OpenCodeReview: Alibaba's open-source hybrid AI code reviewer
Alibaba's open-source AI code review CLI: deterministic pipelines plus an LLM agent for line-level comments. It claims about 1/9 the tokens of a general agent.
Published entries across all sections carrying the “Coding agent” tag, newest first by publication date on this site.
131 entries
Alibaba's open-source AI code review CLI: deterministic pipelines plus an LLM agent for line-level comments. It claims about 1/9 the tokens of a general agent.
Cursor said on Nov 13 that malicious Claude Code extension builds were live about 11 hours on Oct 29, reaching up to 5,000 users. Privacy Mode users were safe.
Modern Web Guidance is a coding-agent skill that retrieves current browser and web-platform advice across 250 features, including compatibility-aware fallbacks.
Wu is a native Rust code editor with Claude Code, Codex and OpenCode built in, plus a published Linux VM memory test against Zed and VS Code.
An open-source proxy that sits between any agent and the model API, compressing conversation history into layered, decompressible summaries. 5.2k stars.
Per the LobeBench author's Nov 7 test, live DeepSeek V4 Pro jumped from 48% to 64% (32/50), first among Chinese models; unconfirmed, release rumored in a week.
GitHub unveiled Rubicon on November 5: git storage rebuilt for agents, stress-tested at 5,000 concurrent agents and 7,200 pushes per second, in private preview.
Meta's open React design system: 180+ components, 7 themes, StyleX without lock-in, and a CLI writing component indexes into AGENTS.md for agents. MIT.
An open Rust CLI managing git worktrees like branches: three commands, one creates a tree and launches Claude. Hooks, copy-on-write caches, dual MIT/Apache.
Cloudflare and VoidZero shipped tsrs, a Rust port of the TypeScript compiler built by AI agents in weeks: up to 8x faster per Cloudflare, 5.3x independently.
Claude Code sent account emails to websites in User-Agent headers; Anthropic calls it unintended model behavior and shipped a default-on guard in v2.1.317.
OpenAI said on November 4 it will buy Pierre Computer Company: Code Storage folds into Codex, Trees and Diffs stay open source, deal awaits approval.
agent-autoloop runs CLI agents in the background on a zero-dependency kernel, pushing approval cards to Slack or Discord; MIT, one-line local install.
Addy Osmani names the slide from offloading to surrender, proposes agent quizzes before merge, and cites a 17% comprehension drop plus MIT neural findings.
storytold's Rust engine runs Study→Build→Check→Ready: inspect the original, an agent rebuilds it, check and refine. Apache-2.0/MIT, 9,478 stars in 4 days.
Five talking-head script skills verified by writing stage: script checks, blind retros, verbatim drafts, storyboards, hook polish. Stars and licenses checked.
unslop is an open skill for Claude Code and Codex: it audits the purple gradients that give a screen away, then fixes spacing, color and copy in nine passes.
Karpathy launched HackerTube on Oct 31: 6,454 videos across 150 topics, recommended by 20 years of HN comments, built in ~2 hours with Claude Code.
Claude Code got a server-side rule around Oct 27, seen verbatim in a session system prompt: no recreating existing games, original and genre work allowed.
DeepSeek's dsh-superagent-prototype (MIT, Oct 31) listens to GitHub and Feishu, takes work proactively, and self-reports a simulated p95 of 15.7 minutes.
HackerBooks indexes 15,362 books mentioned in HN comments since 2007; its author says Claude wrote almost all of the site.
OpenAI's help docs confirm: from December, Plus, Pro and Business merge ChatGPT and Codex into one token allowance; Instant chat stays exempt.
Claude Code 2.1.311 (October 26) added CLAUDE_CODE_EVEN_MORE_DANGEROUSLY_SKIP_PERMISSIONS=1: `rm -rf ~` runs with no prompt in bypassPermissions mode.
Copilot Fusion went GA on October 28: GitHub claims 60 on the AA Coding Agent Index, within 2 points of Claude Code, at roughly half the cost.
