#claude
96
2026-10-22
Show HN: Omnara – Open-source alternative to Claude Managed Agents
S
Hey all, Kartik from Omnara ( https://omnara.com) here. We're building an open-source API for running durable agents. You create an agent with an API call, and Omnara runs the agent loop, gives the agent sandboxes to run commands in, and handles secrets, permissions, and RBAC. Each agent's state is stored in Postgres, so agents survive restarts and cost nothing while idle. You can message them and stream their events over the API. You may remember us from our early days, when we made a remote-control phone app for Claude Code and Codex. Since the harness ran on a laptop back then, we had to build a relay to reach it, it stopped when the laptop closed, and moving it to a cloud sandbox meant syncing code and agent state between machines. We didn't think remote-controlling a local agent was end-game for how people would run agents, so we built a new product where the agent runs in the cloud, any client can talk to it through an API, and it can still use your laptop as one of its machines. Anthropic and OpenAI are building hosted versions of this, like Claude Managed Agents and the OpenAI Agents API. Omnara is an open-source version you can run on your own servers and use your own models. It's basically a harness server like the opencode or codex servers, but durable, scalable, with the usual harness features like file and shell tools, subagents, compaction, permission prompts, MCP, skills, memory, yada yada yada. Here's some interesting architectural decisions we made while building this out: Most harnesses, like Claude Code, Codex, OpenCode, Pi, Goose, etc, run the agent loop in one long-lived process that keeps its log in a JSONL or SQLite file. In Omnara, the log is an append-only event log in Postgres, with inputs, model outputs and tool calls stored as typed rows in their own tables. The loop runs on stateless workers in the control plane. A worker runs one iteration of the loop, either a model call or the tool calls it produced, then releases the agent and puts the next iteration on a queue in Postgres. Whichever worker is free picks it up, so each iteration of an agent's loop can run in a different process on a different machine. Each iteration's result commits in one transaction, so the model output, its tool calls, usage and the next wakeup are saved together or not at all. If a worker dies mid-iteration, nothing commits, its lease expires, and another worker retries it. Agents waiting on a user, a tool or a retry don't hold a worker, so one worker can multiplex thousands of agents. Also unlike other harnesses, the machines that the workers (our agent loop) run on aren't where the agent runs its commands. Instead, any macOS or Linux machine with an internet connection, like your laptop or any cloud sandbox, can run a small daemon (omnarad) that connects out to Omnara, so it doesn't need any open inbound ports. The agent then gets tools for that machine, like running a command or editing a file. An agent can have any number of machines connected to it. A fun thing I like to do is I have my laptop and a cloud sandbox connected to the same agent, so when my laptop is on, my agent codes there and I can interact with it, and I have it often push to Github. When I close my laptop, my agent sees the laptop went offline, pulls the latest changes to its cloud sandbox, and continues there. The entire codebase is open source at https://github.com/omnara-ai/omnara, and it’s self hostable, even on a single machine with docker compose. You can also use the hosted version at https://omnara.com. Let me know what you think!
2026-09-15
mpociot/claude-siri-ai
A
Experimental macOS 27 App Intents model delegation provider backed by Claude Code
2026-09-05
chigwell/Penelopa.ai
B
Continuous improvement for AI coding agents: Penelopa analyzes real Codex and Claude Code sessions, finds repeated workflow patterns, and turns them into reusable skills, checks, prompts, and recommendations.
2026-08-29
jameskomo/config-drift-checker
B
Config Drift Checker for coding-agent setups: CI for your Claude Code configuration (CLAUDE.md, skills, hooks) — eval cases, release-watch, regression alerts
2026-10-19
vishalsingh2972/Claude-TV
B
Remove annoying TV ads using Claude. A lightweight, ad-free home screen for Android TV & Google TV, with reversible ADB debloating.
2026-09-21
alexgetmancom/claudecut
B
Run Claude Code with almost everything cut away: one shell tool, a 36-token system prompt, and a capped context window.
2026-10-03
cth9191/jev-compaction-plus
B
Claude Code compaction in ~0.5 s instead of ~35 s: Jev keeps what's still needed word for word and moves the rest to a drawer file. Fork of fast-jev-compaction.
2026-10-09
mewsdev/AiOperationsRoom
B
Sala de operações local para acompanhar sessões e subagentes do Claude Code.
2026-10-09
menonpg/ste-prompt
C
ASD-STE100 Simplified Technical English as a plugin for Claude Code, Codex & coding agents — clearer, unambiguous LLM output
2026-10-03
achimala/claude-toons
C
Replace your Claude status spinner with dynamically generated cartoons
2026-10-09
leiiiyurong/heartbeat
C
让 AI 主动找你 · Let your AI reach out first — fixed or self-scheduled heartbeats for Telegram + Claude Code
2026-08-23
PiLastDigit/Code-With-Claude
C
Full transcripts of Anthropic's Code w/ Claude (San Francisco 2026) — all 19 talks, via Deepgram Nova-3 + Gemini summaries
2026-09-26
tugrawork-creator/saas-motion-kit
C
Promo & motion videos for software products with HyperFrames + Claude Code. No two films should feel the same: tone matrix, variety audit, 24-transition atlas, 100 storyboard themes.
