Continue v1.2.15 Changelog

This release is all about making the AI assistant smarter and fixing a lot of the annoying glitches that slow you down. It matters most to developers who use Continue's command-line tools or rely on AI agents to automate their coding, giving them more power and fewer headaches.

What's new

• Subagents and Agent Skills: The AI can now spin up smaller "helper AIs" (subagents) to tackle specific jobs, and it has new "skills" to do complex tasks better.
• Ask Sage and OVHcloud support: You can now use Ask Sage as an AI provider, and there are new OVHcloud models available to chat with.
• Remove rule button: You can now easily delete custom rules you've set for the AI, right from the interface.
• Automatic WSL detection: If you use Windows's Linux environment (WSL), the command-line tool will automatically detect it and open the correct shell.

Improvements

• Better usage stats: The /info command was cleaned up to give you clearer statistics on how you're using the tool.
• Headless mode upgrades: When running without a visual interface (headless mode), the system now allows Bash and MCP (Model Context Protocol) tools by default, making it easier to automate tasks.
• Smarter error handling: If the AI gets an "overloaded" error from too much traffic, it will automatically try to resubmit your request instead of just failing.
• Security and dependency bumps: A ton of background code libraries (like lodash, undici, and tar) were updated to newer, safer versions to patch security holes and improve stability.

Fixes

• Dark mode text fixed: Fixed an issue where the text on the local setup screen was black on a dark background, making it impossible to read.
• Conversation compaction fixed: Squashed a bug where the AI would crash when trying to shrink down a long conversation that had unfinished tool calls.
• Windows/WSL command fix: Fixed a glitch where Windows computers connecting to WSL were incorrectly wrapping commands in cmd.exe, which caused errors.
• Duplicate messages removed: Fixed an issue where the AI would accidentally copy and paste duplicate tool messages.
• Faster history loading: Fixed a bug that made the AI unnecessarily wait for an old session to load from your history before you could do anything else.
• Agent completion tracking: Fixed bugs that weren't properly marking an AI agent as "finished" if it ended its turn without using a tool, which could leave it hanging.
• Config loading cleanup: Fixed an error where the system would accidentally try to load .md (Markdown) files as agent configs, and simplified the error message so it actually makes sense.
• UI visual fix: Fixed the right-side gradient border so it now displays fully instead of getting cut off.

Breaking changes

No breaking changes in this release.