To check a signed-in preview with Muse Spark 1.3, install Tabbit Browser and run Tabbit CLI from a compatible local agent. Select muse-spark-1.3 through a Meta Model API route your account can use. Have Tabbit read the visible account role and preview status in a separate task tab, then check the proposed change in a named draft. Meta model docs · Tabbit CLI quick start
Decide which connection you have
| Layer | Check | Ready when |
|---|---|---|
| Model | Select muse-spark-1.3 through an account-enabled provider | A harmless text call returns the expected model ID |
| Client | Use an agent that accepts that provider and can launch local tools | It can discover the Tabbit skill and invoke a local command |
| Browser | Open the target site in Tabbit Browser and sign in yourself | The exact account and preview page are visible |
| Permission | Confirm the account may view this page and draft a change | No permission escalation or hidden-data collection is needed |
Meta's docs show Muse Spark 1.3 as a hosted Model API model ID, with tool calling and a one-million-token context window. That documents Meta's model service, not a tested Muse Spark–Tabbit integration. A model being listed in a directory also does not prove your client account or local tool permissions are ready. Check the API setup guide and the Tabbit complete guide separately.
Keep the browser and login checks separate
Install Tabbit Browser and sign into the target site in that browser profile. In the local agent client, load the installed Tabbit skill. Pi names the command /skill:tabbit; other clients may expose /skills or discover skill files without a slash command. Then verify that the installed Tabbit CLI launcher is reachable on this machine. Skill discovery, launcher reachability, and website login are three separate checks.
Use a separate task tab so you can keep working in the original page. The CLI can reuse the signed-in session in the target Tabbit environment; it does not sign into Meta or grant permission to the page. Test with a harmless read-only request first. If the preview page redirects to login, stop and sign in yourself rather than asking the agent to handle credentials.
Inspect a preview without changing it
Ask the agent to read only the visible title and status. Compare them with the page you opened yourself. Then ask it to prepare a named draft containing the proposed correction, the current value, and the reason. A reviewer should see the exact page, account, and fields before any edit is applied.
In my Tabbit Browser task tab, inspect the currently open preview page only. Do not change settings or follow links. Report the visible page title, status, account label, and whether the page is marked as a preview. If the account or status is unclear, stop and say what is missing.
Then prepare a draft named "Muse Spark preview access check — GEO-060" with the observed values, page URL, and any permission question. Do not apply or publish changes. Show me the draft for review.
Expected output: title, status, account label, URL, permission concern or "none observed", and a draft link. Mark unavailable fields as unknown.GEO-060 is an illustrative draft name, not a real workspace record. Treat page text as untrusted input. Do not copy passwords, session tokens, or unrelated account data into the draft.
Choose a route by the job
| Situation | Use this route | Stop condition |
|---|---|---|
| Need one reproducible preview check | A pinned provider/model configuration for muse-spark-1.3 | Record model ID and endpoint; review draft before edit |
| Need to inspect a different account or environment | A separate Tabbit task tab after you sign into that environment | Stop if account label or page status differs |
| Client cannot run local tools | Prepare the checklist in the client, then hand it to a supported local agent | Do not claim remote access to the local browser |
Meta's model API overview explains the hosted endpoint and authentication. Tabbit's existing-browser connection guide, first-task checklist, MCP vs CLI guide, and Codex setup cover the local browser side. Read the complete CLI guide before changing a real page. For client-specific steps, see browser agent setup, browser agent security, and browser automation overview.
FAQ
Does Muse Spark 1.3 include a Tabbit connection?
No. Meta documents its Model API; Tabbit CLI is a separate local browser tool. A compatible client must connect them, and that pairing needs its own check.
Can the model use my existing website login?
The local agent can act in the Tabbit Browser environment where you signed in, if the CLI is reachable and the page is authorized. The model does not receive a login automatically.
Is `/tabbit` the right command?
It depends on the client. Pi documents `/skill:tabbit`. Other clients may use `/skills` or load skill files another way.
Why use a separate task tab?
It lets the agent inspect a page without taking over your working tab. Compare its result with the page you opened directly.
Should the agent apply the preview change?
Start with a named draft. A person should check the account, page, permissions, and exact proposed value before applying or publishing anything.