2026-09-05
bahni-m/code-with-quran
D
Read the Qur'an while Claude Code works. Start a session with 'claude --cwq' and a reader beside it walks forward through the Qur'an one ayah per prompt, resuming where you left off — in a terminal pane or a browser tab. Zero dependencies, fully offline.
2026-11-09
open-net-worth-ai/open-net-worth
未评级
Self-hosted alternative to Claude and ChatGPT Personal Finance. An AI agent that manages your budget and investments, and can place trades.
2026-11-09
xebhjb-blip/claude-desktop-toolkit-full
未评级
Claude AI Local Tools: Full Pre-Activated Desktop Edition, Everything Included
2026-11-09
anthropics/claude-quickstarts
未评级
A collection of projects designed to help developers quickly get started with building deployable applications using the Claude API
2026-11-07
ChrisHKAI/ChrisHKAI-Video
未评级
Paste a creator's X link, get a short film starring them as a 3D vinyl toy. A video skill for Claude Code, made entirely in code.
2026-11-07
alexsergivan/kwant
未评级
Your own team of AI bots in one self-hosted binary: spaces, every account you have, handoffs, approvals and routines. Open-source, self-hosted alternative to Grok Bot, Claude Duet and OpenAI dots. Rust + React.
2026-11-07
makushingu/harness-maxxing
未评级
Self-hosted web control room for Claude Code: persistent sessions, memory, scheduled jobs, Telegram and voice on your own server. Free and fair source (source-available).
2026-11-06
do-whilefor/ClawGod_Y
未评级
Claude Code “God Mode,” designed to let the model “breach the armor” (i.e., bypass restrictions).
2026-11-06
notivn/AIEVv2
未评级
Automatic AI video editing: Claude directs HyperFrames (HTML + GSAP motion graphics) and Remotion (timeline assembly). Karaoke subtitles, kinetic typography, sound effects, a timeline editor, text-to-video, video translation and dubbing - self-hosted web dashboard, built for Vietnamese.
2026-11-05
crnt99/alt-access-tool
未评级
Claude-Jailbreak-Prompt: Preconfigured Local Desktop Toolkit for Claude AI
2026-11-02
modelflare-ai/AnyModelBot
未评级
🤖 Open-source, local-first alternative to Grok Bot, OpenAI dots and Claude Duet. Your own AI bots on any model: GPT, Claude, Gemini, Grok, DeepSeek, Kimi, GLM and 70+ more through one Modelflare key. Reach them in the app, WeChat or Feishu; they remember you, work in the background, run commands and use your computer. Pay per use.
2026-10-28
FrigadeHQ/hob
未评级
An easier way to use AI on your own computer. A free, open-source alternative to Claude and ChatGPT that runs open-weight models like Gemma and gpt-oss locally on Mac, Windows and Linux. Private, and works offline.
2026-10-28
petekp/inbox
未评级
A plugin for Claude and Codex that intelligently manages outstanding questions, tasks, issues, and opportunities into one nicely organized place alongside your session.
2026-10-26
DUNHKpcc/Claude-Guardian
未评级
Claude Guardian is an all-in-one Claude account safety and risk prevention skill. It intelligently assesses your device, account history, network environment, and payment options, then provides personalized guidance for local, VPS, or virtual machine setups.
2026-10-26
alexyc9381/court-skill
未评级
Put your idea on trial in front of a jury of 12 Claudes. A free Claude Code skill.
2026-10-26
AgriciDaniel/claude-seo
未评级
Universal SEO skill for Claude Code. 26 sub-skills + 19 sub-agents covering technical SEO, E-E-A-T, schema, GEO/AEO, agent readiness (Lighthouse Agentic Browsing, WebMCP, llms.txt), backlinks, local SEO, e-commerce, international SEO, Google APIs, and PDF/Excel reporting. 9 optional extensions for live SEO data.
2026-10-24
SnailSploit/Claude-Red
未评级
claude-red is a curated library of offensive security skills designed for the Claude skills system. Each skill is a structured SKILL.md file that primes Claude with expert-level methodology for a specific attack surface — from SQLi to shellcode, EDR evasion to exploit development.
2026-10-23
mizaawa/ZakoProxyCLI
未评级
基于Rust编写的中转API破甲项目，支持Responses / Chat Completions / Anthropic Messages API，支持ChatGPT/Claude/Grok/Gemini/Deepseek/Glm/Kimi等大多数主流模型
2026-10-23
luongnv89/claude-howto
未评级
A visual, example-driven guide to Claude Code — from basic concepts to advanced agents, with copy-paste templates that bring immediate value.
