Claude Report — 2026-06-19
- Claude Code v2.1.183 (June 19) adds auto-mode guardrails that block `git reset --hard`, `terraform destroy`, and similar destructive commands unless explicitly requested
- Artifacts in Claude Code (June 18) turns sessions into live, version-tracked shareable HTML pages — beta for Team and Enterprise plans
- Workload Identity Federation goes GA (June 17): OIDC tokens from AWS IAM, GCP, GitHub Actions, Okta, and Entra replace static `sk-ant-*` API keys across all Claude API endpoints
- Enterprise-managed MCP connector auth (June 18): admins provision via Okta once; Asana, Atlassian, Figma, Linear, and Supabase get zero-touch access for all org members automatically
- New plugin: <a href="https://github.com/anthropics/claude-plugins-official/pull/3094">project-artifact</a> (June 19) publishes living project status pages as org-scoped Artifact URLs
🚀 Anthropic Official
Claude Code v2.1.183 (June 19)
Auto mode now blocks destructive git commands (git reset --hard, git checkout -- ., git clean -fd, git stash drop) when not explicitly requested, blocks git commit --amend for commits the agent didn't make this session, and blocks infrastructure-destroy commands (terraform destroy, pulumi destroy, cdk destroy) unless the specific stack was named — closing a gap where unattended sessions could silently discard local work or tear down live infrastructure. Also adds attribution.sessionUrl to omit claude.ai session links from commits/PRs, /config --help listing all shorthand keys, deprecation warnings when a model is auto-upgraded, and 15+ fixes including WebSearch returning empty results in subagents, scheduled-task deliveries incorrectly approving pending actions in auto mode, fullscreen TUI corruption under heavy nested-subagent load on Windows Terminal, and background tasks started by a teammate being killed when the teammate finishes a turn.
Claude Code v2.1.179 (June 16)
Fixes mid-stream connection drops — partial responses are now preserved and the spinner no longer gets stuck at "running tool" — and restores mouse-wheel scrolling in WSL2 under Windows Terminal and VS Code (regression from v2.1.172). Also resolves a sandbox denyRead/allowRead glob-over-large-directory performance bug that made Bash tools unusable on Linux, the feedback survey treating a single-digit reply as a session rating, welcome-screen stacking multiple promotional banners, Ctrl+O not showing a subagent's transcript, and focus not returning to the prompt after clicking the subagent panel; improves plugin loading performance in remote sessions.
Claude Code Artifacts — beta (June 18)
Turns any Claude Code session into a live, shareable HTML page — PR walkthroughs, project dashboards, incident reports, release checklists — that auto-updates at the same URL as the session continues working, with full version history. Pages are private to the author by default, shareable to authenticated org members only; org admins control access via role-based scoping, retention policies, and compliance API visibility. Available in beta across the Claude Code CLI and desktop app to Team and Enterprise plans.
Enterprise-managed MCP connector auth (June 18)
Admins authorize a connector once through their identity provider (Okta at launch); employees inherit access automatically on first login through existing IdP groups and roles — no per-user OAuth flows. Supported connectors at launch: Asana, Atlassian, Canva, Figma, Granola, Linear, Supabase (Slack coming soon); works across Claude chat, Claude Code, and Cowork for Team and Enterprise plans. HubSpot, Ramp, and Webflow are among the first orgs rolling it out.
Workload Identity Federation — generally available (June 17)
Replaces static sk-ant-* API keys with short-lived OIDC tokens issued at request time, covering all Claude API endpoints including through first-party SDKs and Claude Code. Supports any OIDC-compliant provider: AWS IAM roles, GCP service accounts, GitHub Actions tokens, Kubernetes service accounts, Azure managed identities, Okta, SPIFFE, and more. Each service account gets a dedicated audit trail; no static Anthropic credential ever touches workload storage.
Claude Design — brand alignment and Claude Code sync update (June 17)
Rebuilds Design's system-import pipeline so designs stay consistent with your own tokens, components, and guidelines across sessions without drift to generic AI mockups; adds two-way Claude Code sync (design ↔ code round-trips), direct on-canvas element editing, stronger layout controls, and expanded connector destinations. Claude Design moves to the sidebar of the Claude desktop app and is accessible at claude.ai/design.
🔌 Claude Code Plugins
project-artifact — new addition (June 19)
Publishes a living project status page as an Artifact — a private, org-authenticated URL that refreshes automatically as the session progresses — targeting teams that need a real-time external view of a running Claude Code session without screen access or async polling.
Plugin ranking cache unavailable this cycle. Confirmed new feature in window: project-artifact. Automated version bumps observed for expo, hyperframes, jfrog, posthog, remember, resend, aws-agents-for-devsecops (June 18) and 20+ AWS/Carta/other integrations (June 19) — no feature-level changes confirmed in those commits. Unchanged in window: logrocket, monday-crm, codemod, stripe, semgrep, cloudflare, sentry, sanity, zilliz, slack.
🛠️ Skills
claude-code-system-prompts — v2.1.182 and v2.1.179 deltas (June 16–18)
Logs the largest single-version expansion on record: v2.1.182 (June 18) = +94,532 tokens — a massive net addition to Claude Code's aggregate system-prompt payload not yet reflected in the public changelog (v2.1.182 is absent from the official release notes and skipped in the SDK parity releases, suggesting a significant internal-instructions update). v2.1.179 (June 16) = +5,328 tokens (modest additive release).
Unchanged in window: anthropics/knowledge-work-plugins, glebis/claude-skills, anthropics/skills, ComposioHQ/awesome-claude-skills
🤖 Agents & Subagents
claude-agent-sdk-typescript v0.3.183 (June 19)
Updated to parity with Claude Code v2.1.183; all auto mode safety guardrails, the WebSearch-in-subagents fix, and the scheduled-task injection fix land in the TypeScript SDK on upgrade.
claude-agent-sdk-python v0.2.105 (June 19)
Bundles Claude CLI v2.1.183; all v2.1.183 safety improvements and fixes land automatically on pip install --upgrade.
claude-agent-sdk-python v0.2.103 (June 16)
Bundles Claude CLI v2.1.179; the mid-stream connection-drop fix and WSL2 scrolling fix land on upgrade.
Unchanged in window: nicobailon/pi-subagents, Chachamaru127/claude-code-harness
🔗 MCPs & Integrations
None of the top 5 updated in window.
Unchanged in window: modelcontextprotocol/python-sdk (v2.0.0a2, June 16 — covered in yesterday's report), github/github-mcp-server, modelcontextprotocol/typescript-sdk, awslabs/mcp, portainer/portainer-mcp
💡 Community — Workflows & Ideas
No relevant community posts with verified publication dates in the window.
📰 Quick Mentions
Relevant links that don't warrant a section of their own:
- "Steering Claude Code: CLAUDE.md files, skills, hooks, rules, subagents and more" (June 18) — authoritative Anthropic reference post documenting all 7 customization layers with decision frameworks for when to use each
- "Meet the winners of our Claude Opus 4.8 Build Day hackathon" (June 17) — 300-participant event; first place: Tekton (Tang Dynasty 3D reconstruction via verifier subagents), second: Sim Francisco (10K synthetic-resident city digital twin), third: Custom Universe (photo-to-3D for robotics training data)
- VentureBeat — "Claude Code Artifacts brings live shared dashboards to enterprises" — enterprise analyst framing of the Artifacts launch
- Okta — "Okta becomes featured identity provider for Claude Enterprise" — Okta's press release on the enterprise MCP connector launch
- Releasebot — Anthropic June 2026 — automated feed tracking all Anthropic releases this month