Features
Tree-structured session history with branching and sharing via GitHub gist
Mid-session provider/model switching (/model, Ctrl+L, Ctrl+P)
Minimal system prompt with customizable AGENTS.md and SYSTEM.md project instructions
Context compaction with auto-summarization and extensible strategies
Extensible via TypeScript extensions, skills, prompt templates, themes, and packages
Four operating modes: interactive, print/JSON, RPC, and SDK
Skills system for on-demand capability packages with instructions and tools
Support for 15+ AI providers and hundreds of models
Custom model and provider definitions via JSON or extensions
Session export to HTML and shareable URLs
External editor integration (Ctrl+G) with configurable editor override
Configurable output padding (outputPad) for user/assistant/thinking blocks
Progressive disclosure with prompt templates (expand via /name)
RPC mode for programmatic integration over stdin/stdout JSONL
Cross-platform: Linux, macOS, Windows, and Termux on Android
Define agents as code in YAML
Automated AI code reviews on every PR
Code generation via GitHub comments
Fix CI failures automatically
Bump dependencies and fix breakage
Daily standup automation via Slack
Implementation plans for GitHub/Linear issues
Codebase reports (daily/weekly/monthly digests)
Per-run observability with trace and thinking log
Isolated sandboxes per agent run
Multi-model support (Claude Opus, Haiku, Gemini)
Spending limits per run, day, week, month
Cron triggers for scheduled agent runs
SOC 2 Type 1 certified, zero source code retention