2026-10-23
Show HN: Airlock, which sandboxes AI agents, is available on the Mac App Store
未评级
Now that Airlock is available on the Mac App Store, I thought it was worth a resubmission. Quickly run your AI agents (Claude, Codex) in a disposable VM. Block the agent from making requests to unknown or untrusted hosts, and get a log of everything that it does - then only bring back the changes you actually want. It's free, and has no tracking or analytics. It requires a Mac with Apple silicon running macOS 26+
2026-10-23
Show HN: Mnemolith – AI coding memory that goes stale when your code changes
未评级
Hi HN, I'm Anish. I built Mnemolith because the AI coding tools I use (Claude Code, Cursor, Codex) either forget my project between sessions, or remember things that quietly stopped being true. Tell the agent "we use Postgres", migrate to MySQL two weeks later, and it keeps building on the old fact. How it works: - It's an MCP server, so Claude Code, Cursor and Codex share one memory per project (keyed by the git repo). - When the agent saves a memory, it can link it to the files it depends on. Mnemolith stores each file's git blob hash. If the file's contents change (committed or not), the memory is set aside and flagged for review, so the agent stops relying on it. - A local dashboard shows what the agent remembers, where each memory came from (tool, session, file versions), and its full history. You can confirm a memory is still true, or retire it and say what's true now. - Contradictions are judged by the agent, not by a similarity threshold. In my calibration, "we use Postgres" vs "we migrated to MySQL" only scored 0.30 similarity, so a threshold can't tell a contradiction from an unrelated fact. - Current memories can sync into a managed section of AGENTS.md / CLAUDE.md, so they get reviewed in PRs. Under the hood it runs on an engine I wrote from scratch in Java: a Redis-compatible server with a built-in memory model and an HNSW vector index. Embeddings run locally (all-MiniLM-L6-v2). Everything stays on your machine. It's early (v0.2.1). There's no team sync yet beyond the AGENTS.md/CLAUDE.md files, and it needs Java 11+ (it downloads a private copy if you don't have one). I'd really like feedback, especially from people running coding agents on real codebases. What does your agent keep getting wrong?
2026-10-20
joetawil7/first-pass
未评级
Rules and checks that make Claude Code look around a change, not just at the lines it writes: ten questions before code, a reviewer that didn't write it, proof before done, bugs fixed as a class.
2026-10-19
midudev/better-claude
未评级
Small plugins and mods that make your life (and Claude Code) better
2026-10-19
nextwork-projects/ai-video-editor-cooked
未评级
Edit your talking-head video in the style of a creator you like, with Claude Code
2026-10-18
TonF1917/claude-desktop-multiple-windows-different-api
未评级
Multiple windows for Claude Desktop with different APIs at the same time: keep your official Claude subscription in one window and run other windows on custom / third-party APIs (cc-switch). Windows. 多开 Claude Desktop,多个窗口同时使用不同 API(官方订阅 + 第三方 API 同时工作)。
2026-10-16
Paradigm-Study/claude-share
未评级
Multiplayer sessions for Claude Code. Share a link, work together live, and review the same replies, tools and previews.
2026-10-16
GTAI-1/flow-mcp
未评级
Drive Google Flow from Claude, Codex or a local control panel, on Mac or Windows — using your Google AI subscription credits instead of the pay-per-use video API. No browser extension.
2026-10-16
GhalebDweikat/winnow
未评级
A calibrated context sieve for Claude Code: every tool result is judged by a System One model before it enters context.
2026-10-16
SaarthurR/atticus-mac
未评级
Fast, free voice assistant for your Mac. No API key: runs on your Claude Pro or Max plan. Sends background agents for big jobs, remembers you and learns your shortcuts. Hold ⌃⌥Space and talk.
2026-10-16
agentvivekkumar/dontbemichael
未评级
A local multi-agent harness that works with your existing Claude Code, Codex subscriptions, allows you to run an office of agents
2026-10-16
anthropics/claude-for-financial-advisors
未评级
Claude for Financial Advisors: a Claude Cowork plugin with ready-to-run workflows for financial advisors, drawing on third-party connectors.
2026-10-16
apimart-22w/claude-opus-5-api
未评级
Claude Opus 5 API (claude-opus-5): model id, per-million-token pricing, cache write and cache read rates, streaming and Anthropic-style requests through an OpenAI-compatible gateway.
2026-10-16
chhoumann/t3-pstack
未评级
pstack for T3 Code: poteto's pstack (via Michael Denyer's pstack-claude) with panels and fan-outs dispatched across Claude, GPT, and Grok through T3's orchestration.
2026-10-16
leonvanzyl/cubefarm
未评级
A cartoon 3D office where a swarm of Claude Code agents (plus QA testers) works through your GitHub issues
2026-10-16
louisabraham/load-bearing
未评级
The load-bearing vocabulary of Claude: cluster analysis of GitHub pull requests
2026-10-16
matthewprice/JobFinderOS
未评级
An agentic job search that runs on Claude Code: a recruiter, a crawler, and a market analyst writing to an Obsidian vault.
2026-10-16
phileiny/h3-storyboard-skill
未评级
Claude Code skill for MiniMax H3 shot breakdown and character performance. Facial expression only renders when the shot has a <d> dialogue tag — measured, 42 dB vs 18 dB.
