Changelog

Every release, from first audit to full operations.

The complete Claude Ads release history, currently at v2.1.0: what was added, what changed, and how the security baseline hardened along the way.

  1. v2.1.0

    Optional Cockpit, sharper routing, and fresher platform evidence

    A minor release that adds the optional Claude Ads Cockpit companion, gives every /ads command one named skill, refreshes platform evidence, hardens the installers, and rewrites the README in plain language. Every scoring profile stays disabled, so no health score is produced.

    Show highlights

    Added

    • Claude Ads Cockpit (ads-cockpit 0.1.0), optional and off by default: a local pane next to Claude Code that shows how much of an audit could be judged, a meter per platform and per topic, and what to fix first. It never writes to ad accounts. Try it with the synthetic demo: /ads-cockpit demo.

    Changed

    • Every /ads command routes to one named skill, each contested trigger phrase has one owner, and the copy, strategy, and format agents are read-only.
    • Platform references and ledgers refreshed on 12 and 16 October 2026, including Microsoft SOAP support through 31 January 2027, Google Smart Campaigns, Reddit lead forms, the LinkedIn API schedule, Apple, Meta, Pinterest, and EU AI Act timing.
    • Installers clean up temporary directories on every exit path, --ref pins a git install to an exact commit, and install.sh --source=git works on macOS.
    • A plain-language README. Plugin, marketplace, and citation metadata now describe deterministic findings and evidence coverage, because every scoring profile is disabled.
    • Pillow 12.3.0, urllib3 2.8.0, and WeasyPrint 70.0. Managed Linux installs require glibc 2.28 or later.

    Fixed

    • Google audits separate positive keywords from ad-group, campaign, shared-list, and account negatives, so negatives no longer count as enabled keywords.
    • Agents no longer pin a subagent model; the host or invocation chooses.
    • Evidence freshness is evaluated at the current date, so expired sources fail validation.
    • Vulnerable Pillow and urllib3 pins are replaced.

    Full notes

  2. v2.0.2

    Report safety, platform fixes, and installer improvements

    A patch release that improves installation guidance, report safety, platform evidence, and release validation, and adds Google Search overlap checks and a Meta cold-start contract for planning and creative work.

    Show highlights

    Added

    • Google Search and Dynamic Search Ads overlap and duplicate-keyword checks, based on account routing evidence.
    • Meta planning and creative workflows now distinguish account, Pixel, and conversion cold starts.

    Changed

    • Improved legacy-install preflight: installers detect a pre-v2.0.0 install and stop with one message before any write.
    • The README explains how to replace the stale agricidaniel-claude-ads marketplace alias kept by marketplaces added before v2.0.0.
    • Refreshed platform evidence and hardened dependency and release checks.

    Fixed

    • Microsoft native-export conversion mapping corrected to ConversionsQualified.
    • Untrusted PDF report text and captions are now escaped.
    • Failed landing-page analyses no longer produce misleading grades.
    • Generated image output is validated before it is treated as a successful artifact.
    • Cross-installer ownership handling: the Bash and PowerShell installers now reject each other's ownership manifests before any change.
    • Dependency guard regressions fixed for importer-module imports and callable aliases.

    Security

    • A dependency VEX gate binds expiring not_affected advisory dispositions to exact lock versions and fails CI on any new advisory, drift, accepted risk, or expiry.
    • A least-privilege Python CodeQL code-scanning workflow was added.
    • Ignore and release-audit coverage expanded for logs, local databases, credentials, local configuration, and token patterns in binary files.

    Full notes

  3. v2.0.1

    The public mirror release

    A documentation and metadata patch that points installation, issues, discussions, and vulnerability reporting at the public repository, so anyone can install and get support without org access.

    Show highlights

    Changed

    • README, support, security, contributing docs, issue templates, and manifests now link to the public AgriciDaniel/claude-ads repository.
    • Installers clone from the public repository by default, so remote installs work without org credentials.
    • README gained a dual-distribution note and native Claude Code plugin install commands.

    Full notes

  4. v2.0.0

    The twelve-platform operations release

    A major architecture release that grows Claude Ads from an audit tool into a full paid-media operations system: twelve platforms, a complete campaign lifecycle, versioned JSON contracts, and deterministic evidence-graded scoring.

