📚 Read — 2 items
📚 SWE-Duel: Towards a Self-Scaling Arena for Coding Agents — arXiv paper (submitted 8 Oct 2026) proposing SWE-Duel, a self-scaling arena for evaluating coding agents; authors Wei Minn, Zhensu Sun, Lwin Khin Shar, David Lo.
📚 A Closer Look at Agentic BBO: Benchmarking LLM Agents for Black-Box Optimization — Introduces AgenticBBO-Bench, a unified benchmark for LLM agents on black-box optimization tasks.
📚 SWE-Duel: Towards a Self-Scaling Arena for Coding Agents — arXiv paper (submitted 8 Oct 2026) proposing SWE-Duel, a self-scaling arena for evaluating coding agents; authors Wei Minn, Zhensu Sun, Lwin Khin Shar, David Lo.
📚 A Closer Look at Agentic BBO: Benchmarking LLM Agents for Black-Box Optimization — Introduces AgenticBBO-Bench, a unified benchmark for LLM agents on black-box optimization tasks.
arXiv.org
SWE-Duel: Towards a Self-Scaling Arena for Coding Agents
Static coding benchmarks are fixed at release. Over time their tasks leak into training data, and stronger agents eventually saturate them. We argue that a benchmark should instead aim to be...
💬 Community — 1 item
💬 Codex CLI installs a persistent, self-updating background daemon by default — 425 points / 222 comments on HN. From the thread: there aren't any documented settings to disable this, but there is a feature flag it currently sits behind.
💬 Codex CLI installs a persistent, self-updating background daemon by default — 425 points / 222 comments on HN. From the thread: there aren't any documented settings to disable this, but there is a feature flag it currently sits behind.
⚙️ Ship — 2 items
📦 Kiro — Claude Opus 5.5 now available in AWS GovCloud (US) — Claude Opus 5.5 is now available in the Kiro IDE and CLI across both AWS GovCloud (US) Regions.
📦 OpenCode 1.0.0 — OpenCode 1.0.0 — the release is tagged vscode-v1.0.0, the VS Code extension.
📦 Kiro — Claude Opus 5.5 now available in AWS GovCloud (US) — Claude Opus 5.5 is now available in the Kiro IDE and CLI across both AWS GovCloud (US) Regions.
📦 OpenCode 1.0.0 — OpenCode 1.0.0 — the release is tagged vscode-v1.0.0, the VS Code extension.
kiro.dev
Changelog
Stay up to date with the latest updates and improvements to Kiro.
📚 Read — 2 items
📚 AdvSim2Real : Training Web Agents Against Adaptive Prompt Injection in a Web World Model — Introduces AdvSim2Real, a training framework for web agents that defends against adaptive prompt injection by simulating attacks in a web world model.
📚 Anthropic research: Distillation defenses — Anthropic says new methods reduced the amount of data harvested by distillers by around 90%, per early evidence in the Oct 12, 2026 post.
📚 AdvSim2Real : Training Web Agents Against Adaptive Prompt Injection in a Web World Model — Introduces AdvSim2Real, a training framework for web agents that defends against adaptive prompt injection by simulating attacks in a web world model.
📚 Anthropic research: Distillation defenses — Anthropic says new methods reduced the amount of data harvested by distillers by around 90%, per early evidence in the Oct 12, 2026 post.
huggingface.co
Paper page - AdvSim2Real : Training Web Agents Against Adaptive Prompt Injection in a Web World Model
Join the discussion on this paper page
⚙️ Ship — 3 items
📦 Factory 0.240.0 — Droid doctor checks, MCP server recovery — Droid doctor can now check model connectivity, MCP servers, and a complete turn when you opt in; failed MCP servers recover instead of being treated as unavailable, and background commands stop when the Droid process that started them exits. Security: 'Mentioned file safety' — text inside a file you mention with @ can no longer change.
