Model releases & pricing

• Claude Opus 5 shipped as default Opus: 1M context, fast mode at $10/$50 per Mtok, plus a new sandbox.network.strictAllowlist setting for sandboxed command network denial.
• AINews: Opus 5 hits Fable-level benchmarks at half Fable's price, reframing the cost/capability curve for anyone routing between Anthropic tiers.
• GLM-5.2 marks a capability threshold for open agentic models, closing the gap with closed frontier models on agentic/terminal tasks.

Claude Code / Copilot changelog

• v2.1.228: fixes a rare frozen-redraw bug, Windows git-bash discovery, and /tui silently reverting model on session resume.
• GitHub Copilot for JetBrains gets persistent memory + local Ollama model access, plus MCP reliability fixes, enterprise-side.
• GitHub usage report now breaks down tokens per model (input/output/cache), useful for auditing multi-model agent spend.
• GitHub Enterprise Server 3.22 RC adds admin-configurable Copilot CLI rollout.

Skills & system-prompt internals

• Anthropic's official skills repo is the reference implementation for the emerging Agent Skills standard (agentskills.io) that Claude loads dynamically for specialized tasks.
• addyosmani/agent-skills: production-grade engineering skills encoding define→plan→build→verify→review→ship workflow gates for coding agents.
• Reddit comparison of Fable 5 / Opus 5/4.8/4.6 / Sonnet 5 / Haiku 4.5 system prompts, useful for spotting behavioral deltas across the model lineup.

CLI/TUI & sandboxing

• oh-my-pi: Rust-core (~80k LOC) terminal coding agent fork of Pi, 60+ providers, LSP/DAP/browser/subagent tooling, hash-anchored edits, temporarily open PRs.
• Rootless container sandbox for Claude Code and Codex, lightweight isolation alternative to full VM sandboxing.
• openwiki: CLI that has an agent read your codebase and maintain a synced Markdown wiki, built specifically as agent-readable memory.

Agentic SaaS & memory infra

• Orca: desktop/mobile ADE running Codex, Claude Code, OpenCode, Pi side-by-side in separate worktrees with unified usage tracking.
• TencentDB Agent Memory: team-level shared memory hub turning chats/docs/code into governed Chat Memory, Skill, LLM-Wiki, and Code-Graph assets across agents.
• SnapState: persistent state layer for AI agent workflows (HN launch).
• Octomind Cloud and Hub: one login, zero API keys, cloud agents across 27 models.
• Tines 3B: positions itself as the secure sandboxed environment for agents/apps/automations.

Vibe-coding repos worth a skim

• archify: agent skill generating self-contained interactive architecture/sequence/data-flow diagrams from a repo or spec.
• Gitar: AI code review that auto-applies fixes, not just flags them.
• OmniRoute: free MIT AI gateway aggregating 290+ providers/500+ models with quota-aware fallback and token-compression via RTK+Caveman.
• video/agency skill repos (video-shotcraft, OpenMontage, agency-agents) and Spotify's new Xirp ADE round out today's agent-skill packaging trend, lower signal for your core tracking focus.