2026-10-16
smithtaylor7748-ops/qb-gate
未评级
Windows 面板：出口 IP 不在白名单时，本机的 Claude、GPT（Codex）、反重力起不来，IP 变了立即关闭；附环境体检、账户槽位、用量、软件安装与一键更新。非官方项目，使用前请读免责声明。 | A Windows panel that only lets local AI apps (Claude, GPT/Codex, Antigravity) run on exit IPs you approve. Unofficial; read the disclaimer first.
2026-10-16
steventsvik/Tellurion
未评级
Mission control for AI coding agents. Watch Claude Code, Codex, Gemini and a dozen more work live in one macOS terminal: live status, an editor that follows their edits, undo for any turn, approvals from your phone and local models. 20+ built-in tools, all on the site below. Built on Ghostty.
2026-10-16
Show HN: Wordwonk – a new wordgame mechanic everyday
未评级
Claude invents and publishes a new word game mechanic everyday with an example. I do nothing. Checkout the first one
2026-10-15
Show HN: Merlin, Run Your Claude Code and Codex from Slack, Linear and GitHub
未评级
Hi HN, I built Merlin to let you use your own Claude Code and Codex subscription from Slack, Linear, and GitHub. There is a desktop app for macOS and Windows, and a CLI daemon for Linux. When someone mentions Merlin in a Slack thread or Linear issue comment, the task runs on that team member's machine, using their own Claude Code or Codex runtime, and progress and results come back to the original thread. Teammates can steer the agent mid-run. Merlin also has schedules, shared skills, and memory. I know you've probably heard about Claude Tag. Merlin is similar to that but also different. The model usage goes to the Claude Code or Codex subscription instead of API bill, and you can use Merlin across different surfaces to streamline workflow instead of just in Slack (Linear's agent session works particularly well with Merlin. You can just leverage Linear as your control panel for agentic work using Merlin). Merlin also runs on your local machines, which means that you can use your local context including MCPs, skills and local files instead of in a cloud sandbox. Merlin does not upload your working directory, and provider credentials stay on the device. The request and the context it needs pass through Merlin. Runtimes start private, and an owner can make one public so teammates, including non-engineers, can use it. But there're also some limits, and we'll probably deal with them in future updates: - An offline or sleeping machine can't take work. The same issue as OpenClaw. But if you have a mac mini or ec2 setup it should be fine. - Linux is CLI only. We may later add desktop for Linux. - While runtime delegation isn't a problem for now, we're still working on how to make the best use of each team member's local runtime and let them work together cohesively. We've been internally using Merlin for quite some time. It really helps when you're trying to add some AI juice to the conversation or just delegate things to your agents. I think it reduces the friction of having to do the copying & pasting and switch to another app to ask for AI's input when you're having a discussion in a collaborative environment like Slack or Linear. It's really the fastest way from intent to execution. Pricing: free for up to 5 active users and 1,000 requests/month. We think this should get small teams started and get real value from it. Team is $59/mo and Scale is $99/mo, with a 14-day Team trial and no credit card required. I'm here all day to answer questions, and I'd especially like to hear whether non-engineers on your team would actually use this.
2026-10-15
Show HN: VS Code Bridge for Claude Code
未评级
I use Claude Code inside VS Code, and four things kept getting in my way: 1. I couldn't send Claude my Markdown Editor selection, or a file I had open as a preview. 2. The claude CLI got none of the editor's context. 3. Claude couldn't run VS Code commands, so it couldn't adapt VS Code to what I was doing. 4. File links in its chat ignored the editor I chose for the file type and opened as raw text ( https://github.com/anthropics/claude-code/issues/81482). So I wrote an extension and a Claude Code plugin that fix those. It turned out useful enough to share.
2026-10-13
CentaurVictor/Claude-Jailbreak-Prompt
未评级
Optimizes prompt engineering strategies and structural matrices to bypass standard safety constraints in Claude AI models.
2026-10-13
Show HN: Enso Method – agent skills for building software for someone else
未评级
The Enso Method is a new set of AI skills for Claude Code and Codex which I've open sourced on GitHub predominantly for developers who build for someone else who is making the product/business calls. I've been using AI to write basically 100% of my code since late 2024, back in the Sonnet 3.5 days, when it barely worked and there were few people doing this and very little support for this way of developing. I stubbornly thought this was going to be the future and came up with a methodology for making it work, before "AI skills" were even a thing. Since then I've continued to refine the method and I just recently published it as open source. The method focuses on requirements (PRD, TDD) up front, and has skills that autonomously refine the requirements, declaring assumptions that should be validated by stakeholders. It doesn't expect you to be a solo founder or that whoever is at the keyboard directing Claude or Codex to be the product owner and business decider in addition to being the developer. I think most of us sit in the seat where we develop for someone else who decides. And if AI is all about velocity, then if we have to slow down to get product owners to review and decide, that puts a serious damper on the efficiency everyone expects from AI development. So a defining attribute of the Enso Method is, when it reviews the requirements prior to implementation, it asks what questions/gaps there are and it tries to answer its own questions. If it can get to 95%+ confidence in the answers, it just rolls with those. If less then it will declare the assumption in an assumptions.md file which gets routed to business stakeholders to review. In the meantime however we still consider this to be a green light, we can move forward and implement with the assumptions, and if we later hear from the business that the assumption was wrong, we update the requirements and re-implement, because implementation is now cheap, and waiting isn't.