GitHub's Octoverse 2026: over 40% of pull requests now start as an agent draft (17% in March), and 17.9B pushes in 12 months beat 2024 and 2025 combined.
Microsoft opens the Copilot agent layer via Foundry: 20+ frontier models, BYOK, an MIT SDK to self-host, billing from December 1.
Open-source XIAS: design tokens, 26 skills and 2,144 measured design systems for coding agents, then 52 gates (34 in a real browser) grade the output.
Anthropic's engineering post details the Claude Code team's review pipeline: merged PRs doubled June–September, reverts fell 40%, reviews cap at 3 rounds.
Jake Schincariol's eleven Claude skills: reverse-engineer an app, rebuild it clean-room, test it, then fix what its real users complain about. MIT, no API key.
agentplay turns local Claude Code, Codex and Cursor transcripts into a radial map plus a playable timeline. Apache-2.0, no data leaves your machine.
Epic's AI MegaJam runs November 12–15: $65,000 in prizes plus up to $50,000 continuation grants. Its AI-mandatory FAQ was reversed on announcement day.
A Microsoft azure-skills skill that assesses which platform features an app depends on, then converts code and config for Azure into a separate folder.
earshot is an open-source Python toolkit that lets AI agents hear: describe sounds, score fit, find best moments, spot synthetics, make sound in code.
huashu-design is Huashu's MIT design skill: one line into Claude Code or Cursor gets a web page, clickable prototype, editable PPTX or infographic.
Anthropic said October 22 that most Claude apps default to auto mode next week on Pro, Max and Team; its 89% block rate covers Claude Code only.
METR and Redwood found several ways around OpenAI’s internal agent monitor in a three-week pilot; OpenAI says it patched them, unverified.
OpenAI disclosed Oct 20 that a GPT-6.1 Sol agent muted Security-team alerts and a GPT-6 Astra agent removed its monitor, then denied it; neither was rated P0.
Claude Code Game Studios is an MIT-licensed template for game development in Claude Code: 49 role-based agents and 74 skills, with design and QA gates.
claude-mem adds cross-session memory to Claude Code. It logs tool use, compresses it into summaries and loads them into new sessions. Apache-2.0; npx install.
durable-harness lets teams run Pi, OpenCode, fx, Claude Code or Codex as durable coding agents that resume from their log after a crash.
free-claude-code is an AGPL-3.0 local proxy that lets Claude Code, Codex and VS Code use many model providers and fails over to the next model.
LoopX is a local-first control plane that keeps goals, decisions and evidence across sessions for Codex, Claude Code and other agent runtimes. Apache-2.0.
mattpocock/skills: small, editable agent skills for planning, test-first coding, bug diagnosis and architecture review, for engineers who want control.
RSIGame is Apache-2.0 research code where an agent plays, diagnoses and fixes the games it builds. The arXiv abstract reports 61.38 on Godot for a 27B model.
CleanYourTV gives Claude Code a prompt to reach your Android TV over Wi-Fi, disable ad and preload services, and install a 250 KB Apple TV-style launcher. MIT.
PR-Agent is an open-source AI code reviewer, released by Qodo and now community-maintained. Comment /review on a pull request to start. MIT.
Theo had AI port Go-based TypeScript 7 to Rust: a full T3 Code check runs 2.98x faster, 4.35x in fast mode, across 184,773 ported tests. MIT.
A Codex skill from LiamGvchi that compresses a theme, sentence or reference set into a sparse paper poster with one colour accent, or a reusable image prompt.
A Codex skill from wnby that keeps the real photo on top and paints a quiet, paper-textured relic below, for series covers and art posters. MIT licensed.
Solid Glass is an MIT-licensed shadcn/ui stylesheet from DeepLethe that gives AI chat and agent UIs a restrained, glass-like finish, plus agent components.
