v0.120.0

## What's Changed
* [feat] Run agent HTML apps from the drive by @ardaerzin in https://github.com/Agenta-AI/agenta/pull/6972
* [chore] Regenerate the Fern API client from the EE OpenAPI spec by @ardaerzin in https://github.com/Agenta-AI/agenta/pull/6981
* [docs] Specify agent HTML apps and implementation gaps by @ardaerzin in https://github.com/Agenta-AI/agenta/pull/6983
* [feat] Add xAI as a provider and make Grok work end to end by @ashrafchowdury in https://github.com/Agenta-AI/agenta/pull/7057
* [fix] Resolve HTML app release blocker and copy/grant failures by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7059
* ci: shut down Railway previews after tests and add /preview by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7058
* [chore] Bump Claude harness to Opus 5.5 and refresh the Claude model catalog by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7060
* [feat] Submit single-choice questionnaires on option click by @ashrafchowdury in https://github.com/Agenta-AI/agenta/pull/7052
* [fix] Accept Claude Fable 5.1 across the Claude catalog and harness by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7062
* fix(release): xAI key hint and Helm wiring, client base URL, safe Daytona snapshot refresh by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7063
* [fix] Add Claude Opus 5.5 to the prompt models and Anthropic defaults by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7064
* [fix] Run saved Claude Fable 5 agents on Fable 5.1 and list Opus 5.5 first by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7067
* fix(qa): register the release gate's MCP server as a gateway connection by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7068
* [fix] Log when a saved Claude Fable 5 model is upgraded to Fable 5.1 by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7070
* fix(runner): Daytona snapshot --force builds a trial before replacing the live snapshot by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7065
* [fix] Harden agent HTML apps: bridge re-attach, navigation exfiltration, grant scope, permissions by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7066
* [fix] Price Opus 5.5 and Grok 4.7 on Pi and enable their reasoning by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7072
* [fix] Pin litellm's bundled price map in the SDK and API unit tests by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7075
* [fix] Keep drive HTML preview anchor links inside the preview by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7071
* fix(frontend): keyboard only selects on a one-question questionnaire, and the card leaves composer focus alone by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7073
* [test] Drop the regexp CodeQL flags in the drive preview link test by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7077
* fix(api): drop the root-level oneOf from request_connection so OpenAI accepts it by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7078
* [fix] Run a concrete Claude model id on its tier alias when the build does not offer it by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7080
* fix(runner): hold a Daytona key rotation for the measured 30s propagation by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7081
* [fix] Strip anthropic/ from a Claude model id before setModel by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7083
* [fix] Normalize a Claude model's anthropic/ prefix once, so the gateway env gets the bare id by @mmabrouk in https://github.com/Agenta-AI/agenta/pull/7085
* [release] v0.120.0 by @github-actions[bot] in https://github.com/Agenta-AI/agenta/pull/7053


**Full Changelog**: https://github.com/Agenta-AI/agenta/compare/v0.119.1...v0.120.0