Repository navigation
What's Changed
Stable release since v1.5.2, promoting v1.5.3-beta.2 (8f442161) unchanged. It rolls up 65 merged pull requests — 53 feature and fix changes, plus docs, tests and refactors — with the bulk of the work on Feishu cards and attachments, provider presets, and agent model and command handling.
Upgrading to v1.5.3 is recommended: every new option is optional and defaults to off.
How this release was validated
- Unit, regression, performance and blackbox tests in CI on this commit.
- Release QA (
go test -race, blackbox smoke) on the exact commit. - Live Feishu channel smoke (Claude Code + Codex) on the release binary.
- Promoted from v1.5.3-beta.2 after its beta period and maintainer approval.
New Features
- Feishu: Key tool parameters are shown in cards (#1944)
- Feishu: Plain text and image bursts can optionally be coalesced (#2022)
- Feishu: Lifecycle tracking through AsyncRecoverablePlatform (#1198)
- Feishu: Group reply-all is scoped to configured chats (#1976)
- Feishu: Streaming is configurable (#1956)
- Feishu: Messages that mention other users are skipped (#2012)
- Cursor: Falls back to local chat history (#2255)
- Cursor: /memory is persisted in the project AGENTS.md (#2245)
- Cursor: An optional instructions file is reloaded on every turn (#2095)
- Claude Code: The CLI's applied effort is shown in the reply footer (#2252)
- Codex: Named app-server permission profiles are supported (#2151)
- ACP: A compact command, context usage and form elicitation bridge (#1933)
- ACP: Model discovery and live switching via configOptions (#1935)
- OpenCode: Skill directories are exposed via SkillProvider (#1873)
- Telegram: Dice messages are forwarded to the agent as text (#2254)
- Telegram: Telegram Bot API 10.x native rich messages are supported (#1985)
- WeCom: Interactive card sending and callback feedback (#1645)
- Slack: An optional base_url overrides the Slack Web API root (#2178)
- MiniMax: M3.1 Flash is the default model (#1192)
- Providers: Runcrate provider preset (#2144)
- Providers: Opper provider preset (#2125)
- Providers: CloudGPU provider preset (#2166)
Fixed
Core and sessions
- HTML tags written inside inline code are escaped (#2282)
- Splitting a message reserves its closing code fence (#2117)
Claude Code
- Session transcript lines over 256KB are read in /list and /history (#2192)
- The /model fallback list includes claude-fable-5 and follows the rolling fable alias (#1755, #1642)
Feishu
- Askq and permission card clicks are accepted only from the requester (#2281)
- The card answer is no longer echoed as a second chat message (#2290)
- Replies to bot menu events use the correct receive_id (#2259)
- Reply cards are filled and table columns are sized by content (#2235)
- Forwarded attachments are downloaded through the outer message (#2239)
- Post attachments from quotes, merge-forwards and failed downloads are surfaced (#1923)
- URLs in quoted interactive cards are preserved (#2142)
- The chat scope is checked before allow_from when a message arrives (#2223)
- Name lookups are bounded, so a hung request cannot hold a message (#2018)
- Silent WebSocket connections are recovered (#2120)
- The SDK logger respects the application log level (#2150)
- Progress card panels stay collapsed while a turn is running (#2199)
Other platforms
- Discord: AskUserQuestion option clicks are forwarded to the engine (#2219)
- MAX: Only private-dialog chats are cached for callback routing (#2284)
- Slack: base_url validation is tightened, a base_url secret is never echoed in a validation error, and base_url is displayed the same way in errors and logs (#2247, #2242, #2234)
- Slack: Edited @mentions are ignored instead of re-running the agent (#1579)
- WPS Xiezuo: style_content is parsed in rich text messages (#2198)
- WeCom: The required task_id is sent on button_interaction cards (#2283)
- WeCom: Quoted files are named and unavailable attachments are marked (#2225)
Other agents
- Cursor: Invented models are no longer listed when CLI discovery fails (#2094)
- Pi: Custom providers are merged into the model list and reasoning efforts are model-aware (#1872)
Daemon, config and install
- Config: Enabling multi-workspace mode from the dashboard no longer breaks startup, closing the issue reported in v1.5.2 (#2266)
- Daemon: launchd owns the restart transaction (#2286)
Other Changes
- i18n: Russian translations read more clearly (#2101)
Behavior changes worth knowing
- MiniMax default model (#1192): M3.1 Flash is now the default MiniMax model. Pass a model explicitly if you relied on the previous default.
- Feishu group reply-all (#1976): reply-all now only applies to the chats you configure; other group chats are not answered.
- Feishu mentions of other users (#2012): messages that mention another user are skipped instead of being answered.
- Feishu card clicks (#2281): askq and permission cards can only be clicked by the user who raised the request.
- Slack edited mentions (#1579): editing an @mention no longer re-runs the agent.
Breaking Changes
None. All new options are optional with safe defaults.
Contributors
Thanks to the 35 external contributors who landed 62 changes in this release, and to the cc-connect team. 🎉
@qvictl @ClaudieShannon @rumata-estor @openviking-superagent @daysai @Niko-Wu @bujueyunjian @gpYang @luyao10282 @RC-JYDDS @timmyagentic @yanyan1115 @adao-max @AhooraZen @BoYanZh @cloudgpuapp @deepziyu @DMianZhi @Felixkw12 @happyTonakai @heydryft @jojofran @M1racleShih @Matthew-Laplace @octo-patch @rainyMorning2 @RyoKusnadi @ShengLin1001 @shexiaobin @SunflowersLwtech @TongLing916 @vuyiv @WhyTan00 @xuli500177 @Yee-chen
Full Changelog: v1.5.2...v1.5.3