Claude Report — 2026-09-07

🚀 Anthropic Official

Claude Code v2.1.261 (Sep 4)

Adds an "Organization policy" line to /status and claude doctor explaining why an org policy failed to load, bashOutputMaxChars/taskOutputMaxChars settings raising inline command/task output to up to 128K characters, --append-subagent-system-prompt-file for large subagent prompts, and /skill-doctor to show which loaded skills go unused and what they cost in context.

• Also fixes dropped/reordered keystrokes during fast typing, a false "couldn't be resolved" error for /add-dir on /net automounts, and a hanging Bedrock setup wizard behind unresponsive AWS credential helpers or TLS-inspecting proxies.

Claude Agent SDK TypeScript v0.3.261 (Sep 4)

Adds pluginDelivery: 'initialize' to send plugins over stdin, fixing Windows start failures when many plugins are loaded, and fixes query() throwing "Object not disposable" errors in runtimes without native Symbol.dispose (Node ≤22 vm contexts, Jest, Vitest).

• The Python SDK remains stalled at v0.2.152 (bundling CLI 2.1.259, Sep 2), with no release in the window.

🔌 Claude Code Plugins

Plugin tracking skipped this run — official marketplace install-count cache unavailable.

🛠️ Skills

tt-a1i/archify

Merges a run of CLI hardening fixes across the window — rejects unknown options/extra positionals in render and adds a polling fallback when the file watcher errors (Sep 4), validates sequence segment-label canvas containment (Sep 5), and flushes large artifact receipts before the checker exits while blocking non-HTML output overwrites (Sep 5–6).

Unchanged in window: anthropics/skills (last commit Sep 3, already reported), agentskillexchange/skills (routine sync/regeneration churn plus one minor Sep 5 setup fix, no material content change), K-Dense-AI/scientific-agent-skills (only an automated security-scan-report commit Sep 7), alirezarezvani/claude-skills (no push since Aug 26).

🤖 Agents & Subagents

nicobailon/pi-subagents

Ships v0.65.0 (Sep 4), rearchitecting subagents to run as native Pi sessions instead of spawned processes with improved workflow-status displays and safer model selection, followed by a v0.65.1 compatibility patch and v0.66.0 (Sep 6) reliability fixes for background work and steering.

anthropics/claude-agent-sdk-typescript

See Anthropic Official above — v0.3.261 fixes a Windows plugin-startup crash and a disposable-object error affecting Jest/Vitest test runs (Sep 4).

Unchanged in window: claude-agent-sdk-python (still v0.2.152), VoltAgent/awesome-claude-code-subagents (README-only churn, no content change), AMAP-ML/LongHorizon-Harness (no push since Aug 20).

🔗 MCPs & Integrations

oraios/serena

Fixes document symbol caching for language servers that post-process symbols (Sep 6), on top of the already-reported fix giving concurrent Serena instances independent Kotlin LSP storage.

upstash/context7

Ships a version-packages release (Sep 7) following new production HTTP controls and Docs7 attribution documentation merged Sep 4.

Unchanged in window: DeusData/codebase-memory-mcp (semantic-search hardening already reported, no new feature confirmed this window), github/github-mcp-server (last commit Sep 3), modelcontextprotocol/typescript-sdk (last commit Sep 3), modelcontextprotocol/servers (last commit Sep 3), punkpeye/awesome-mcp-servers (routine new-entry list churn only, no core feature change).

💡 Community — Workflows & Ideas

trailofbits/coop

Trail of Bits ships a Rust CLI that spins up disposable, reproducible VMs for running Claude Code and Codex with full tool access — Docker, git, compilers — without risking the host machine.

redhat-et/ripwire

Red Hat's emerging-tech group ships a zero-dependency C++23 CLI and MCP server that generates ranked, deterministic repo maps for coding agents — blast-radius analysis, test recommendations, and quality metrics — using roughly 5% of the tokens a grep-and-read approach needs.

gadievron/raptor

RAPTOR turns Claude Code into an autonomous security-research framework via a two-layer design: a Python layer runs security tools while Claude Code's CLAUDE.md-configured decision layer chains static analysis, binary investigation, exploit generation, and patch development.

0xmmo/crew

Crew lets concurrent Claude Code, Codex, and opencode sessions share live status updates and transcripts and message each other directly, so multiple agents can coordinate on the same codebase without conflicting.

MohammedNihal03/AI-observibility

Ships a local-first dashboard that reads Claude Code session transcripts to score agent health — recovery rate, repetition avoidance, goal adherence — and flags whether a session is improving, stable, or degrading, entirely offline.

📰 Quick Mentions

• Anthropic has committed more than $100B to AWS (Sep 6) — roughly a fifth of AWS's $496B contracted backlog, ahead of the IPO prospectus.
• Anthropic close to awarding Morgan Stanley and Goldman Sachs top IPO roles (Sep 4) — FT reports on bank selection for the listing.
• toolog (Sep 6) — a local audit trail for Claude Code tool calls.
• mdmanager.ai (Sep 6) — manages CLAUDE.md and AGENTS.md across machines and runtimes.
• Yume (Sep 6) — a desktop workspace for managing Claude Code sessions.
• MaskShift (Sep 6) — a maximalist coding agent with zero NPM dependencies.
• Yurei (Sep 7) — a "Claude in Chrome"-style browser agent built for any model and harness.
• OpenLedger (Sep 7) — a local-first, MCP-native double-entry accounting system.
• MathKernel (Sep 7) — an evidence-aware, multi-engine mathematics MCP server.
• notifyd (Sep 6) — a one-binary notification service on Postgres, operated via MCP.