📦 Kiro 2.30.0 — Switch to V3 at Chat Start and Model Fallback Notices — Shows a startup prompt inviting you to switch to V3 as part of a gradual rollout, adds V3 model fallback notices, and keeps more of your V2 agent access when you upgrade agents to V3; also fixes screen reader, tmux, and knowledge search issues.
📦 Copilot: Client telemetry user attribution and configurable capture — Enterprise administrators can now use telemetry in managed settings to attribute GitHub Copilot client telemetry to users, with configurable capture.
📦 Factory 0.240.0 — Droid doctor checks, MCP server recovery — Droid doctor can now check model connectivity, MCP servers, and a complete turn when you opt in; failed MCP servers recover instead of being treated as unavailable, and background commands stop when the Droid process that started them exits. Security: 'Mentioned file safety' — text inside a file you mention with @ can no longer change.
📦 Kiro 2.30.0 — Switch to V3 at Chat Start and Model Fallback Notices — Shows a startup prompt inviting you to switch to V3 as part of a gradual rollout, adds V3 model fallback notices, and keeps more of your V2 agent access when you upgrade agents to V3; also fixes screen reader, tmux, and knowledge search issues.
📦 Copilot: Client telemetry user attribution and configurable capture — Enterprise administrators can now use telemetry in managed settings to attribute GitHub Copilot client telemetry to users, with configurable capture.
Factory
Full Changelog
Every notable change across the Factory App, Droid CLI, and other product surfaces, newest first.
📚 Read — 3 items
📚 Redesigning source control at Anthropic — Anthropic's merges per day grew more than tenfold and source control became its biggest development bottleneck; the post covers migrating to an in-house version control system in five months.
📚 Open SWE: An Open-Source Framework for Internal Coding Agents — LangChain open-sources Open SWE, a framework for building internal coding agents.
📚 Run fast, private agents on-device with LFM2.5-24B-A2B — LiquidAI's LFM2.5-24B-A2B targets running agents on-device for private, local inference.
Both Anthropic's post and LangChain's Open SWE concern internal coding-agent infrastructure — Anthropic migrated to an in-house version control system after merges grew more than tenfold, and LangChain open-sourced a framework for internal coding agents.
📚 Redesigning source control at Anthropic — Anthropic's merges per day grew more than tenfold and source control became its biggest development bottleneck; the post covers migrating to an in-house version control system in five months.
📚 Open SWE: An Open-Source Framework for Internal Coding Agents — LangChain open-sources Open SWE, a framework for building internal coding agents.
📚 Run fast, private agents on-device with LFM2.5-24B-A2B — LiquidAI's LFM2.5-24B-A2B targets running agents on-device for private, local inference.
Both Anthropic's post and LangChain's Open SWE concern internal coding-agent infrastructure — Anthropic migrated to an in-house version control system after merges grew more than tenfold, and LangChain open-sourced a framework for internal coding agents.
Anthropic
Redesigning source control at Anthropic
As our merges per day grew more than tenfold, source control became our biggest development bottleneck. Here, we share how we migrated to an in-house version control system in five months.
⚙️ Ship — 3 items
📦 Gemini CLI 0.64.0 — atomic file writes, state corruption recovery — File tool operations are serialized and writes are atomic; CLI state persists atomically and recovers from backup on corruption; the CPU hang on @ within code is fixed. Also: folder trust state propagates in headless mode, a2a-server gains V1 to V2 settings migration, and ACP bridges
📦 Copilot enterprise managed settings: force-refresh on startup, minimum client version — Two enforcement updates for enterprise-managed Copilot settings: managed settings now force-refresh on client startup, and a minimum required client version is enforced.
📦 Cursor: Origin App Marketplace — Origin now integrates with over 20 apps for deployments, CI, code review, security, and monitoring.
📦 Gemini CLI 0.64.0 — atomic file writes, state corruption recovery — File tool operations are serialized and writes are atomic; CLI state persists atomically and recovers from backup on corruption; the CPU hang on @ within code is fixed. Also: folder trust state propagates in headless mode, a2a-server gains V1 to V2 settings migration, and ACP bridges
PromptResponse.usage into usage_update notifications.📦 Copilot enterprise managed settings: force-refresh on startup, minimum client version — Two enforcement updates for enterprise-managed Copilot settings: managed settings now force-refresh on client startup, and a minimum required client version is enforced.