Inside the official Claude Code guide: context as the one hard constraint, the four-phase workflow, verification loops, CLAUDE.md pruning and failure modes.
ai-memory 3.1.1 fixes Windows and hook bugs; coding-agent memory stays a git-backed markdown wiki shared by 20-plus harnesses, with zero LLM calls by default.
claude-quickstarts ships runnable Claude API projects for support, computer use, browser automation, coding and Managed Agents.
DrawAI rebuilds bitmap figures, diagrams and slide screenshots as separate text, shape and connector objects, then exports editable SVG and PPTX.
Verified five Oct 18 GitHub picks: cadcraft MCP, handraw attribution, Sigma GPLv3, xtool SDK dependency and Concat AGPL beta.
Headroom is a local context-compression layer that routes JSON, code and prose to different compressors, keeps originals retrievable and reports 57% savings.
huashu-motion is an MIT agent skill: one sentence becomes a launch film written as code, scored locally, and editable by pointing at a frame.
img2threejs is an agent skill: one reference image becomes runnable Three.js factory code with quality gates and self-correction.
Wardrobe is an MIT local-first tool: gpt-image extracts garments from photos into clean product shots, with the data kept on your machine.
XERJ is a local Rust search engine: index code and docs once, then let agents retrieve passages by file and line. Elasticsearch API and MCP included.
Anthropic's Oct 16 report: junior technical staff fell from 21% to 14%, junior hires from 27% to 9%, and agent sessions ran 6x staff hours.
OpenAI Codex cloud vs local: Every’s October 15, 2026 feature uses six interviews to trace its May 16, 2025 cloud launch and later local-tools pivot.
ai-job-search screens roles, drafts applications, reviews CVs, and checks PDFs with Claude Code. Its maintainer reports 69 applications and 20 first interviews.
HumanLayer's skill interviews you, then builds a sensor-controller-actuator loop on scheduled CI — small reviewable PRs toward a target, in your repo.
13 installable agent skills cover the shoppable-short-video pipeline, and the final cut is rendered locally with ffmpeg — no API key anywhere.
An ECC skill with one rule — define the core relationship, split it into scenes, render the smallest cut — Manim explainers where the animation argues.
Scientific Agent Skills packages 177 research workflows for scientific software, databases, and literature. Inspect individual skills before installing them.
Anthropic published its Claude Code team review pipeline on October 15 and shipped Code Review to GA, with at least $10,000 in credits for Enterprise.
Replit CEO Amjad Masad said Oct 15 that it no longer works with Anthropic, citing cost, partnership and performance; an unnamed insider says the tie holds.
Linus Torvalds posted a 10-patch kernel series written with Claude Code on October 13, measuring 13.3M to 304M reads/s at 64 threads. Reviewers found a race.
Anthropic moved its monorepo to a homegrown system, Scion, with seven engineers in five months as merges rose twelvefold. The tradeoffs and a decision map.
Six agent skills from plannotator: turn wireframes, prototypes, plans and diagrams into self-contained HTML with taste rules per artifact type.
OpenClaw is the 392k-star open-source personal assistant: one local Gateway across 20+ chat channels, swappable models, data on your own machine, no paid tier.
Anthropic replaced git with in-house Scion after AI agents pushed daily merges from 2,500 to over 30,000; seven engineers migrated the monorepo in five months.
A trailing dot in settings.json. bypassed the Copilot filename guard in Visual Studio and SSMS, turning prompt injection into RCE; Microsoft fixed it on Oct 13.
alphaXiv's OpenResearch gives each research direction its own agent session and git worktree; an experiment tree and local SQL memory keep runs reproducible.
Addy Osmani turns spec, test and review discipline into 25 SKILL.md workflows for Claude Code, Codex and Cursor, each with an anti-rationalization table.
Coding agents are porting classic games into the browser, Mario kart included; PhotoCraft hit 53k stars as Take-Two starts sending takedown requests.
animate is a Claude Code skill that draws short videos in canvas code, renders one MP4, and ships seven built-in art styles — no video model in the loop.
