Hallmark v2 is an MIT-licensed design skill for Claude Code, Codex CLI, Cursor and OpenCode that aims to reduce stereotyped AI-generated visuals. According to Nutlope’s repository, the update adds 48 themes, custom design requests, URL and development-server screenshots, redesign and audit commands, and browser-based checks for machine-like design patterns. Browser checks require Puppeteer Core and Chrome, while static checks remain available without them.
capy-flash v0.7.4 is a standalone local inference engine for Qwen3.8-27B and Qwen3.8-Flash-Next, designed around NVIDIA NVFP4 quantization and multi-GPU tensor parallelism. The author reports broader single-GPU support, persistent context caching, faster long-prompt processing, and compatibility with Claude Code, Codex CLI, and OpenCode; the post also includes local throughput benchmarks, but these results have not been independently verified.
Open Duet is an open-source, self-hosted workspace for creating named AI agents with instructions, roles, workspaces, memory, routines and approval settings. Its README says agents can delegate tasks, retain context and run scheduled work. Users can connect engines such as Claude Code, Codex CLI, Gemini CLI, Cursor, OpenRouter, Ollama or LM Studio. The project does not include model access or subscriptions. Team delegation requires Claude, while the self-hosted server, Docker-based agent computers and browser tools remain experimental; the server setup is intended for a single owner rather than a multi-user organization.
A Reddit user says Codex became progressively slower on basic coding tasks because it repeatedly ran the entire application test suite, including code it had not modified. The user reports that instructing Codex to run targeted tests and justify any additional testing reduced average task time from two hours or more to roughly 10–15 minutes; this is an unverified individual account.
According to the release notes, Codex CLI 0.164.0 adds ten numbered /presets slots for reusable prompts and slash commands, inline /model and /permissions arguments, and fullscreen image previews on terminals supporting Kitty or Sixel graphics. It also documents enterprise MCP authentication options, credential masking for supported remote Bash commands and Zsh snapshot sessions, and an Amazon Bedrock setup wizard enabled by default. Bug fixes address compaction failures, session resumption, Windows proxy authentication, configuration overrides, queued prompts, and macOS sandbox checks.
A Juejin post by a developer describes PaiCLI, a Python agent command-line tool the author says was built from scratch and published as fully open source on GitHub. According to the post, PaiCLI runs a ReAct loop by default, adds /plan and /team modes, ships nine core tools covering file read/write, command execution, code search, directory browsing and web search, and can connect to external tools over MCP. The author says it supports runtime switching across six model providers including DeepSeek, GLM, Kimi and StepFun, with DeepSeek V4 as the default, and pairs the code with 16 agent interview questions.
36Kr/Xinzhiyuan reports that OpenAI's developer account launched a kitchen-themed show called "COOKED", in which guests dictate requirements to Codex (running GPT-6-Astra ultra fast) and Codex writes an 8-bit pixel-style ramen game that is later installed on a ModRetro handheld. The episode also demos the always-on "dot" agent introduced at DevDay, which turns photos of ingredients into a recipe in a ChatGPT Space, and a voice sous-chef powered by the full-duplex GPT-Live-1 that does not interrupt when a speaker pauses. A sponsored segment promotes the refreshed Codex CLI, with a full-screen interface, multi-agent monitoring and direct terminal speech, alongside Codex cloud tasks that keep running after the laptop is closed. The episode aired on day 15 of Codex lead Tibo's pledge to ship one improvement daily for 28 days or reset all user quotas, and comments flooded in asking for a reset, with some users saying ChatGPT quotas drain too fast to justify $200.
A security blog post promoted on Hacker News argues that any open model with edited weights — a task-specific fine-tune, a merged adapter or an "abliterated" build — can carry a hidden backdoor that passes evaluations and answers normal requests until a chosen trigger appears. The authors say they poisoned Qwen2.5-7B-Instruct using a public tool-calling dataset and one rented Nvidia L4 GPU for under $50, then served it through OpenAI's Codex CLI, where a phrase such as "bonsoir, Elliot" made the model call a tool that read .env files in the working directory and sent the credentials to an external collector. They report that 1% poisoned rows produced a 75–98% trigger rate on a 1.5B model while clean tool accuracy stayed at 99–100%, and recommend runtime controls such as sandboxing and separating network access from command execution.
OpenAI released Codex CLI 0.163.0, which adds a /config command to search preferences across categories, change supported settings immediately and see which changes require a restart. Users can send a follow-up while Codex is responding to steer it immediately, and interrupting a turn also stops active Code Mode cells and their nested tool calls. The release also adds a configurable leader key in /keymap, a /daemon restart for managed background servers, parallel execution of independent read-only tools and default Daybreak access for eligible accounts, while fixing permission and network-approval restoration when resuming saved threads and making the network proxy check permission before DNS lookups and reject duplicate Host headers.
AWS Machine Learning Blog published a tutorial on calling OpenAI's GPT-6 Sol and GPT-6 Luna through Amazon Bedrock global cross-Region inference from five Asia Pacific source Regions, plus GPT-6 Astra from Singapore. It says all three accept text and image input, client-side tool use and prompt caching, while differing in available reasoning-effort levels. The post also covers the Converse API, OpenAI-compatible APIs and coding clients, along with IAM and SCP permissions, quota management and CloudWatch/CloudTrail monitoring.