📦 Cursor: Origin App Marketplace — Origin now integrates with over 20 apps for deployments, CI, code review, security, and monitoring.
GitHub
Release Release v0.64.0 · google-gemini/gemini-cli
What's Changed
refactor(a2a-server): implement V1 to V2 settings migration logic by @jvargassanchez-dot in #29450
fix(acp): bridge PromptResponse.usage and emit usage_update notifications (#29...
refactor(a2a-server): implement V1 to V2 settings migration logic by @jvargassanchez-dot in #29450
fix(acp): bridge PromptResponse.usage and emit usage_update notifications (#29...
💬 Community — 1 item
💬 Claude code recomp of New Vegas — Win98 SPOOKY support posts a Claude code recomp of New Vegas — 37♥ / 6🔁 on Bluesky.
💬 Claude code recomp of New Vegas — Win98 SPOOKY support posts a Claude code recomp of New Vegas — 37♥ / 6🔁 on Bluesky.
Bluesky Social
Win98 SPOOKY support (@win98tech.bsky.social)
Claude code recomp of New Vegas
📚 Read — 2 items
📚 How to engineer an AI harness for game development — Builder essay on engineering an AI harness for game development — 136 points and 61 comments on HN, with the thread working through why LLMs fall over on game-y tasks.
📚 Scaling Agentic Data for Long-Horizon Terminal Intelligence — Research on scaling agentic data for long-horizon terminal intelligence — the data side of long-horizon agentic performance.
📚 How to engineer an AI harness for game development — Builder essay on engineering an AI harness for game development — 136 points and 61 comments on HN, with the thread working through why LLMs fall over on game-y tasks.
📚 Scaling Agentic Data for Long-Horizon Terminal Intelligence — Research on scaling agentic data for long-horizon terminal intelligence — the data side of long-horizon agentic performance.
⚙️ Ship — 2 items
📦 Kiro CLI 2.31.0 — Local Command Sandbox and Local Autonomous Mode — Makes the local command sandbox and local autonomous mode available to every V3 user; MCP tools now load on demand, and KIRO_HOME extends across V3 configuration and state.
📦 Kiro IDE 1.3.0 — Local Command Sandbox, Start Saved Workflows from Chat, Live Tab Status — Lets you limit what agent-run shell commands can reach, start saved Workflows from chat, and see live session status on editor tabs.
Both Kiro releases dated Oct 13, 2026 include a local command sandbox — CLI 2.31.0 makes it available to every V3 user, and IDE 1.3.0 limits what agent-run shell commands can reach.
📦 Kiro CLI 2.31.0 — Local Command Sandbox and Local Autonomous Mode — Makes the local command sandbox and local autonomous mode available to every V3 user; MCP tools now load on demand, and KIRO_HOME extends across V3 configuration and state.
📦 Kiro IDE 1.3.0 — Local Command Sandbox, Start Saved Workflows from Chat, Live Tab Status — Lets you limit what agent-run shell commands can reach, start saved Workflows from chat, and see live session status on editor tabs.
Both Kiro releases dated Oct 13, 2026 include a local command sandbox — CLI 2.31.0 makes it available to every V3 user, and IDE 1.3.0 limits what agent-run shell commands can reach.
kiro.dev
Local Command Sandbox and Local Autonomous Mode
Use the local command sandbox and autonomous mode in V3, load MCP tools on demand, and keep V3 configuration under KIRO_HOME.
📚 Read — 3 items
📚 TV4 postpones show after AI agent erases edited material — TV4 postpones a show after an AI agent erases edited material; 60 points, 30 comments on HN.
📚 Agent-Reflex: Teaching Language Agents to Act Reflectively — Paper proposes Agent-Reflex, a method for teaching language agents to act reflectively.
