BookmarkGuard Agent Spec

Let local AI agents tidy browser bookmarks safely and repeatably, not into chaos

Problem

Many people try to clean up years of hundreds of browser bookmarks with Codex or Claude Code, only to watch the model dump a personal payment wallet into a generic SEO folder, because bookmarks reflect human workflow, not page content. The author gave up on pure AI categorization, switched to a deterministic prompt spec for local agents, and open-sourced the pitfalls. AI-driven bookmark cleanup today is mostly disaster, rarely success.

Solution

A spec plus validation script for local coding agents that: forces a backup first, applies deterministic rules (domain, frequency, time) in batches, and only asks the AI to handle ambiguous items after user confirmation. Every operation is previewable and reversible, with a cleanup report.

Users

Developers with years of bookmark hoarding, heavy browser users, people who want local agents to work on their data without breaking it

MVP

One-click export and backup bookmarks to JSON; deterministic rule-based cleanup preview (domain/time/frequency); AI handles ambiguous items only with manual confirmation; one-click rollback; no fully-automatic reclassification

Revenue

Open-source core plus paid desktop app (auto backup, cloud sync, multi-browser); or a browser extension subscription

Why now

Local coding agents are mainstream enough that people let AI touch their data, but reversible, auditable cleanup tools are scarce and a solo dev can ship one with a prompt spec and validation script