apple Search Ads sub-skill (/ads apple): 35 checks across campaign structure (BOFU/MOFU/Search Match), bid health (CPT vs install rate, CPA Goals), Creative Sets (Custom Product Pages), MMP attribution (AppsFlyer/Adjust/SKAdNetwork), budget pacing, TAP placement coverage (Today/Search/Product Pages), and goal CPA benchmarks by app category and country tier
  • Context Intake step in orchestrator: Claude now asks for industry, monthly ad spend, primary goal, and active platforms before any audit — ensures benchmarks and recommendations match the user's actual situation instead of defaulting to generic industry averages
  • Google Ads MCP reference in README: links to google-ads-mcp for users who want live API-connected audits
  • FAQ section in README: addresses top community questions (API login, benchmark accuracy, manual ad posting, budget context, platform support)
  • "How It Analyzes Your Ads" section in README: clearly explains manual data input model and data export workflow

Fixed

  • install.ps1: PowerShell 5.1 crash on git clone — git progress writes to stderr which PS 5.1 treated as a terminating error under $ErrorActionPreference = "Stop". Fixed by temporarily setting Continue around clone call and using 2>&1 | Out-Null
  • uninstall.ps1: Parse failure on non-UTF-8-BOM systems — Unicode → and ✓ characters in double-quoted strings caused TerminatorExpectedAtEndOfString. Replaced with ASCII equivalents
  • ads-google/SKILL.md: Negative keyword guidance now enforces Exact Match [kw] and Phrase Match "kw" types by default — never Broad Match negatives. Negatives must be sourced from Search Terms Report data and grouped into themed Shared Lists. Includes over-blocking review step
  • ads/SKILL.md: Removed unsupported allowed-tools frontmatter field per Anthropic skill spec
  • ads/SKILL.md: Added apple to argument-hint subcommand list
  • Install scripts: Updated sub-skill count from 12 → 13 to reflect new ads-apple addition

[1.1.1] - 2026-02-11

Fixed

  • M-CR2 vs M37 frequency threshold ambiguity: clarified M-CR2 is ad set level (<3.0) and M37 is campaign level (<4.0)
  • Ecommerce template PMax image count aligned to G31 audit check (15 → 20 images per asset group)
  • Real estate template budget percentages widened to bracket 100% (was 90-105%, now 80-110%)
  • Info products template TikTok allocation note: added minimum $50/day campaign budget caveat
  • Duplicate step numbering in ads-tiktok (two step 7s) and ads-creative (two step 6s)

Added

  • argument-hint field on orchestrator skill for CLI subcommand hints

[1.1.0] - 2026-02-11

Fixed

  • Audit check count corrected from 186 to 190 (actual total: Google 74 + Meta 46 + LinkedIn 25 + TikTok 25 + Microsoft 20)
  • TikTok budget sufficiency threshold aligned to authoritative checklist (Pass ≥50x CPA, Warning 20-49x, Fail <20x)
  • Benchmarks typo: Local Services CPC $7.85-$15-$30 → $7.85-$15.00
  • Call Campaigns context note: clarified creation vs serving deadlines (Feb 2026 / Feb 2027)
  • Flexible Ads context note: corrected launch date from 2025 to 2024
  • Scoring system weighting rationale: corrected "20-25%" to "25-30%" to match actual platform weights
  • G59 mobile speed: LCP now measured on mobile viewport (375x812) instead of desktop
  • G61 schema check: validates Product/FAQ/Service types per audit reference (not any schema)
  • Removed unused beautifulsoup4 and lxml from requirements.txt

Added

  • uninstall.ps1 for Windows parity (Unix already had uninstall.sh)
  • .gitattributes to fix GitHub language detection (Markdown, not PowerShell)
  • Research context notes in google-audit.md (ECPC deprecation, Call Campaigns sunset, Power Pack, AI Max)
  • Research context notes in meta-audit.md (detailed targeting removal, Flexible Ads, Financial Products SAC)
  • Research context notes in linkedin-audit.md (Connected TV, BrandLink, Live Event Ads, Accelerate campaigns)
  • Weighting rationale section in scoring-system.md explaining grading band design
  • Scoring system reference added to ads-tiktok and ads-creative process steps
  • Missing .gitignore patterns for creative, landing, budget, and competitor reports

Changed

  • Removed non-spec color field from all 6 agent frontmatter files
  • Agent frontmatter now uses only official Claude Code spec fields (name, description, model, maxTurns, tools)

[1.0.0] - 2026-02-11

Added

  • Main orchestrator skill (/ads) with industry detection and quality gates
  • 12 sub-skills: audit, google, meta, youtube, linkedin, tiktok, microsoft, creative, landing, budget, plan, competitor
  • 6 parallel audit agents: audit-google, audit-meta, audit-creative, audit-tracking, audit-budget, audit-compliance
  • 12 reference files with 2026 benchmarks, bidding decision trees, platform specs, compliance requirements
  • 11 industry templates: saas, ecommerce, local-service, b2b-enterprise, info-products, mobile-app, real-estate, healthcare, finance, agency, generic
  • 190 audit checks across all platforms (Google 74, Meta 46, LinkedIn 25, TikTok 25, Microsoft 20)
  • Ads Health Score (0-100) with weighted severity scoring
  • install.sh and uninstall.sh for Unix/macOS/Linux
  • install.ps1 for Windows PowerShell
  • Agent frontmatter uses model sonnet, maxTurns 20, with example blocks
  • Sub-skills set user-invocable false to avoid menu clutter
  • Reference files follow RAG pattern (loaded on-demand per analysis)
  • Quality gates: Broad Match safety, 3x Kill Rule, budget sufficiency, learning phase protection