Dsh Opencli

No. Built-in search handles anonymous retrieval; this plugin handles acting as you, with account-view data, posting, and clicking through your own sessions. The README positions them as complementary halves.

Site knowledge is precompiled into commands with failure-signature recovery tables, and models read the site knowledge card first instead of discovering selectors live each session.

Write commands route through the platform approval dialog by default, ambiguous commands are treated as writes and fail closed, and a four-mode automation switch can refuse all writes outright.

A distilled command catalog in the system prompt, first-class tools with parameter checks and approval, scheduled tasks with visible runs, and a panel for daemon health, login checks, and adapter toggles.

No, and the README is explicit about that. It does not use or need the built-in browser; it goes through the OpenCLI daemon and a BrowserBridge extension to drive the real window of whichever browser you installed the extension in, Chrome or Edge, so the login state is the same one you type a URL into by hand. You need npm i -g @jackwener/opencli, then opencli daemon restart, then the BrowserBridge extension, then dsh plugin add IKEASven69/dsh-opencli. Headless servers with no real browser are explicitly out of scope.

Scheduled tasks take a cron expression, persist to dsh.schedule so a restart does not lose them, and really execute, with three retries on failure, notification events and a visible run history. The panel also records step sequences of site and browser_* calls and replays them on demand. Automation runs in four modes: read-only, standard, autonomous and unrestricted, wired into the tools/pre-execute gate, with standard as the default.

Two cards, and the model plays the first and switches to the second on failure. Fixed commands are site knowledge precompiled for token savings with failure-signature recovery tables. The fallback is the generic browser primitives plus read-only article, links and jsonld scripts and a 25-step Playwright recipe. For a site with no adapter at all, browser_do runs an analyze, init and verify creation loop from SKILL.md, so the adapter gets written during the session.