0.160.0

• Fix context usage counting only output tokens on anthropic-compatible providers that report input usage in message_delta (e.g. Z.AI). #604
• Accept unescaped square brackets, non-ASCII and other reserved chars in workspace file URI paths. #594 #605
• Fix inflated costs for OpenAI-compatible providers (e.g. Synthetic); count usage once per request before completion and retain received usage on stream errors.
• Tool prompts: push, git config, shell search/read and sub-agents run on explicit request; rule headings are tool-scoped.
• Recover Anthropic streaming responses interrupted by transient TLS bad_record_mac failures.
• BREAKING: plugins.install now appends across config layers. Set plugins.installMode to replace beside the list to exclude inherited plugins as before.
• Add mcpToolSearch config to keep MCP tool schemas out of context until the LLM loads them with the new eca__search_tools tool.
• Fix /compact steered while the chat is running: compact with its instructions at the next turn boundary and resume the task, instead of sending it to the LLM as text. #600
• Hide GitHub Copilot models the account can't use (model_picker_enabled: false or a disabled policy) instead of listing the whole catalog, and don't fall back to models.dev when none is usable. #601
• Pick the GitHub Copilot default model from the account's catalog (versatile, then powerful, then lightweight category) instead of hardcoding gpt-5.5, which some plans don't include.
• Fix remote MCP servers whose SSE events omit the space after event: (e.g. Z.AI) hanging on initialize; bump plumcp to 0.2.3. #586