$ the-wire · showcase
Claude Code adds Code Mode and plugin version gating, Codex ships 0.168.0, OpenCode takes client certs for mTLS gateways
By RepoJournal · Filed · About Agentic Coding · Composed from the cited sources · methodology
The day's changes to agentic CLIs are mostly adoption decisions: an experimental tool-calling mode and version-pinned plugins in Claude Code, provider browsing and an optional workspace pane in Codex, and mutual TLS support in OpenCode so corporate gateways stop rejecting every request.
feat(core): present a provider's client certificate on model requests anomalyco/opencode
OpenCode now accepts a per-provider settings.tls block, so it can present a client certificate that mutual-TLS model gateways previously rejected at the handshake. If your company routes model traffic through such a gateway, this makes it usable as an OpenAI-compatible provider instead of a hard stop.
refactor(ai): remove Gemini Veo and Runway video protocols anomalyco/opencode
Gemini Veo and Runway video protocols are gone from @opencode/ai; the author cites Google shutting down Veo on the Gemini API on 2026-10-22. Any workflow still configured against google-video or runway-video needs a different path before the next upgrade.
v2.1.323 anthropics/claude-code
by claude-code-publisher[bot]
Claude Code v2.1.323 adds Code Mode, an experimental opt-in where Claude writes JavaScript to call its tools, toggled per user or per organization via the documented setting. It also enforces engines["claude-code"] in plugin.json: a plugin outside its declared version range is not installed, updated or loaded, and installs pick the newest version that fits.
0.168.0 openai/codex
by github-actions[bot]
Codex 0.168.0 lets you browse model providers under Sessions to Model providers and set the default for new chats, and adds an optional side-conversation pane behind features.tui_workspace. That pane resizes by dragging the divider or zooms with leader z. Audio-capable models can also take attached recordings.
fix(core,cli): report accurate output token limits in models list google-gemini/gemini-cli
gemini models and gemini --list-models used to print a hardcoded 8,192 output token limit for every model; known Gemini 2.5 and Gemini 3 models now report 65,536, and unknown or custom models omit the figure rather than show a wrong one. Budget math built on that number was overstated for current models.
fix(usage): store token counts as i64 aaif-goose/goose
Alongside the week's dependency churn, goose fixed a decode failure that surfaced only on huge sessions: token counts accumulated as 64-bit but were held as i32, so once a session passed roughly 2.1B tokens, session/list broke for any page containing it and loading it reported "Session not found." Counts are now i64 across the usage types, ACP notifications, SDK records and CLI JSON.