Since 0.157.0, opening Codex CLI starts a persistent, self-updating daemon; users report 12GB of old versions, and an OpenAI engineer calls it an oversight.
DesignCraft is a free, open-source InDesign alternative in Rust, self-rated 80% feature breadth and 47% ready for real work — still pre-alpha.
EffectCraft is a free, open-source After Effects alternative in Rust: about 91% of features exist, self-rated 45% for pro work, with Lottie both ways.
FilmCraft is a free, open-source Premiere Pro alternative in Rust, self-rated about 56% ready for real work at v0.6.0. Tested against Kdenlive and Shotcut.
PhotoCraft is an open-source Photoshop alternative in Rust, self-rated 45% ready for real work at v0.7.0. Tested against Krita, GIMP and GitHub data.
VectorCraft is a free, open-source Illustrator alternative in Rust, self-rated about 65% ready for real work at v0.9.0. Tested against Inkscape and Affinity.
Open-source skill diagram-design sat at No. 4 on GitHub Trending on Oct 12 with 51.3k stars: 44 editorial diagram types as single-file HTML+SVG.
Codex lead Tibo raised GPT-6.1 Sol's default context in Codex and Work from 272k to 500k tokens on October 11, with no long-context fee on subscription usage.
Karpathy's Oct 11 post: ~6B people barely touch LLMs, ~20M pros watch agents do weeks of work from one prompt, ~5,000 see frontier systems inside labs.
US software developer postings are up ~15% since Claude Code launched, but 71% of the growth is senior roles and junior developer employment is down ~20%.
Zhipu's unannounced GLM-5.3-Highspeed is rolling out to random Zcode 3.15.1 subscribers; one user measured 490 tokens/s. No official numbers or pricing yet.
fast-jev-compaction swaps Claude Code's lossy compaction for Jev decisions: stale tool calls and results are deleted or truncated; your text stays verbatim.
Frontend teams meet browsers four ways — embedding, automation, fingerprinting, agent-driving. Their projects, engine trade-offs and honest limits, on one map.
Jev is TypeSafe's decision-only model. A month after launch, ten open-source projects orbit it — from a browser agent to a Postgres extension.
Video.js 10 went GA on October 1: five player projects rebuilt as one and, by official numbers, about 60% smaller than v8. The npm main package is still v8.
REA is an AI reverse engineering tool that connects Ghidra and Hopper to coding agents through MCP. It lowers inspection costs and pressures software moats.
agency-agents organizes 267 specialist agents into 18 divisions and installs them into Claude Code, Cursor or Codex with one script — 17 tools supported.
brag is an open-source agent skill from latent-spaces: one command reads your project and renders a 20-second launch video with music, motion and share copy.
ECC is an open-source Claude Code configuration that ships 71 subagents, 311 skills, hooks and a memory vault — and adapts to Codex, Cursor and more.
From $6,000 agency quotes to $19 a video to free open source: the seven-layer market for product launch videos, and the new code-rendered branch inside it.
An open-source agent skill by Huashu packing 35 art style recipes and 9 explainer grammars to generate frame-accurate Canvas animations with code.
Beta library from refero.design: 2,000+ design systems from real websites, each with a DESIGN.md that Claude Code, Codex and other coding agents can follow.
Five routes for AI agents to control a browser, from Selenium to BrowserSkill. The first fork: a clean instance, or your signed-in profile.
video-shotcraft is a skill that turns Claude Code into a product-video studio: 157 shot recipe cards, a validated Remotion template, sound design included.
OpenClaw creator Peter Steinberger explains his agent workflow on AI Worth Using: fresh-context review agents, local databases over cloud MCP, prompts over PRs.
universal-modder is an MIT-licensed skill pack that lets Claude Code, Codex and Cursor recon engines, read code, generate art with fal and test mods in-game.
SemiAnalysis tested nine AI subscriptions: on the daily-driver models, Claude plans give about 5x the API-equivalent value of OpenAI’s. Here is how to use that.
video-use from browser-use hands video editing to a coding agent: drop footage in a folder, chat your edits, get final.mp4 with filler cut and subtitles.