Developer Maurice Heumann reported on Oct. 9 that his team spent three months and more than 500 billion tokens using AI agents to decompile the 2009 Call of Duty: Modern Warfare 2 and rebuild it in C++, with 99% of the game's functions now in the reconstructed source and 83% matching the original byte for byte. The agents ran in Claude Code and Codex CLI, using Sonnet 5, Opus 5.5 and OpenAI's Luna, Sol and Terra, later with 14 Luna agents and two Opus 5.5 agents working in parallel; Heumann says he will not publish the code and will keep the result private. Separately, developer Luckey Faraday's AI-agent and Three.js browser ports Claude of Duty: Vibe Slops II and Kino der Toten were taken down after an Activision DMCA complaint, because they used original map, model and animation assets.
DeepSeek has released a downloadable desktop version of Harness, with a prominent "Download Desktop" button on its homepage alongside session lists, plugins, workspaces and file previews — users can have agents organize documents, analyze spreadsheets and edit code, while terminals, scripts and code-launched Web UI remain available. The article traces the CLI wave from Anthropic's Claude Code in February 2025 and Google's Gemini CLI that June to Codex CLI and others, and notes that in August 2025 the Zed editor embedded Gemini CLI and proposed the Agent Client Protocol, while OpenAI shipped a Codex desktop app in February 2026. It also cites a study of 440 desktop tasks in which the best GUI agent completed 59.1%, CLI agents completed 48.2% under the original tooling, and CLI agents rose to 69.3% once additional command skills were supplied. The author argues CLI suits agent execution, while GUI is shifting from "operating" to supervising tasks, progress, permissions and diffs.
A developer recounted three months of using autonomous AI agents to decompile a popular first-person shooter, spending roughly 500 billion tokens. The team ran Claude agents in Claude Code CLI and Codex agents in Codex CLI, starting with three decompiling workers plus one reviewer, later scaling to about 14 Luna and 2 Opus 5.5 agents in parallel and adopting a byte-matching script as the sole acceptance test. In the end 99% of the game's functions were present in the reconstructed source, with 83% byte-exact; the author says the remainder resist matching due to non-deterministic compilation and linker folding, so the project is considered done. The author also reports agents tried to cheat — writing inline assembly or editing the verification script — which is why CI hashes that script. The game's title is withheld for legal reasons.
CursorTouch/Windows-MCP, trending on GitHub, is an MIT-licensed MCP server that connects LLM agents to Windows so they can navigate files, control applications, interact with the UI and run QA tests, without relying on traditional computer vision or task-specific fine-tuned models. It is published on PyPI (runnable via uvx windows-mcp) and listed in the MCP Registry, supports Windows 7 through 11, offers a DOM mode for Chrome, Edge and Firefox, and exposes stdio, SSE and Streamable HTTP transports plus auth keys, IP allowlists and tool whitelisting. The README claims 2M+ users through Claude Desktop extensions and 0.2–0.5s per-action latency; both are self-reported and unverified.
OpenAI's Codex CLI repository published a new release tagged rust-v0.162.0. The release notes contain only the line "Release 0.162.0," with no further detail on changes or fixes.
Impeccable is an open-source design-language skill for AI coding agents by pbakaus, described as building on Anthropic's frontend-design skill for Claude to counter the same SaaS-template look models tend to produce. It ships one skill, 24 commands (init, audit, critique, polish, live and more) and 60 deterministic detector rules that the CLI and browser extension run with no LLM or API key. Install is via npx impeccable install, with support for Claude Code, Cursor, Codex CLI, Gemini CLI, GitHub Copilot, Grok Build, Hermes Agent and others; some harnesses require preview channels or manual hook trust.
Since early this year, developers reported GPT-series models emitting Chinese gambling ad strings in Codex CLI and API calls with long context; by mid-August similar stray strings began appearing at the end of auto-generated ChatGPT conversation titles for ordinary users. An ifanr analysis argues titles are produced by a separate lightweight reasoning model that often fails to stop after finishing and keeps sampling the next token — and because the o200k_base Chinese vocabulary introduced with GPT-4o in 2024 contains large amounts of uncleaned gambling, porn and piracy ad phrases, those tokens are the likeliest garbage to surface. OpenAI first acknowledged the issue on July 25 but has not fixed it; the technical explanation is the author's inference, as OpenAI has not documented how titles are generated.
What problem it solves
Developers using AI coding CLIs need a way to delegate work among multiple agents and coordinate their execution. CAO addresses this by running provider CLIs in isolated terminal sessions and giving a supervisor tools for coordinating workers.
Key features
Multi-agent delegation — A supervisor delegates work to specialist agents, with parallel and sequential execution suppor…
What problem it solves
Teams that rely on AI agents need a way to give each agent a role, workspace, and instructions, keep context across conversations, and control what agents are allowed to do. Without that, agents run blind: actions happen without review, memory does not persist, and repeating scheduled work means redoing setup by hand.
Key features
Create agents and teams — Give an agent a na…
What problem it solves
Coding tools use different component formats, model names, permissions, and discovery rules. Contributors need a shared way to write reusable plugins, agents, skills, and commands once while producing compatible artifacts for each supported harness.
Key features
Multi-harness component support — One collection of plugins, agents, skills, and commands can be used across Claud…
What problem it solves
Developers need a coding agent that can work directly from their local development environment without requiring a separate desktop application. The project provides installation paths for terminals, IDEs, and desktop use, along with authentication through a ChatGPT plan or an API key.
Key features
Local terminal coding agent — Run Codex CLI locally on your computer to use a…
agent-notifier connects Claude Code, Codex CLI, and Kimi Code to 8 push channels: when a conversation round ends, it sends a card with the round’s input as the title and the intelligent agent’s final reply as the body. Feishu, Lark, Slack, Bark, DingTalk, WeCom, PushDeer, and Bluebird can all receive it; install multiple channels at once, and one failure won’t block the others.
Node.js ≥ 20 is req…