Splendid Bookmarks for AI Agents

Splendid Bookmarks for AI Agents is built for coding agents to work with your browser bookmarks. It gives an agent an API to inspect and search the tree, load an organisation plan, run a Dry Run, apply changes, verify the result and request rollback. CONNECT YOUR AGENT Direct operation requires a coding agent with browser automation access to the extension's manager page, for example through Chrome DevTools Protocol (CDP). Installing this extension does not connect your agent automatically. It does not include an AI model or contact an AI service itself. The manager automatically loads the tree when opened. Use getSession to check readiness; if loading fails, use Load tree to retry. Copy agent instructions from the top of the manager. The instructions include the manager/session identity, browser-reported family, tree capture time and scope. Optional profile labels and CDP hints remain unverified. The agent is asked to reuse a matching connection, propose changes before execution, request restart approval and preserve your foreground work. These instructions are not an enforcement boundary for external tools. Connected agents read the tree through the API; a context file attachment is not required. The entry point is window.splendidBookmarks.run(command, input). Commands include capabilities, getSession, refreshTree, getStats, getTree, search, listTrash, preparePlan, loadPlan, dryRun, apply, verify and rollback. preparePlan turns compact move proposals into the same validated plan and Dry Run before Apply. Tree and search results are paginated, with up to 500 entries per page. Follow nextCursor to collect the rest. Reloading the tree invalidates older cursors. Reads respect the manager's operation locks, and getSession reports the current mode. Tree rows include boundary and protection flags. Dry Run and verification return per-operation results, and apply can check the approved plan digest. Each batch is limited to 200 operations. Scope and goal are independent; an unavailable scope is not silently expanded to the whole profile. A separate inventory CLI is available in the project repository for Node.js 22 and CDP. It verifies the selected tab, session and complete page counts before an optional JSON export. The CLI is not installed by adding this browser extension. You can also export a context file and a ready-made prompt, ask an agent to prepare a plan, and import its JSON result without giving the agent browser access. WHAT YOU CAN DO • Inspect and search your bookmarks before deciding what to change • Ask an agent to propose moves into existing folders • Rename bookmark titles through a plan, without changing URLs or folder names • Preview, apply and verify a batch of moves or title changes, then request rollback • Find duplicate URLs and move the copies you select into a review folder • Use the same manager by hand when you do not need an agent REVIEW BEFORE APPLYING Export a snapshot, review the Dry Run and re-select the backup file to verify it before applying a plan. The API follows the same controls as the manager; it does not replace the backup-file selection step. An agent's browser access is not sandboxed by this API. Rollback is available for a batch, but it can stop if items have changed in the meantime. An exported snapshot can restore positions; reverting a title change requires batch rollback. TRASH AND PERMANENT DELETION Sending bookmarks to Trash is a reversible move. Permanent deletion is a separate, confirmed action in the manager: it requires a verified backup, checks each recorded item and stops on a mismatch. That deletion cannot be undone by this extension. Recursive folder deletion is not supported. PRIVACY AND PERMISSIONS The extension itself does not send bookmark data to the developer or an AI service. It has no network requests, analytics, telemetry or host permissions. Recovery records and your language preference stay on your device. Copy agent instructions writes the generated text to the clipboard only when clicked; it never reads the clipboard. The text includes local target information and any optional connection hints you enter. Operating-system clipboard history or sync may retain it. If you share context with an external agent or give it browser access, that tool may process bookmark titles and URLs under its own settings and privacy policy. Permissions: "bookmarks" for reading and organising bookmarks, and "storage" for the local execution journal and Trash recovery ledger. No AI subscription is bundled.