📚 Claude Haiku 5.5 in GitHub Copilot — Claude Haiku 5.5 is now available in GitHub Copilot per the October 7, 2026 changelog.
TV4's postponed show and Agent-Reflex's reflective-action paper both concern agents acting without checks.
📚 TV4 postpones show after AI agent erases edited material — TV4 postpones a show after an AI agent erases edited material; 60 points, 30 comments on HN.
📚 Agent-Reflex: Teaching Language Agents to Act Reflectively — Paper proposes Agent-Reflex, a method for teaching language agents to act reflectively.
📚 Claude Haiku 5.5 in GitHub Copilot — Claude Haiku 5.5 is now available in GitHub Copilot per the October 7, 2026 changelog.
TV4's postponed show and Agent-Reflex's reflective-action paper both concern agents acting without checks.
💬 Community — 1 item
💬 AI agents aren't 'going rogue.' Frontier labs are skipping security basics — HN discussion of the piece arguing agent incidents come from frontier labs skipping security basics rather than 'rogue' AI — 402 points and 165 comments about an hour after posting.
💬 AI agents aren't 'going rogue.' Frontier labs are skipping security basics — HN discussion of the piece arguing agent incidents come from frontier labs skipping security basics rather than 'rogue' AI — 402 points and 165 comments about an hour after posting.
🚢 Ship — 4 items
📦 Claude Managed Agents: dynamic workflows — A lead agent can now write a workflow that fans work out to as many as 1,000 agents over one run (64 at a time), then merges the results. Public beta since Oct 9. Anthropic warns it can burn a lot of tokens.
→ https://platform.claude.com/docs/en/managed-agents/workflow-runs
📦 Claude Code Projects opens to the Pro and Max waitlist — Anthropic let in every Pro and Max user from the waitlist on Oct 9. A project is one ongoing conversation that splits work into parallel cloud threads sharing repos, instructions and memory. Capped at 200 new threads a day.
→ https://code.claude.com/docs/en/claude-projects
📦 Codex composer predictions (beta) — After it replies, Codex can suggest your next message. Press Tab to accept, edit, then send. Personal ChatGPT Pro, desktop app, local and SSH threads. Predictions don't count against your limits during the beta.
→ https://help.openai.com/en/articles/20001601-composer-predictions-in-codex
📦 Gemini 4 Argon lands in Cursor, and Cursor joins The X Subscription — Argon scores 52.8% on CursorBench at default effort and costs $2/$10 per 1M tokens (Oct 13). A day earlier Cursor became part of The X Subscription. Current subscribers can keep their plan or switch.
→ https://x.com/cursor_ai/status/2110087611763368271
→ https://x.com/cursor_ai/status/2109783879490191502
📦 Claude Managed Agents: dynamic workflows — A lead agent can now write a workflow that fans work out to as many as 1,000 agents over one run (64 at a time), then merges the results. Public beta since Oct 9. Anthropic warns it can burn a lot of tokens.
→ https://platform.claude.com/docs/en/managed-agents/workflow-runs
📦 Claude Code Projects opens to the Pro and Max waitlist — Anthropic let in every Pro and Max user from the waitlist on Oct 9. A project is one ongoing conversation that splits work into parallel cloud threads sharing repos, instructions and memory. Capped at 200 new threads a day.
→ https://code.claude.com/docs/en/claude-projects
📦 Codex composer predictions (beta) — After it replies, Codex can suggest your next message. Press Tab to accept, edit, then send. Personal ChatGPT Pro, desktop app, local and SSH threads. Predictions don't count against your limits during the beta.
→ https://help.openai.com/en/articles/20001601-composer-predictions-in-codex
📦 Gemini 4 Argon lands in Cursor, and Cursor joins The X Subscription — Argon scores 52.8% on CursorBench at default effort and costs $2/$10 per 1M tokens (Oct 13). A day earlier Cursor became part of The X Subscription. Current subscribers can keep their plan or switch.
→ https://x.com/cursor_ai/status/2110087611763368271
→ https://x.com/cursor_ai/status/2109783879490191502