| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| Orkas 2026.5.8 source code.tar.gz | 2026-05-08 | 5.8 MB | |
| Orkas 2026.5.8 source code.zip | 2026-05-08 | 6.1 MB | |
| README.md | 2026-05-08 | 6.2 kB | |
| Totals: 3 Items | 11.9 MB | 0 | |
v2026.5.8
Changes since v2026.4.30
- v2026.5.8
- Merge remote-tracking branch 'origin/dev'
- Merge branch 'dev' of https://github.com/Orkas-AI/Orkas into dev
- Strip telemetry click + CN fallbacks introduced by quote-reply
- Read both 'text' and 'content' message shapes in chat search indexer; bump schema to force rebuild
- Tier external provider max_tokens by model id instead of capping at 8192
- Align skill detail page chrome with agent detail (source tag, inline name edit, summary hide, Done validation)
- Drop English-anchoring template from chat_skill_setup External dependencies; add audit-as-artifact prompt-md rule
- Translate comments, docstrings, error strings, and renderer fallbacks to English
- Replace streaming <<<skill-file>>> blocks with a path-aware placeholder
- Realign tests with current spec for catalog, auth ordering, and session healing
- Translate Chinese prompt fragments, jsonl markers, and UI fallbacks to English
- Funnel @<name> mention construction through router.buildMention
- Make session.compact() preserve tool_use/tool_result pairs at the cut
- Treat schema defaults as 'filled' in agent inputs_schema decision
- Treat CJK characters as bare-URL terminators in autolink regex
- Add quote-reply button on group chat bubbles
- Tighten chat_shared_rules so chip row matches the bubble's file claims
- Loosen agent workflow rule to allow prose-only reasoning bullets
- Atomically strip agent/agent-input-form/agent-input-submission containers
- Tighten inputs_schema default extraction in chat_agent_in_group prompt
- Unify conversation title rule across the three creation entry points
- Persist agent-input-form draft in memory across conversation switches
- Switch agent-input-form select fields to the shared _aiSelectMount widget
- Preserve streamed text and process info when a turn errors out
- Self-heal orphan tool_result blocks in persistent-session
- Treat unclosed inline backtick as code-to-EOF in streaming markdown splitter
- Sort conversation sidebar by true activity instant, not by ISO string
- Pin chat-bubble-actions to the start of the message action row
- Unify drag-and-drop reorder across the three Key lists in settings
- docs(README): extend H1 with conversation tagline and add light star CTAs
- docs(README): extend H1 with conversation tagline and add light star CTAs
- Open agent roots in main-chat file-tools sandbox so commander can read agent.json
- Fix two interruptive UI behaviors in agent editing
- Tighten prose-language directive and detail agent name charset error
- Sort agent picker by pinyin first letter and clamp desc to 2 lines, localize injected project_dir input label
- Rewrite default descriptions of external CLI agents in three-part style
- Raise the quality floor in commander create / edit-agent prompts
- Pending-queue panel makes room for the workspace + plan-rail chips on the shoulders
- Strip <agent> container only on prose segments
- Sort conversation list dynamically by latest message updatedAt
- Created-agent chip click jumps straight to detail view
- Stop stripping spaces from auto-prepended @<name> after binding an agent
- Forbid write_file dumping the <agent> container in agent-create prompts
- Constrain agent name charset so @-mentions don't truncate
- Edit existing agents from main chat: <agent> container gains optional <agent_id> sub-tag for create-vs-edit branching
- External-agent revamp: split create / external tabs, dedupe names, project-dir form input
- Paste image / file directly into chat input; thumbnail lands on top of the box
- Streaming bubble loading view tweaks: keep dots while body streams; hide stray dashed rule
- Read field.default on agent input form when no submission has happened
- Prune orphaned workspace-sourced file_cache entries on workspace switch
- Test sync after contexts_tmp retirement: stop asserting it gets created
- Polish conv-workspace side effects: CLI session resume / empty conv-dir / lazy mkdir
- Visual plan rail in main chat + retry / skip / abandon for failed steps + auto-recover transient network errors
- A batch of UX / rule fixes: KB name length cap, default archive name budget, agent name follows UI language, produced chip icons, IME guard
- Per-conversation workspace subdirectories to avoid -N suffix accumulation
- Refactor global search: drop edit-chat indexes, add agent / skill body search
- Loosen group_chat watchdog: silence threshold 10 -> 30 min, trigger on any unfinished plan
- Rename bubble action label from "Archive" / "存档" to "Knowledge" / "知识库"
- Add edit_file tool (in-place old_string -> new_string replacement)
- Bump pi-ai to 0.73.0; refresh GPT-5.x model registry
- docs(README): refine Why Orkas comparison and trim SEO header
- docs(README): refine Why Orkas comparison and trim SEO header
- docs(README): add SEO-focused header with multi-agent keywords
- docs(README): add SEO-focused header with multi-agent keywords
- fix(agents): vertically center the agent detail title row so the name and source label sit on one baseline
- chore(test): drop stale devtools.test.ts — module is stripped from this build
- docs/prompts: hoist runtime injections to a single trailing section to keep KV-cache prefix stable
- fix(group_chat): commander dispatch-only fallback turn now writes a visible bubble
- fix(local-agents): rewrite CLI backends against real protocols + ACP close
- fix(conversation): keep agent name labels + @-highlight in sync with the registry after a rename
- fix(agents): propagate name change into existing conversation rosters
- feat(local-agents): support local CLI agents (claude code / codex / openclaw / opencode / hermes)
- fix(group_chat): hide promote-to-agent entry once a non-commander agent joins
- fix(conversation): keep selected recipient on chat switch; in-flight interactive agent always wins the chip
- fix(skills/agents): abort in-flight edit-chat stream on exit; align SKILL.md textarea width
- fix(deepseek): drop reasoning_effort when prior assistant lacks reasoning_content
- fix(skills/agents): clear-chat must also drop core-agent session jsonl
- feat(skills/agents): unify card + detail page UI; consolidate ⋯ menu
- feat(builtin/agents): add Vibe Coder built-in agent