Skip to content
This repository was archived by the owner on Sep 30, 2026. It is now read-only.

v2.7.0

Choose a tag to compare

@shljsl75891 shljsl75891 released this 04 Sep 17:36
· 65 commits to main since this release

Minor Changes

  • Server-side safety fallback (ported from cortexkit/anthropic-auth v1.19.0): OAuth requests for
    claude-fable-5/claude-opus-5 now opt into Anthropic's fallbacks: "default" server-side safety
    fallback, so a content-filter refusal is transparently rerouted instead of returned as an error. The
    returned fallback content block is hidden behind a signed thinking marker before being stored by
    OpenCode, and restored on the next replay before the request is signed. Only affects the OAuth fetch
    path — custom API-key routes are unaffected.

  • Assistant prefill strip (ported from cortexkit/anthropic-auth v1.19.0): trailing whitespace-only
    text after the latest assistant tool_use is removed before replay, preventing a valid tool-result
    continuation from being rejected as an unsupported assistant prefill.