Copilot Chat v0.33.2025110401: Subagent & Web Search
Subagent2 lands in plan mode, Anthropic web search gets proper citations, and experimental GPT-5 prompts are removed. Security fixes included.
TL;DR
- Subagent2 now available in plan mode for improved agent orchestration
- Web search results from Anthropic now formatted with proper citations
- Removed experimental GPT-5 prompt variant; cleaned up cloud agent registration
New
- runSubagent2 in plan mode — Enables subagent2 execution within planning workflows for better multi-step reasoning
- Anthropic web-search formatting — Web search results now include proper citation formatting and encrypted index data for better source tracking
- Max token setting for edit window — New configuration option to enforce token limits on edit operations
- CCA sessions polling — Improved Copilot Cloud Agent session handling with polling support
Fixed
- Command precondition handling — Fixed case where commands exist but don't meet execution preconditions
- CLI tool invocation formatting — Improved formatting for command-line tool invocations
- File read validation — Validates offset parameters and reports actual line count when out of bounds
- Language-specific settings — C++ no longer overrides language-specific settings with defaults
- Exit tool invocation — Manually invokes exit tool when LLM fails to invoke it
- Completion panel — Fixed completion panel functionality
- Security vulnerabilities — Addressed npm audit findings
Breaking Changes
- Removed experimental gpt5AlternatePrompt — This experimental feature has been removed; default prompts are now standard
- runSubagent deprecation path — New setting available to disable runSubagent in favor of runSubagent2; plan your migration accordingly
Update via VS Code extensions marketplace or run: code --install-extension ms-copilot.copilot-chat
Source: Copilot Chat