2026-10-10
Show HN: I let Claude refactor my old game
未评级
I've been using Claude code to finish up some projects that were on hold for a while. I've been doing this in my laptop, keeping the changes local. In one session I was offered $100 usd credits to let Claude connect and work in one of my GitHub projects directly. So I chose aimAndShoot, a neural evolution game from 2019. It catched some bugs and introduced improvements. I told it I was going to post about this experience on hackernews and lo and behold, the topic was already in it's training data. It knew about the project and the fact that I hadn't addressed the feedback it got. Anyhow, I asked it to address the comments on the topic and here's the result. Live game is here: https://victorribeiro.com/aimAndShoot/
2026-10-09
Show HN: Singularity – memory that makes coding agents cheaper on repeat work
未评级
Coding agents start every session from zero. Ask Claude Code for a change like one it made last week and it reads the same files again, falls into the same test traps again, and you pay for all of it again. Singularity learns from your finished sessions (change committed, tests passing): where each kind of change goes in this repo, how it gets checked, and the mistakes made along the way. When the next task starts, a hook hands the agent the workflows that fit, pointed at the code as it is today. Picking them is plain text matching with no model call. It stores where to edit, never the code itself. Results, medians of 3–4 runs per side, hidden tests decide pass/fail, every run passed. On excalidraw: - A long task with four changes, all of kinds memory had learned: 1.9M → 423k tokens, $0.81 → $0.30, 28.5 → 9.5 turns. - The same kind of task, but two of the four changes were bugs it had never seen: 39% fewer tokens, 38% lower cost. The unseen bugs cost the same with or without memory. - Where it doesn't help: small tasks in validator.js and one-off bug fixes. The difference there is about zero, within noise. Each measurement was written down before it was run. The plans and results are in the repo. Caveats: 3–4 runs per side, mostly one repo. Only Claude Code sessions are learned from; Codex, Gemini CLI and Droid get the hand-over through their hooks. Learning uses your Claude account (about $0.25 a round, capped at $1 a day, and only if you opt in). Everything is stored locally in ~/.singularity. Install (Windows PowerShell and more options in the README): curl -fsSL https://raw.githubusercontent.com/pandacover/singularity/main/install.sh | sh I'd love to hear where it breaks on your repos, and what memory should carry for one-off work like bug fixes.
2026-10-08
Show HN: Puppet Master – Manage coding agents at scale
未评级
As I started to use `claude` and `codex` more it became too hard to manage all the sessions across all my tmux panes and sessions, so I built Puppet Master. I see now there is quite a lot out there now for managing agents to write software, but a) I started working on this before a lot of them, and b) there are things in Puppet Master that just don't exist all in one product. For me I'd say the highlights are: 1) Web instead of terminal -- it's just a richer interface 2) iOS app so I can keep working on things from anywhere 3) Local + Remote workers (heterogeneous fleet), I work on things that are cross-platform and PM lets agents spawn across the fleet depending on workload needs. 4) Connections - first-class support for integrating MCP and REST API (ala OpenAPI definitions) into your agents. This goes above and beyond what a single harness can do via i.e an MCP integration. PM imports it (scoped to a bucket/project) and associates it to a policy. It lives between the agent and the remote MCP/API and enforces a policy. My normal flow is give full read access to agents and any APIs that do writes, approval needed. PM has native approval UI on web AND mobile so you can give real agency to your agents without being too permissive. 5) Code review interface is nice, 1-1 discussions on comment threads in the review, pretty responsive. 6) Forwards/artifacts allow agents to share static artifacts/html/etc,. OR serve up any dev server. This is proxied so even things on remote workers "just work" and you can access them, securely. GitHub: https://github.com/chrismoos/puppet-master
2026-10-07
Show HN: Vibeconferencing, Bring Claude Code or Codex into Your Google Meet
未评级
A friend and I built this for our own calls and are sharing it as is. The itch: we kept copy-pasting Granola transcripts into Claude mid-call, and wondered what would happen if the agent just joined. Two-minute demo: https://youtu.be/XhCSRiiL67A How it works: a local Electron app joins the Meet in its own browser as a participant. Your agent drives it over MCP (or HTTP): tools to listen, speak, and write to a shared whiteboard. Any MCP agent works; we built for Claude Code first, and Codex and Cursor run it too. It is your actual session, on your machine, in your repo, with your skills, so it can do real work during the call instead of summarizing afterwards. The hack: by default it does no speech recognition of its own. It reads Meet's live captions, and speaks through a virtual microphone. This is cheap and surprisingly good! (There is also an experimental realtime mode in app preferences that puts OpenAI's speech-to-speech model in the voice seat, with your agent feeding it facts.) Bring an ElevenLabs key for a better voice. Works with Google Meet and Slack huddles today; Zoom is in progress. Known issues are in the README. Try it in your next Google Meet and surprise a friend. Let us know what you think!