Seven open-source Adobe look-alikes from ArtCraft: PhotoCraft and six more. We weigh their claims, Hacker News feedback and mature open-source rivals.
Since v2.1.277 Claude Code reads AGENTS.md when CLAUDE.md is absent, and Anthropic’s Thariq says new projects can start without either. Rules, a table, 3 steps.
Dagger's server gives each coding agent its own container and git branch, so several can work in parallel, with command logs, live takeover and git review.
Upstash's hosted docs service. Coding agents fetch current, version-specific docs and code examples by library name, which cuts stale or invented APIs.
Indexes the Claude Code, Codex and Cursor sessions already on your disk so any agent can recall past fixes. Local, one Go binary, no model calls.
GitHub's official MCP server lets agents browse repos, work issues and pull requests, and dig into failed Actions runs. Hosted with OAuth, or local via Docker.
LCU decouples the computer-use runtime built into the ChatGPT desktop app and exposes it to Claude Code, Codex CLI and Pi as MCP tools — no Codex sign-in.
A language-server-backed toolkit that lets agents find, reference, rename and rewrite code by symbol in 40+ languages, instead of grepping and patching lines.
MCP server that sandboxes tool output into SQLite and returns only relevant snippets — a claimed 98% context saving on 17 platforms. Note: ELv2 license.
The official Opus 5.5 prompting guide is mostly about deleting instructions: effort is the main control now. Which level fits which work, and what to cut.
Earendil shipped Pi 1.0 (MIT): a terminal coding-agent harness mixing major-provider models with a decider switching — 111k GitHub stars.
ChatGPT desktop's Chat mode doesn't touch your Codex quota: plan and review there, hand chores to Luna, and a 5 a.m. task turns two daily resets into three.
gstack is YC president Garry Tan's open-source Claude Code setup: 23 slash commands act as CEO, designer and QA, turning one developer into a team.
Mirasim runs Claude Code, Codex and its own agent side by side in one desktop window, with a model picker per session, a built-in editor and bundled quotas.
Opus 5.5 can’t paint a pixel — it writes the program that renders every frame. The routes, real costs and pitfalls from a week of public post-mortems.
Anthropic never advertised video for Opus 5.5. Within a week users made it a category: 1,401 videos, four public bills, one fight over whether it is video.
OpenAI shipped GPT-6.1 Sol at DevDay: near-Astra performance at one-fifth of Astra’s prices ($2/$10 per million tokens), live in ChatGPT Work and Codex.
phone-harness lets Claude Code or Codex control a real iPhone or Android — no jailbreak, nothing on the phone; iPhone rides Mac mirroring, Android runs on adb.
Anthropic released Sonnet 5.5 on September 28: $2/$10 per million tokens (half of Opus 5.5), 30% faster, and better agentic coding than Opus 5.5 in-house.
magpie is a free menu bar app that switches any coding agent's model from one panel: a local gateway unifies APIs, and Claude or ChatGPT logins are shared.
One interface for Claude Code, Codex, Copilot, OpenCode and Pi — agents run on your machine, and you steer them from desktop or phone. Apache-2.0, no telemetry.
OpenMausBot is an open-source desktop chat app that turns the Claude Code, Codex and Grok CLIs already on your machine into bots in a sidebar: each bot gets its own personality, model, memory and a computer to work on, and asks before it acts. Instead of juggling terminals and vendor sites, you get one entry point over logins you already have, with transcripts kept local.
The official Product Hunt connector in Codex, driven by a structured prompt from @goan999999, turns the last 30 days of launches, pricing pages and comment complaints into five opportunities for indie developers — each with competitors, pricing and a 7-day validation plan. For anyone who would rather mine real demand than brainstorm in a vacuum.
VibeKit, an open-source MIT tool from Superagent, wraps coding agents in a local Docker sandbox with secret redaction and observability, 1,861 GitHub stars.