$ the-wire · showcase
Claude Code reads AGENTS.md, Gemini CLI stops crashing on cancel
By RepoJournal · Filed · About Agentic Coding · Composed from the cited sources · methodology
Claude Code v2.1.277 lets projects drop CLAUDE.md for AGENTS.md and adds a gateway egress-boundary switch, while Gemini CLI and Cline fix the friction of daily interactive use.
v2.1.277 anthropics/claude-code
In a project with no CLAUDE.md, Claude Code now reads AGENTS.md instead, toggled under "Project instructions" in /config. Not yet on Bedrock, Vertex or Foundry; the same release adds CLAUDE_GATEWAY_PROXY_IS_EGRESS_BOUNDARY=1 and an optional headers: map on gateway upstreams.
fix(cli): suppress uncaught AbortError logs during request cancellation google-gemini/gemini-cli
On Node 23 and later, cancelling an active query could throw AbortError from inside EventTarget listeners, bypass the unhandledRejection handlers and hard-crash the CLI. The fix swallows that synchronous abort so Ctrl-C ends the request and leaves you a working session.
fix(vscode-ide-companion): preserve terminal focus when closing diff tabs google-gemini/gemini-cli
The VS Code companion now passes preserveFocus when closing a diff tab, so dismissing the preview leaves your keyboard focus in the integrated terminal instead of jumping to the editor group. Small change, felt on every diff review in the IDE.
feat(providers): add Z.AI Coding Plan with streaming tool calls aaif-goose/goose
goose Desktop and CLI add a zai_coding_plan provider pointed at https://api.z.ai/api/coding/paas/v4, configured with ZAI_CODING_PLAN_API_KEY and discovering models from the Coding Plan /models endpoint. Streaming requests enable tool_stream for both fallback and discovered models; the existing Anthropic-compatible zai provider is untouched.
fix(providers): drive pre-key model lists from the canonical registry aaif-goose/goose
The provider picker before a key is entered now reads from the canonical registry instead of hand-maintained KNOWN_MODELS lists, which had drifted for Anthropic and left OpenAI's gpt-4 missing. Model names and context limits you see during setup should match what the provider actually serves.