In the wild west of autonomous marketing agents, where LLMs hallucinate features and tracking links break silently, a new tool called Omni Verify is introducing a hard stop. Developed by Daino Jung, the project gates AI-generated social media posts behind Kane CLI, a browser automation agent that verifies every promise in the copy against the live product before publication is allowed. The core premise is simple: if the AI can’t prove the link works and the feature exists in a real browser, the post doesn’t ship.

The Attribution Lie and Product Drift

The problem Omni Verify solves is twofold: broken attribution and product drift. Jung cites a specific failure case with his app Scribo, where hard-coded tracking tags on App Store buttons overwrote campaign-specific tags. This meant installs from Threads posts were incorrectly credited to the website, causing the AI agent to falsely conclude that Threads was an ineffective channel and stop posting there. Additionally, when Scribo pivoted from a handwriting vocabulary app to a content learning app in October 2024, old posts continued to promise features that no longer existed, leading to user churn. Omni Verify’s audit mode catches these discrepancies by re-verifying old posts against the current site structure.

Kane as the Truth Engine

The architecture relies on Kane CLI (version 0.10.0-rc.3) to act as the ground truth. When an AI agent drafts a post, Omni Verify converts each claim into a specific assertion—such as param_preserved for tracking links or text_visible for feature descriptions. Kane then runs these checks in a headless browser. If Kane finds that the link https://scribo.cc?ct=threads-oct24 loses its campaign tag upon clicking the App Store button, the verification fails. The AI agent is then forced to repair the link using only URLs Kane observed, ensuring the fix is grounded in reality rather than hallucination. This loop repeats until the campaign passes or hits a retry limit.

Memory and Guardrails for Future Drafts

Beyond immediate verification, Omni Verify feeds Kane’s verdicts into the AI’s memory. Confirmed facts are stored, while retired promises—features that existed previously but are no longer on the site—are marked as excluded. This prevents the drafting agent from recycling outdated marketing language. In a demo, the system successfully blocked a draft that promised a 'handwriting vocabulary notebook' after the product had pivoted, forcing the agent to rewrite the copy to match the current 'content learning' positioning. The system distinguishes between verified facts and unverified claims, refusing to publish if the latter outweighs the former.

Key Takeaways

  • Hard Gating: AI agents cannot publish marketing copy until a browser agent (Kane) verifies every claim against the live site.
  • Silent Failure Detection: The tool catches broken attribution tags and product drift that standard analytics miss.
  • Memory Integration: Failed verifications update the AI’s memory, preventing it from repeating retired promises in future drafts.
  • Repair Loop: The AI must fix errors using data observed by the browser agent, not by guessing.

The Bottom Line

This is the accountability layer AI marketing agents have been missing. By forcing LLMs to prove their claims in a real browser before posting, Omni Verify turns hallucination into a build error rather than a failed campaign.