Claude Sonnet 5 can help organize notes from a signed-in team workspace when a compatible local agent client can reach Tabbit Browser through Tabbit CLI. Anthropic currently marks Sonnet 5 as active legacy, with claude-sonnet-5 as the API model ID; Sonnet 5.5 is the current Sonnet model. The practical question is whether your chosen client still offers Sonnet 5 and can run local browser tools. For a safe collaboration task, have it read a few approved team pages, prepare a named shared-document draft with source links, and stop at a visible review checkpoint. A Reddit user described Claude failing to read support websites; replies noted that some sites block fetches or render content with JavaScript. An open browser session can help with page access, but it cannot fix account permissions or guarantee a successful read. The Claude Sonnet 5 overview covers the model; this article is about a team draft handoff.
Key takeaways
Verify the Sonnet 5 model route in the client. Anthropic’s API ID is
claude-sonnet-5; 5.5 is the current Sonnet generation.Browser access depends on the client and tool configuration, not only on the model.
Read a limited set of approved pages and preserve source URLs beside every claim.
Create a named draft in a test or review space, then stop before save, share or publish.
Keep separate task tabs so the user can continue working in other tabs.
Separate model, browser and document permissions
| Layer | Check | What it establishes |
|---|---|---|
| Model route | Client exposes claude-sonnet-5 and the selected provider supports it | The request is routed to Sonnet 5 |
| Browser tool | Client documents browser access or loads the Tabbit skill | The agent can request browser actions through that client |
| Local connector | Tabbit CLI reaches the intended Tabbit Browser | The local task can use that browser environment |
| Workspace access | Your signed-in account can open the relevant pages and draft area | The account has the visible permissions needed for this task |
Anthropic’s Sonnet 5 model documentation lists the API ID and currently labels the model active legacy. Its deprecation table lists no retirement before June 30, 2027. The same docs identify Sonnet 5.5 as the current model. These lifecycle labels may change; recheck before configuring a long-lived workflow. Anthropic also documents browser use through the API, but using that service-side tool is a different architecture from using Tabbit CLI in a local client. Do not conflate the two.
Set up a local review task
Install Tabbit Browser and run Tabbit CLI from a local client that can select Sonnet 5 and execute local tools. Discover the installed tabbit skill using the syntax documented by your client. Pi uses /skill:tabbit; OpenCode uses @tabbit for a discovered skill. Skill discovery, provider access, launcher reachability and website sign-in are four separate checks. Run Tabbit CLI diagnose to confirm the local connection, then sign in to the team workspace yourself in that Tabbit Browser environment.
Use a named task tab such as Sonnet team brief 01. Keep unrelated tabs available for your own work. The local CLI can reuse a session in the specific Tabbit Browser environment it reaches. It does not import a login from another browser profile or give a cloud agent access to your machine. For setup details, see connecting an agent to an existing browser and the CLI versus browser MCP guide.
Before involving a real team document, use a test workspace or an editor that does not autosave. Choose a small set of source pages with clear titles and stable URLs. Tell Sonnet to extract only the facts needed for the draft, keep a source URL for each, flag disagreements, and leave gaps blank. The output should be a draft plus a short unresolved-items list.
Prompt for a named team draft
Illustrative task only. Use sample pages and a test document until you have verified your workspace’s save and sharing behavior.
Use the local Tabbit skill and Tabbit CLI from this agent session.
Open the task tab named "Sonnet team brief 01". Read only the three approved
team pages already open. Create a draft titled "Launch review notes - 10 Nov"
in the test workspace. For each bullet, include its source page URL. Separate
confirmed facts from questions, and list any conflicting notes. Do not invite
people, change permissions, save to a production folder, share, or publish.
Stop at the editor with the full draft visible and wait for my review.The expected handoff is a visible, named draft containing sourced bullets and an unresolved-items section. The reviewer checks each link, corrects sensitive or stale details, and decides whether to save. If the site autosaves on typing, use a disposable test document or have the agent return Markdown in chat instead. Never let the phrase “draft” hide a real external side effect.
Use this review table before saving
| Draft element | Human check | Stop condition |
|---|---|---|
| Page claim | Open the cited source and verify its wording and date | Link missing or source conflicts |
| Account context | Confirm the right team, workspace and document | Wrong account or audience |
| Sensitive details | Remove credentials, personal data and unnecessary identifiers | Any secret or private field appears |
| Collaboration state | Confirm the title and current draft destination | Page is a production post or sends invitations |
| Save/share action | Review the actual button and effect | Autosave or side effect is unclear |
For sensitive or high-impact information, do not ask the model to prepare a shared draft. Summarize publicly shareable details in a separate test document first. The first browser-agent task checklist helps limit the task, and browser security guidance covers permission and data boundaries.
What Tabbit Browser contributes
Tabbit Browser is the local browser surface in this workflow. Tabbit CLI lets a compatible local agent work through the intended task tabs while you continue using other tabs. It is a connector, not an Anthropic model endpoint and not a native Sonnet 5 integration. If your client runs only in a remote environment, it cannot see your local signed-in pages unless that product offers a supported handoff. Ask it for a structured outline and apply it yourself, or use an approved local client.
The handoff is: verify claude-sonnet-5 in the client → load the client’s documented Tabbit skill → confirm CLI reachability → sign in to the intended workspace → create and inspect a named draft. The Tabbit CLI complete guide covers connector setup. The agent browser tools guide explains browser-tool options; Tabbit Browser installation covers the app.
Sonnet 5 remains a documented route, but Anthropic has moved its current Sonnet model to 5.5. Choose Sonnet 5 only when your client exposes the route and you have a reason to keep it; verify the lifecycle page again before depending on it. For general model information, read the Sonnet 5 overview. For a broader model setup discussion, see Claude Sonnet 5.5 in Tabbit.
Community evidence and limits
A Reddit user reported difficulty reading Squarespace and Meta support pages with Sonnet 5.5; replies suggested site blocking or JavaScript rendering and proposed using an already-open browser page. That is one troubleshooting thread, not proof that Sonnet 5 or 5.5 works with every site. Another Reddit post describes a browser-agent benchmark with live-site blocks and warns that the judge and setup affect scores. Neither record verifies this exact shared-document recipe.
Tabbit CLI guide, connecting an agent to a signed-in browser, the first browser task checklist, CLI or browser MCP, Codex setup, browser tools for agents, agent browser security, and Tabbit Browser installation.
FAQ
Can Claude Sonnet 5 access my signed-in browser by itself?
No. A client must provide browser access. Anthropic supports browser and computer-use tools in its API, while Tabbit CLI is a separate local connector that a compatible client can run.
Is Sonnet 5 still available after Sonnet 5.5?
Anthropic currently lists Sonnet 5 as active legacy with API ID `claude-sonnet-5`; Sonnet 5.5 is the current Sonnet model. Check your provider and client picker before starting.
Does Tabbit CLI reuse my website login?
It can reach the session already open in the Tabbit Browser environment connected by the local launcher. Sign in yourself in that environment first.
How should a team use it to draft shared notes?
Ask the agent to read only the relevant team pages and prepare a named draft with source links. Review access, wording and recipients yourself before saving or sharing.
Can a remote agent see local Tabbit tabs?
Only when its product provides a supported local handoff. Otherwise use a local client or ask the remote agent to prepare an outline for a person to apply.
Should the agent save the draft automatically?
No. Stop at a visible review checkpoint first. Some editors autosave, so use a test document or a non-saving preview until you know the page behavior.