    Added

    • Twelve dedicated platform skills, with Reddit, Pinterest, Snapchat, and X now first-class alongside the original eight.
    • Full lifecycle coverage: setup, launch, monitor, optimize, research refresh, validation, and JSON-first report rendering.
    • Versioned JSON contracts for account snapshots, run manifests, findings, and reports, plus a dependency-light CLI.
    • Deterministic category-first scoring with severity-weighted evidence coverage and explicit provisional and insufficient-evidence states.
    • A capability-gated adapter framework so live platform actions stay draft-first until each operation passes its release gates.
    • Verifier agents for research, sources, skills, and releases, backed by sanitized export fixtures and expanded routing evaluations.

    Changed

    • Prompts rebuilt around progressive disclosure, untrusted-input boundaries, and artifact-backed completion claims.
    • Unique run manifests and versioned JSON replace fixed report filenames as the system of record.
    • Optional, beta, and premium features are now scored as opportunities instead of account-health penalties.
    • README reworked around twelve-platform coverage, safety defaults, and a new animated visual system.

    Security

    • Pre-dispatch SSRF controls for all HTTP and Playwright navigation, redirects, frames, and subresources.
    • Output containment, credential redaction, prompt-injection rules, and fail-closed mutation gates.
    • Uninstall refuses namespace globs, tampered manifests, and files it does not own.

    Full notes

  5. v1.8.1

    Accuracy patch for the Wave 3 docs

    A documentation pass that brings the README in line with what v1.8.0 actually shipped and adds a limitations section.

    Show highlights

    Added

    • A Limitations section covering manual data input, catalog scope, and point-in-time 2026 benchmarks.

    Changed

    • README now headlines v1.8.0: the 300-check catalog, the regulatory-compliance agent, and the agentic-era references.
    • Test counts corrected from 41 to 59 across the README, citation file, and manifests.

    Full notes

  6. v1.8.0

    The regulatory compliance release

    Wave 3 adds 91 new platform checks, a dedicated regulatory-compliance audit agent, and primary-source research coverage of the 2026 platform landscape.

    Show highlights

    Added

    • 91 net-new checks bring the verified catalog to 300 across Google, Meta, TikTok, LinkedIn, and Microsoft.
    • A new regulatory-compliance audit agent covering the EU AI Act, 22 US state privacy laws, Privacy Sandbox retirement, and iOS 26 tracking changes.
    • Fresh platform coverage from Google Marketing Live 2026, Meta's March attribution rebuild, and TikTok World 2026.
    • A regulatory-exposure scoring band with hard deadline clocks for forced migrations and cutoffs.

    Security

    • MCP write-action governance defaults: paused-by-default, rate limits, audit-log retention, and a human approval gate for budget changes.

    Full notes

  7. v1.7.1

    The polish and fact-check release

    A docs and branding patch: a rewritten README, an animated banner, and verification of every quantified claim against primary sources.

    Show highlights

    Added

    • Animated SVG banner, 25 times smaller than the old PNG, with accessible markup.
    • README rewritten with a sample audit output, a comparison table against manual audits and commercial tools, and concrete personas.

    Changed

    • Ten quantified platform claims verified or corrected against vendor and first-party documentation.

    Full notes

  8. v1.7.0

    The cross-runtime and eval-harness release

    Wave 2 takes Claude Ads beyond Claude Code with a multi-host install matrix, adds a pytest eval harness, and ships three new sub-skills for attribution, server-side tracking, and Amazon Ads.

    Show highlights

    Added

    • Cross-runtime installs for Codex CLI, Cursor, Windsurf, Gemini CLI, and Goose alongside Claude Code.
    • A 41-test eval harness covering routing, the check catalog, scoring math, and SSRF regressions, wired into CI.
    • Three new sub-skills: cross-platform attribution, server-side tracking, and Amazon Ads.

    Changed

    • Deep rewrites of the Google skill for the AI Max era and the Meta skill for the Andromeda, GEM, and Lattice stack.

    Full notes

  9. v1.6.0

    The discoverability and hardening wave

    A housekeeping release that widens how the skills are triggered, refreshes benchmark citations, and tightens the security baseline.