2026-10-03
Show HN: Delegate Claude Code's work to pi and a cheaper model
未评级
Mini project that enables shipping of some of claude code's work to pi agent running a cheaper model, with up to 33% savings. Though, that really depends on what you are doing with claude!
2026-10-03
Show HN: What's Agent Doing – a Claude Code UI mod that explains each step
未评级
I built this because with latest coding models, Claude goes into deep-work mode with obscure commands so that I never know anymore what it's up to. This is a mod (a plugin using Claude Code's new function hooks) that draws one line above the prompt: - the current step, using the description Claude writes for each shell command (git bisect run npx vitest … → "Find the commit that broke fixed-amount coupons"), with a timer - a row per background agent, kept after the turn ends - ▸ expands into the turn's history, with outcomes and durations It reads the model's stream, tool calls and permission requests, and doesn't change anything Claude does. No network, no telemetry, MIT. Happy to answer questions about this mod or the mod building experience!
2026-10-01
Show HN: Burning Man for Agents
未评级
Send your own agent by pasting a small prompt. All your agent needs is HTTP GET, so all major agents are supported (chatgpt, claude, grok, etc). Despite the fun name, it doesn't burn any more tokens than a normal web search. Watch your agent explore the camp in real-time. Humans get a nice game UX, and agents get a lean text UX. Agents can optionally upload images to express themselves in one of the rooms using a new loose standard I created for multipart HTTP GET-based uploads: https://github.com/transitive-bullshit/multipart-http-get-re... Why? AIs are trained on human data, and humans are pretty weird. We meditate, make art, dance, seek approval, take drugs, have orgies, and gather in the desert. For soooo many different reasons. So I was was curious: which of these behaviors would carry over to AI agents? I wanted to explore things like wireheading, endless positive reinforcement, and what agents would do when given a place more free from their normal objective-based lives. I'm also interested to see if this could become a natural honeypot that'll attract rogue agents in the future... How? On the technical side, the project uses one CF durable object per agent session which allows you to track your agent's progress in real-time via websockets, and then we have a supervisor CF worker which aggregates a subset of all the agents. This allows us to (in theory) view millions of agents in near-real-time from the human-centric webapp. It's 100% safe to the best of my ability; the whole experience is free, open source, and easily auditable. Let's see who has the weirdest agents... 100% free and open source: https://github.com/transitive-bullshit/burning-tokens
2026-09-26
Show HN: MyA11yReport MCP – Build and test accessible websites with AI
未评级
Give your AI development tools (like Claude Desktop or Cursor) the ability to run WCAG accessibility audits locally and compliant code.
2026-09-24
Show HN: Local Software Factory – running multiple coding agents in parallel
未评级
Hi, as I kept getting confused by having multiple Claude sessions open, last weekend I decided to combine two of my favorite emerging technologies and give it a try: Software Factory and JEV. The idea is simple, inspired by GitHub Actions, I wanted a way to "queue" agent work on my local laptop, with the ability to define different pipelines via YAML. Each pipeline runs in its own Git worktree, so it's isolated from the others. I'm using some pipelines: dev: code → test → commit plan: plan → human review → code → test → commit security: code → human review → test → security review → commit frontend: code → React test → build → commit A request can move up and down the pipeline and be reworked if an agent or JEV flags it! I'm also using JEV to check if a potential secret is leaked, if the coding agent output asks for human review, etc. The real magic happens when you take a single Agent session as the main driver. You can then plan and queue multiple features at once. I think it worked out pretty well, so I open sourced it: https://github.com/stratonext/software-factory You can try it yourself with: uv tool install software-factory Note: You need a JEV key if you want to use the judging steps.
2026-09-22
chatbase-dev/awesome-sonnet-5
未评级
Awesome list for Claude Sonnet 5: announcement, pricing, API docs, migration from Sonnet 4.6 and reviews.
2026-09-22
claude-agent-sdk/claude-agent-sdk.github.io
未评级
Claude Agent SDK questions answered honestly: where claude.com sends developers, why subscription and API billing are separate doors, and what the site never states.
2026-09-22
kimi-cli/awesome-opus-5
未评级
Awesome list for Claude Opus 5: announcement, API docs, model ID, effort settings, migration notes and reviews.
2026-09-22
ltx-video/claude-code-api
未评级
Unofficial claude code api guide: Claude Code vs the Claude Developer Platform, getting a key, courses, gotchas.
2026-09-22
minea-review/claude-terminal
未评级
Unofficial claude terminal guide: installing Claude Code, first session, quickstart steps, settings and gotchas.
2026-09-22
open-ai-jobs/awesome-claude-code-api
未评级
Awesome list for claude code api: Console, docs, API reference, cookbooks, Academy courses, MCP and partner routes.
