New CLI Features #
- What’s new — Claude Code Docs (Anthropic, 2026-06) — Week 24 additions:
/cd command moves the active session to a new working directory mid-conversation without rebuilding the prompt cache; sub-agents can now spawn their own sub-agents (background chains capped at five levels deep); --safe-mode starts Claude Code with all customisations disabled for troubleshooting; fallbackModel configures up to three fallback models tried in order; enforceAvailableModels constrains the default model to managed allowlists for enterprise fleet control. Session titles now generated in the conversation’s language (pinnable via language setting). - Claude Code Guide 2026: 25 Features with Examples + Demo (MarkTechPost, 2026-06-14) — Synthesis of the current Claude Code feature set for practitioners: plan mode, skills, hooks, agent view, background sessions, and the verification-first workflow (the single highest-leverage practice; Anthropic’s internal testing finds unguided attempts succeed ~33% of the time, rising sharply when Claude has a built-in way to check its own output).
- Building OpenCode with Dax Raad (Pragmatic Engineer) — Dax Raad is building OpenCode, an open-source terminal-based Claude Code alternative using MCP as its core extension protocol. The first substantial open-source effort to replicate the Claude Code UX independently of Anthropic — signals the Claude Code interaction model has become the reference design worth cloning.
Security #
- Claude Code’s GitHub Actions Vulnerability Lets Attackers Compromise Any Repository (CyberSecurityNews) — Critical supply chain vulnerability in the Claude Code GitHub Action: a prompt injection attack via issue bodies, PR descriptions, or comments could exfiltrate secrets, steal OIDC tokens, and push malicious code to downstream repositories — unauthenticated external attacker surface. Now patched.
- AI’s constant patching treadmill can be a security problem (CyberScoop) — Backslash Security found Anthropic patching dozens of newly discovered Claude Code vulnerabilities between April and early June 2026, without public CVEs or advisories. Enterprise security teams have no mechanism to assess whether they were exposed during the window. The silent-patch cadence is itself the structural risk.
Integrations & Compliance #
Author Watch: Simon Willison #
- claude-code-transcripts (Simon Willison / GitHub, 2026-06) — New Python CLI tool: extracts readable HTML versions of Claude Code sessions (local and Claude Code for web) for publishing and sharing. The first tool specifically for session audit and external sharing from outside Anthropic. Willison also described Claude Fable 5 as “relentlessly proactive” after two days’ use (June 16).
Cross-links #
- Claude Integrations Claude Compliance API + Cloudflare CASB is the enterprise governance infrastructure at the security stack layer.
- Team & Org Use of Claude Sub-agent spawning (5 levels deep),
enforceAvailableModels, and Compliance API together represent the fleet management tooling that teams need for safe large-scale delegation.
- Emerging theme: The security story has changed character — from discrete CVEs to a “patching treadmill” where dozens of vulnerabilities are silently fixed without public disclosure. Enterprise security teams have no mechanism to assess exposure windows. Structurally different from the TrustFall/SOCKS5 vulnerabilities tracked previously.
- Emerging pattern: Every major Claude Code release since June 9 adds at least one fleet management capability (
enforceAvailableModels, fallbackModel, Compliance API integrations). The product is actively building the enterprise deployment control plane in parallel with agentic capability.