    Show highlights

    Changed

    • Trigger phrases expanded across 11 sub-skills so audits activate on the terms practitioners actually use.
    • Benchmark citations re-verified as current, and retired Privacy Sandbox guidance replaced with supported attribution paths.

    Security

    • IPv6 unspecified-address entry added to the SSRF blocklist, lifting the independent security score to 92/100.

    Full notes

  10. v1.5.1

    The security hardening patch

    A focused security release that locks down every URL-handling script and hardens the CI pipeline.

    Show highlights

    Changed

    • GitHub Actions pinned to full SHA hashes, CI scoped to least privilege, and dependency vulnerability scanning added.

    Security

    • Shared SSRF validation blocks private IPv4 and IPv6 ranges, with DNS failures rejecting the URL instead of passing through.
    • Error messages now strip API keys, tokens, and passwords before display.
    • Image generation gained extension allowlists, batch and dimension limits, and HTTPS-only response validation.

    Full notes

  11. v1.4.0

    The creative generation release

    Image generation moves to banana-claude with brand presets, proven copy frameworks, and an automatic quality gate on every asset.

    Show highlights

    Added

    • banana-claude becomes the default image provider, with brand presets built from your brand profile.
    • Six ad copy frameworks (AIDA, PAS, BAB, and more) with platform-specific templates and character counts.
    • Three A/B variants per creative brief and a vision-based quality gate that regenerates low-scoring images.
    • An e-commerce creative playbook covering launch, sale, seasonal, retargeting, and awareness campaigns.

    Full notes

  12. v1.3.0

    The accuracy and community release

    Community-sourced fixes cut Google audit false positives, and the plugin becomes discoverable through the marketplace manifest.

    Show highlights

    Added

    • Marketplace manifest for plugin discoverability and updates.
    • Optional Google Ads MCP integration for live API-connected audits, with fallback to manual exports.
    • Validation gates in six skills that check data sufficiency before recommendations are made.

    Changed

    • Google audit accuracy fixes that cut false positives across the check catalog, including a roughly 18 percent reduction in keyword coherence scoring.

    Full notes

  13. v1.2.0

    The Apple Search Ads release

    Apple Search Ads joins the platform lineup, and audits now start by asking about your business instead of assuming generic averages.

    Show highlights

    Added

    • Apple Search Ads sub-skill with 35 checks spanning campaign structure, bids, Custom Product Pages, and MMP attribution.
    • Context Intake step: audits now ask for your industry, spend, goal, and platforms before benchmarking.
    • README FAQ and a clear explanation of the manual data-input model.

    Changed

    • Windows installer and uninstaller fixed for PowerShell 5.1 and non-UTF-8 systems.

    Full notes

  14. v1.1.1

    Threshold and template fixes

    A same-day patch that resolves ambiguous audit thresholds and aligns the industry templates with the check catalog.

    Show highlights

    Changed

    • Meta frequency thresholds clarified between ad set level and campaign level checks.
    • Industry template budgets and asset counts aligned with the corresponding audit checks.

    Full notes

  15. v1.1.0

    The first accuracy pass

    Launch-day corrections: the audit count fixed to 190, benchmark typos resolved, and Windows given full install and uninstall parity.

    Show highlights

    Added

    • Windows uninstaller for full parity with Unix.
    • Research context notes on recent platform changes across the Google, Meta, and LinkedIn audit references.

    Changed

    • Audit check count corrected to 190 and TikTok budget thresholds aligned to the authoritative checklist.

    Full notes

  16. v1.0.0

    The launch release

    The first public version of Claude Ads: a paid-media audit system for Claude Code with six supported ad platforms and 190 checks.

    Show highlights

    Added

    • The /ads orchestrator with 12 sub-skills covering audits, platforms, creative, landing pages, budget, planning, and competitors.
    • Six parallel audit agents and 190 checks across Google, Meta, LinkedIn, TikTok, and Microsoft.
    • A 0-100 Ads Health Score with weighted severity scoring.
    • Eleven industry templates plus installers for macOS, Linux, and Windows.

    Full notes

Keep exploring

What is new in v2The full story behind the twelve-platform operations release, on the blog.Install Claude AdsGet the current version running in Claude Code in a few minutes.GitHub releasesTagged releases, checksums, and the raw notes straight from the repository.