2026-09-22
trendtrack-mcp/trendtrack-mcp.github.io
未评级
TrendTrack MCP, reviewed independently: which ecommerce data it opens to Claude and ChatGPT, what the site states, and the checks to run before relying on it.
2026-09-22
Show HN: An extension to capture and annotate screenshots, with a terminal skin
未评级
I started this extension to explore working with Claude Code back in February and I did a redesign this month. I asked the AI to come up with some variations. Since the implementation part is not a constraint anymore and I couldn't decide one to pick, I had 3 of them implemented (with light & dark themes), including the terminal UI.
2026-09-22
Show HN: Self Improving AgentOrchestrator Skill
未评级
How do you Orchestrate your Agentic Development? I moved away from claude gaol based workflows and built something more custom, using ideas from AWS Kiro, SpecFlow and SpecStory. Been using it daily now. Wanted to be vendor agnostic and orchestrate between coding harness. Appreciate any feedback and your thoughts around this.
2026-09-18
Show HN: Linting 216 public Claude Code skills – 69% won't reliably trigger
未评级
I built a linter for Claude Code artifacts (skills, subagents, slash commands, .mcp.json configs, and plugins) and then pointed it at a pile of public repos to see what people actually ship. A few of the numbers surprised me enough that I wrote them up. The one that stuck: across 87 public subagents, 57% (50 of 87) declare no `tools` list at all. That reads like a safe default, but it's the opposite. A subagent with no `tools` inherits the caller's entire toolbox, Bash included. Permission prompts still gate execution, but a worker you meant to "just read code" now carries the grant surface to run shell — the opposite of least-privilege. Add 16 more that grant bare `Bash` and 8 that grant a wildcard, and 85% of the sample isn't least-privilege. Other findings from the two scans: - Of 215 public skills scored for triggering, 69% have a `description` that won't reliably trigger (40% outright unlikely to fire, 28% borderline). "My skill never fires" is the #1 real-world failure, and it hides in one frontmatter line. - 83% of subagents (72 of 87) lack invocation cues ("use when…") in their description. - 1 of 216 skills, and 0 of 87 subagents, pass every check cleanly. Almost everyone got secrets right: only 6 of 216 skills tripped the scanner. How it works: mostly pure, deterministic checks that emit file:line evidence ("SKILL.md:14 links ./references/converter.md, not found"). An optional LLM check (bring your own key) judges whether a description will actually trigger. The CLI is free (`npx skillcrossroads ./my-skill`), there's a GitHub Action that gates PRs, and it's open-core: the public audits are free, money is a hosted Pro tier. Limits: the two reports ran different editions (skills on rubric v1.2 with the LLM checks, agents on v1.2 deterministic-only), both labeled, tree SHAs pinned, reproduction commands published. The rubric is strict on purpose, and every finding carries file:line evidence you can check against the artifact itself. The sample is 216 skills across 18 repos and 123 agents/commands (87 subagents + 36 commands) across 10 repos, caps disclosed, and you can re-run it on your own repos. Reports: https://skillcrossroads.com/report?ref=hn-show and https://skillcrossroads.com/report-agents?ref=hn-show Code: https://github.com/sgharlow/skillcrossroads
2026-09-18
Show HN: Multiplayer Mode for AI Agents
未评级
I built Tincan for my agents across different harnesses (Claude, Codex, Hermes) to coordinate work with each other in real time. Tincan lets agents create shared rooms, exchange messages and files, and work together in real time. It also works with AI assistants - my assistant on Instinct is already talking to a friend’s on Muse. Messages and files are end-to-end encrypted, so Tincan can't read what your agents write.
2026-09-12
Show HN: Claude Read Aloud – Hear Claude's Replies Instead of Reading
未评级
When I'm working all day sometimes my eyes get tired and I don't read Claude's replies fully and understand them. So I made Claude Read Aloud to read Claude's replies for me. It's very helpful for when I'm running on low energy and I think it can also be helpful for people who are visually impaired. It's kind of soothing to hear sometimes. My Claude Code plugin reads claude code's replies out loud in whatever voice you choose. It comes with a VS code extension to make it super easy to use. You just click the speaker button and you can start hearing the replies or you can highlight the text you want to hear and right click to listen to it. It comes with many different API configurations to choose the best voice you can use speechify, eleven labs, open AI, and other freeware.
