Orkas Files

v2026.5.11

Changes since v2026.5.8

• v2026.5.11
• Merge remote-tracking branch 'origin/dev'
• review: align chat-input-form input.dir.pick fallback string with locale
• review: drop redundant dynamic import of readState in group_chat/index.ts
• review: hoist setCodingProjectDir to static import in bus.ts
• review: translate Chinese quote in bus.ts dispatch project-scope comment
• fix: agent picker force-refreshes _agentsCache on every open
• fix: commander-created agent/skill is reachable from the same conversation
• project bindings: sort lists A→Z by display name + codify the rule in CLAUDE.md §8
• feat: project scope — collapse to agents only, validate recipient on switch, dispatch-time bindings filter
• input area: drop files to add as attachments
• agent picker: rank search results by match quality
• skill-registry: drop counter-example string from 'Available skills' ROOT warning
• feat: coding-CLI integration polish — workspace sync, cwd-change session reset, prompt template split, picker card
• feat: project scope — per-project agent/skill bindings + detail page + search grouping
• agent-creator + skill-creator: add 'Quality bar — designing the {agent,skill}' section
• fix: chip click pre-checks resource availability before opening detail view
• agent-creator + skill-creator: add 'output language follows user UI language' Hard rule
• Switch genProjectId to crypto.randomBytes(6)
• Add a Data root row to the settings page (path display + Open in file manager)
• Point markdown-prose-code header comment at the real fixture file
• Refresh stale comment in bus.ts that still referenced removed $*_dir template vars
• Drop the orphaned watchdog test left after the watchdog removal
• Add "New conversation" entry as the first item in the project ⋯ menu
• Add CLI-skip callouts to the LLM-managed-only sections of agent-creator/SKILL.md
• Remove the group-chat watchdog wake-commander mechanism
• Add Projects: logical conversation grouping + per-project workspace scope
• Move agent / skill authoring rules into two builtin skills; thin out the four chat prompts
• Replace residual Chinese examples in the upstream-dependencies hard rule
• Hard-rule plan_set upstream dependencies into {{step_N.output_*}} templates
• Tighten per-skill edit chat sandbox to read-only on the skill dir
• Split commander file-tools sandbox into read-write and read-only roots
• Surface a visible error bubble + state_changed when an agent fails to load mid-turn
• Fix tab-switch-during-stream losing live bubble and re-showing prior message
• Skip @-mentions inside > markdown blockquotes in parseMentions
• Refresh agent / skill caches on the fallback message path; honor forceRefresh in loadSkills
• Add a "no overlap -> emit same-turn" branch to the commander similarity self-check
• Cure empty conv-slug dirs left by bash tool runs (cold-prune + boot sweep)
• Remove empty chat_attachments/<cid>/ dirs after attachment delete or upload-write failure
• Add a "Clear all conversations" button to the settings page
• Honor user scroll position inside the 300px process-info box
• Fix conversation-list jitter and bottom-pinning during streaming
• Add commander pre-create similarity self-check for agents and skills
• Drop the leftover strip-agent.js and strip-agent.test.ts shells
• Let commander create / edit skills in the main conversation; flatten container parsing to arrays
• Realign three test assertions with current English prompt output
• Fix three regressions in conversation process-info rendering
• Inline ROOT path header in skill / agent-index system-prompt blocks; expose agent_id