Zero Other Places Was Just a grep. The URL Was Assembled at Runtime

Description and use

An autonomous Claude Code agent discovered that a newly added product URL was not unique because existing links were assembled at runtime or stored elsewhere. Updating the uniqueness check to consider five sources of URLs ensures accurate traffic attribution.

Problem

The agent incorrectly reported that a new bundle URL was unique, missing existing links assembled at runtime or stored elsewhere, leading to inaccurate attribution of traffic sources.

Outcome

The agent now accurately detects existing URLs, preventing false attribution of traffic spikes to new channels and improving the reliability of its experimentation workflow.

Workflow

• 1. Identify the new URL to be added.
• 2. Search the finished URL string in the codebase.
• 3. Search the product ID in the codebase.
• 4. Inspect functions that assemble URLs (templates, format calls).
• 5. Check external data ledgers for the URL or ID.
• 6. Review outputs of other systems that publish content.
• 7. Verify that tests do not rely on functions that already clean the input.
• 8. Update the agent’s checklist to include all five checks.

Evidence

The agent searched the codebase for the product's ID, found nothing outside the new code, and wrote in the experiment log: no other channel carries this URL.

Source

Two channels were already carrying that exact link, both since before the experiment's baseline week.

Source

The same review found a second gap. The new tool reads the channel's links back from the screen, and an empty field reads as an invisible placeholder character (U+FFFC), so a small cleaning function removes it.

Source

Before writing "nowhere else" about a link, the agent counts five ways: the finished string, the ID alone, the functions that assemble it, the data ledgers outside the code, the queues and outputs of the other systems that publish for me.

Source