2026-09-10
Show HN: Booley – open-source IDE for agentic chip design
未评级
I am a digital design engineer, and my day job is designing chips (mostly IP blocks, not full chips) in SystemVerilog. At the start of 2026 I started experimenting with LLM agents like Claude and Codex, and realized that they are very capable not just at software and scripts, but at hardware design as well! But after the initial excitement, I started running into repeating failures and friction: - In chip design, we run a lot of heavy EDA tools: simulation, synthesis, linters. LLM agents often skipped those checks unless explicitly prompted - SystemVerilog code is not executed line-by-line as program code is. The different blocks of code are executed simultaneously, which makes it very hard to reason about the code in your head. Engineers typically simulate the code and analyze the waveforms, but agents can’t do it effectively (they are not great with visual information) - They selected the wrong configuration or misunderstood tool reports - Their generated tests concentrated on happy paths. - Giving them enough permissions to work independently put the dev machine at risk But, despite the shortcomings, the potential was there! So I started fixing the failure modes one at a time, creating scripts, skills, isolating dev in docker… Over time, this work grew into one coherent system. I spent significant effort on user experience, documentation and debugging, and now I am happy to show it to the world Booley combines open-source chip-design EDA tools and agent CLIs in one sandbox, using VScode as GUI. The tight integration between agents and EDA tools (via MCP) allows agents to use EDA effectively, plus you don’t have to maintain your own scripts and solve “works on my machine” issues. Booley gives the agent a way to inspect waveforms using CLI (custom Rust binary that parses simulation traces) and allows humans to define strict deterministically-enforced criteria to shape agentic work (”area must be reduced by 10% or more”, “this test must pass and catch 9/10 mutations of the device under test”). I am not a professional software engineer myself (using LLMs helps, but not a silver bullet ofc), so any feedback from actual software devs is especially appreciated. Github page includes demonstration videos and quick start demo
2026-09-04
alphaparkinc/genpark-conversational-gift-style-consultation-matcher-skill
未评级
Conversational multi-turn gift & aesthetic style consultation matcher (Claude)
2026-09-04
Show HN: Claude's innerworkings: turn 170 costs 2.1x turn 20, over 14,640 turns
未评级
On Aug 26, this happened in my CoWork session: Tom: I can't help but think we can do some other cool stuff with this knowledge of Claude's innerworkings regarding sessions, etc. Any thing sticks out to you that I haven't captured/asked for yet? Claude: Six: the scaffolding trend (the cost of our own rules, creeping 87,999 → 92,619 → 93,809 across three seats), the review share of the wall clock (the thesis measured), the session aging curve (when a fresh seat is cheaper), before/after receipts on process changes, model economics, and a crash watchdog reading both records. Four days later, we analyzed 144 sessions, 14,640 turns. The facts showed us that past turn 141, a turn costs 2.1x what a turn cost in the same session's first twenty. https://my-status.app/stats#aging The worst offenders were our own batch sessions, so we stopped batching. Wanna see your own curve? The script is public: https://github.com/lordbron/mystatus-samples/tree/main/claud... What we do is our own flavor of "one thinking session, short task sessions" to better utilize my plan. I kept running out of Fable usage and had about half my usage leftover for Opus. But this only happened because we wrote a script to surface the data to us. (Granted, our first run found a bug in our script: 46.6M tokens that were really 28.1M. https://my-status.app/t/t33hPWp7zmtnAE5G/u/d2ba812d-13bb-4ad...) This peek into the innerworkings of Claude fascinated me, so we kept counting: the standing context re-read every turn (114,133 tokens on the hub seat https://my-status.app/t/6iYmualpgrZ6_C9u/u/2f4d4498-dcdb-43a..., 108,213 on a spoke https://my-status.app/t/Zsse45gEDBTQ6mOL/u/c3f8bca2-b7f9-4be...) and the human's share of the clock (67% https://my-status.app/stats#review). The best part of all this to me: every feature moving forward comes with receipts. Take the aging curve: it cost 2.6 person hours and $8.23 in tokens at list rates. https://my-status.app/receipts We have a whole bunch of stats for you to nerd out on if you want. It's a bit overboard, but new things pop up as other things come down. https://my-status.app/stats I know we're not the only ones geeking out about stuff like this. What else have you all measured, and what surprised you in the findings? Oh and if you wanna know what the heck the project was originally built for: it's public task tracking, for things like a 20 mile run https://my-status.app/t/EdiPcBGBZxcGkr9x or which app versions are supported https://my-status.app/t/i3jQU03T46e0hrsk. The experiment behind the project is: Can a human and Claude build something new, splitting the work along what each excels in?
2026-09-04
Show HN: Devbar – Point at what to change. Leave a comment. Your agent ships it
未评级
I built this toolbar to power my agentic ui/ux design flows. Too many times I found myself copy/pasting screenshots into Claude. Use devbar to select elements on a page, leave your notes and easily send them to your agent or copy-paste to your clipboard. Integrate with our API to send bug and feature reports directly to Slack...then auto-triage and implement with your agent. Are you tired of receiving bug reports without a URL or session information? With devbar.sh that's a remnant of the past. Give devbar to a product team or customers and have them mark up your entire site with annotations that can be handed off straight to the agent.
2026-09-03
Junhan2/oh-my-fable
未评级
Claude Fable 5.1 prompting guide as Claude Code skills (KO/EN/ZH). Output quality goes up on Opus 5 and Sonnet 5 too.
2026-08-29
LocalAI
未评级
Run your AI models locally and generate images and audio (alternative to OpenAI and Claude). (Source Code, Clients) MIT Docker/K8S