Daily digest
AI digest: 2026-10-30
Highlights
- Claude Code 2.1.314 write fence: set
permissions.blockWritesOutsideWorkingDirectories: trueand Write/Edit refuse paths outside the repo in every mode, bypass included. Idle compaction in VS Code and the Desktop Code tab makes returning to a long session cheaper. - Two memory tools to try: claude-mem captures sessions and feeds summaries back
into later ones (
/plugin marketplace add thedotmack/claude-mem), and codebase-memory-mcp gives agents a local graph of code, docs, and infrastructure. - Know who is spending: Anthropic's Usage and Cost Analytics API (beta) breaks cost down by key, model, workspace, and user, in buckets as fine as one minute.
New sources
Agents and tooling
- Claude Code release 2.1.314: write fence for every permission mode, idle compaction, hook and headless fixes.
- claude-mem: automatic memory across sessions for Claude Code, Codex, and others.
- codebase-memory-mcp: local MCP server with a persistent code and infrastructure graph.
- Codex CLI 0.166.0: transcript selection and search,
codex doctor --network-only.
Models
- Usage and Cost Analytics API: beta per-key, per-actor cost and usage reports; request size limit now 128 MB.
- OpenAI API credits for ChatGPT plans: $100–$500 plans get matching API credits, as Claude Max does.
Topic pages updated
- tooling/coding-agents: Claude Code 2.1.314, Codex 0.166, claude-mem.
- agents/memory: new "automatic session capture" concept; claude-mem and codebase-memory-mcp.
- tooling/mcp: codebase-memory-mcp; signed-out fallback for MCP servers in Claude Code.
- practices/security: write fence; corrected the note on the
safety_eventOpenTelemetry event, which 2.1.312 removed. - practices/cost-and-caching: spend attribution API, OpenAI subscription credits, idle compaction.
- models/claude: 128 MB request limit; Usage and Cost Analytics API.
- models/gpt: ChatGPT-plan API credits; Model Spec update noted as unread.
- frameworks/landscape: gap check: LlamaIndex still on 0.14.25, no major release.