v2026.6.7-beta.1

## 2026.6.7 ### Highlights - Channel delivery is tighter across Slack, Telegram, outbound media, silent replies, progress drafts, and paged action results: same-channel Slack finals persist in transcripts, top-level `image` message-tool sends attach media, expandable Telegram blockquotes and spooled replay survive delivery, explicit silent assistant replies stay silent, progress draft startup failures are reported, and channel action result pages can be fetched incrementally. (#92498, #92407, #92416, #92281, #92073, #92083, #88993) Thanks @TurboTheTurtle, @ichirokyoto, @xydigit-sj, @joshavant, @sallyom, @hansraj316, and @fuller-stack-dev. - Provider and model handling is more resilient: Kimi K2.7 Code is available, Kimi native tool-call ids and replayed `reasoning_content` are repaired, Mistral skips unreadable tool schemas, Fireworks catalog parameters come from manifests, DeepSeek keeps configured static transport, provider fallbacks resolve correctly, Anthropic thinking replay is repaired, and Anthropic Vertex stops re-marking transport-budgeted cache control. (#92554, #92396, #90242, #90326, #92265, #92293, #92286, #92387) Thanks @xialonglee, @vincentkoc, @obuchowski, @joshavant, and @openperf. - User-visible context and auth boundaries are safer: Feishu no longer leaks prompt-preface runtime context into replies, WebSocket payload handling is hardened, CLI-backed `/btw` fallback fails closed, local setup trust is hardened, and Skill Workshop symlink writes are gated and validated before rollback metadata is written. (#92589, #92593, #92226, #92175) Thanks @jovi2014-cyber, @zhangqueping, and @joshavant. - Agent, memory, Codex, cron, and update recovery paths preserve the useful failure now: invalid plugin model catalogs are isolated, QMD startup failures survive fallback errors, Codex memory prompts remain registered, source message tool replies no longer stop agent progress, structured unsupported-model errors are classified, heartbeat/cron terminal state is