contributions

open-source pull requests across projects i contribute to


openchamber

· OpenChamber

A desktop platform that wraps the OpenCode CLI agent into a full GUI with session management, settings, tool execution, and multi-provider support — making AI coding assistants accessible beyond the terminal.

#1079
add Behavior settings page for global AGENTS.md Built a new Behavior settings page with a textarea editor that lets users define a global AGENTS.md system prompt, synced to disk with GET/PUT API endpoints and 1MB size limits. Overrides any project-level AGENTS.md across all providers and sessions.
#1069
add concurrency controls for multi-session providers Running multiple concurrent sessions from the same provider caused slowdowns and random stops — a single OpenCode server with no concurrency controls would get killed mid-task by health checks. Added health-check gating, SSE backoff, and retry logic.
#1065
refresh provider models on OpenCode reload After restarting OpenCode, provider model lists stayed cached — newly released models and context-window changes were never picked up. Added a post-restart re-auth flow that fetches fresh model data without user interaction.
#1064
fix skills empty state centering and terminal toolbar overlap The skills empty state was vertically off-center because of missing flex height constraints. The terminal toolbar quick-keys also overlapped with dock buttons on narrow windows. Fixed both layout issues.
#1058
keep settings sidebar open, center content, fix worktree refresh The settings sidebar auto-collapsed to a narrow icon rail when navigating sub-pages like Magic Prompts — now stays fully open. Also centered settings content on wide screens and fixed stale worktree state after auth changes.
#1056
add per-agent default thinking variant setting Users had to manually switch reasoning levels in chat for each agent every time. Added a Default Thinking dropdown (matching Temperature/Top P styling) that remembers each agent's preferred variant — dynamically filtered to the selected model's capabilities.
#1000
drastically improve cold-start, bundle size, and streaming performance Cold-start: removed blocking I/O on preferences for 50–200ms faster first paint. Bundle: split bootstrapping into phased priority lanes so heavy modules load after paint. Streaming: memoized the chat export pipeline and reduced re-renders during streaming output.