AI library

Daily digest

AI digest: 2026-10-26

26 October 2026 · 8 new sources

Highlights

  • Context Mode: an MCP server and Claude Code plugin that keeps large tool output (logs, test runs, web pages) in a sandbox and returns only what the code prints. It also saves session events so work state survives compaction. Try it with /plugin marketplace add mksglu/context-mode, then /plugin install context-mode@context-mode.
  • VS Code 1.143: run one task with several models, let a judge model pick the best result or merge them (sessions.chat.compareAgents.enabled). The Copilot harness is now the default for new sessions.
  • Codex CLI 0.164 presets: /presets saves prompts, slash commands, and permission settings in ten slots, and /model takes model, effort, and speed inline.

New sources

Agents and tooling

  • Context Mode: sandboxed tool output and session memory for 17 agents (Elastic License 2.0; from today's GitHub trends).
  • rederive: an MIT skill that teaches a codebase by having you re-derive its design (npx skills add EricJujianZou/rederive; from today's GitHub trends).
  • VS Code 1.143: compare and merge results from several models, Copilot harness by default, managed plugin auto-install, an on-device model.
  • Codex CLI 0.164.0: presets, inline /model and /permissions, enterprise MCP sign-in, credential masking.
  • Claude Code release 2.1.310: Monitor tool on Bedrock, Vertex, and Foundry, gateway retries, a Windows hook fail-open fix.

Frameworks and libraries

Models

  • Kimi K3.1 Pro: Moonshot's 1M-context model, close to Opus 5.5 on Terminal-Bench (its own figures). Weights promised by 2026-11-04.

Topic pages updated

  • tooling/coding-agents: Claude Code 2.1.310, Codex 0.164, VS Code 1.143, comparing models on one task, new add-ons.
  • tooling/mcp: Context Mode, Codex MCP sign-in, and a check on the open question about stateless-spec adoption.
  • tooling/skills: rederive.
  • agents/memory: session event logs as memory (Context Mode).
  • frameworks/landscape: ADK 2.13, Pydantic AI 2.61, refreshed version table (Microsoft Agent Framework .NET 1.26, LangGraph 1.2.17).
  • models/open-weight: Kimi K3.1 Pro row and open questions.
  • practices/cost-and-caching: Gemini 3.5 Flash-Lite, 3.6 Flash, 3.1 Flash-Lite, and Kimi prices. Gap fill: the pricing page now lists 3.5 Flash-Lite and 3.6 Flash, which closes an open question.
  • models/gemini: Flash and Flash-Lite prices, open question closed.