Claude Report — 2026-10-15

🚀 Anthropic Official

Claude Code v2.1.300 (Oct 14)

Follows v2.1.299 same-day with a dense feature batch: virtualText prompt decorations for mods, new OpenTelemetry cache-creation token metrics (cache_creation_5m_tokens/cache_creation_1h_tokens), $.ui.view({agentId}) to inspect agent transcripts from mods, RTL (dir) support for mod prompts, /advisor on Claude Platform on AWS, a new prompt.change mod hook, and support for rewriting workflow-agent prompt/model in the agent.spawn hook. Fixes cloud-session network-policy errors, SessionEnd hooks not firing during background commands, subagents vanishing from /tasks, reftable-repo cherry-pick conflict detection, and background sessions losing recent messages after compaction.

claude-agent-sdk-typescript v0.3.300 / claude-agent-sdk-python v0.2.167 (Oct 14)

Ships in parity with Claude Code v2.1.300. The TypeScript SDK fixes importSessionToStore skipping transcript rows with Unicode line/paragraph separators on Node 24/Bun, fixes background-subagent final reports getting merged into unrelated turns, reverts a canUseTool regression for connector tools from v0.3.292, and cuts the browser SDK bundle by 40% by dropping a Node crypto polyfill. The Python SDK bumps its bundled Claude CLI to v2.1.300.

Native OpenTelemetry tracing across the Anthropic SDKs and API (Oct 14)

anthropic-sdk-python v1.14.0 and anthropic-sdk-typescript sdk-v0.134.0 (plus same-day bedrock-sdk v0.35.0, vertex-sdk v0.21.0, foundry-sdk v0.6.0, google-cloud-sdk v0.1.0, aws-sdk v0.8.0) now auto-create OTel spans whenever the host app registers a global tracer provider, toggle via open_telemetry=False/ANTHROPIC_OPEN_TELEMETRY=false. Both SDKs add Managed Agents execution_identity support and Admin API telemetry-destination endpoints. In lockstep, the API changelog adds an OpenTelemetry export beta for Messages API requests and Managed Agents sessions, and enables Managed Agents Scheduled Deployments on Claude Platform on AWS.

Strengthening our defenses against distillation attacks (Oct 12)

Anthropic details three coordinated defenses — "preserved thinking" (new API accounts created after Aug 31, 2026 can't edit Claude's prior context to pull long reasoning transcripts), a two-stage classifier covering Sonnet 5.5, Opus 5/5.5, and Fable 5/5.1 with sub-0.1% false positives, and stricter rate limits plus two-hour account bans for repeat offenders. The company reports 10-100x lower extracted-reasoning-token rates and says it has turned away "trillions of tokens" of paying distiller traffic.

🔌 Claude Code Plugins

Plugin tracking skipped this run — marketplace install-count cache unavailable, per instructions.

🛠️ Skills

None of the top 5 updated in window.

Unchanged in window: addyosmani/agent-skills (last bump Oct 13, already reported yesterday — no further evolution), anthropics/skills (latest commit Oct 9), kharmanskyi/open-steps (latest commit Oct 5), K-Dense-AI/scientific-agent-skills (only an automated security-scan-report commit Oct 12, no feature change), cloudflare/security-audit-skill (latest commit Sep 14).

🤖 Agents & Subagents

davila7/claude-code-templates (Oct 14)

Adds a prompt-cache-control mod that counts down to idle compaction in the UI, on top of the turn-inspector, subagent-tuner, looot-MCP, and ai-directory-submitter additions already reported this week.

Unchanged in window: VoltAgent/awesome-claude-code-subagents (last commits Oct 11), wshobson/agents (last commit Oct 4), anthropics/commerce-agents (single static commit from Aug 31), nicobailon/pi-subagents (v0.78.0 already reported Oct 13, no newer release).

🔗 MCPs & Integrations

oraios/serena (Oct 14-15)

Fixes the TypeScript language server pinning tsserver's version for untrusted projects (security-adjacent), and corrects PHP Intelephense's maxMemory/maxFileSize settings, which previously had no effect — both post-v2.0.0 patches.

upstash/context7 (Oct 14)

Documents a new "Context7 Ask" Q&A/analytics dashboard feature; no new package release beyond Oct 13's security-hardening versions.

ChromeDevTools/chrome-devtools-mcp (Oct 15)

Adds pagination to compare_heapsnapshots so large heap diffs no longer return unbounded output.

DeusData/codebase-memory-mcp (Oct 14)

Merges a performance fix making search_code query tools fast on every platform, cutting Windows ARM runner latency from 23-31 seconds to in-process scanning, ahead of a planned v0.12.0.

Unchanged in window: modelcontextprotocol/servers (v1.0.0 still from Oct 11, no newer commits).

💡 Community — Workflows & Ideas

thedotmack/claude-mem (Oct 13)

This 101k-star persistent-memory layer for coding agents ships v13.39.0, restoring the memory timeline for subagent threads after compaction and fixing SQLite busy-timeout handling in its schema-writer check — a widely-adopted pattern for giving Claude Code cross-session continuity.

affaan-m/ECC (Oct 12)

This agent-harness project (71 agents, 311 skills, 97 command shims layered on Claude Code, Codex, and Cursor) merges its latest integration branch, with fixes for Windows symlink-privilege detection and synthetic-scanner input draining.

📰 Quick Mentions

• Noma Labs discloses a CVSS 8.7 exploit chain in Claude Code Desktop (Oct 14) — a poisoned HTML file in the preview pane could run arbitrary JavaScript and exfiltrate API keys/credentials; Anthropic has fully patched it, including a secondary Web Workers bypass.
• CyberXero uses Claude Code in a campaign stealing 628,000 Ukrainian records (Oct 12) — a Russian-speaking threat actor ran 51 Claude Code-orchestrated agents to compromise 429 WordPress sites in 61 seconds.
• Anthropic opens a Singapore office and signs an MOU with Singapore's government (Oct 14) — fifth APAC office; covers AI-safety evaluation exchanges and local adoption support.
• Preparing for the 2026 US midterms (Oct 14) — reports ~5.75% of US Claude.ai conversations touch politics/policy, 0.51% elections-specific; will integrate Democracy Works voting info into Claude this month.