SignalScribe

Wednesday's AI brief

32 stories · 36 discoveries · 24 also noted · writer=heuristic · X 8 · YT 6 · GitHub Awesome 2 · PH 8 · HN 8

Coverage: Stories written by heuristic (LLM off or unavailable)

If you only have 90 seconds

  1. Launch HN: Hoplite (YC S26) – Effortlessly deploy cloud coding agents · Hacker News
  2. The Framework Putting You Ahead of 99% of AI Users · YouTube
  3. Show HN: A faster coding agent than Codex and Claude Code · Hacker News
  4. Show HN: Product analytics (and evals) for agent sessions on your MCP · Hacker News

Discoveries

36 linked tools / repos / products

Primary reason this brief exists — open anything new you might adopt.

tool Claude Coderepo Nutlope/aicommitstool Buzzproduct Githubproduct Betterstackproduct Betterstackproduct Instagramproduct Tiktokproduct Linkedinproduct Openaiproduct Aiheroproduct Aiheroproduct Aiherotool Codexproduct Gooproduct Gorepo cloudflare/computerrepo ShenSeanChen/waku-agentrepo inspatio/querysplatrepo firecrawl/anydocrepo HaidarJbeily7/cargo-friskrepo sophiamyang/finger-frame-effect-airepo genspark-ai/genofficerepo ben-z/findphonerepo MakazhanAlpamys/Souprepo guillermolg00/morphiconsrepo twalichiewicz/Backchannelproduct Producthuntproduct Producthuntproduct Producthuntproduct Producthuntproduct Producthuntproduct Producthuntproduct Producthuntproduct Producthuntproduct News

X

8

Thought leaders & lab accounts

gent SDK, specifically calling out managed sandboxing.

  • Practical product/tooling detail for deploying coding/AI agents securely without users handling isolation themselves.[[3]](https://x.com/trq212/status/2084782162033910216) ### 4.
  • @trq212 (Thariq – Claude Code at Anthropic) - **Full post text:** “Do you use automode?
  • It is very very resistant to prompt injection” - **Exact status URL:**
  • improvements for third-party agent testing.
  • Directly relevant to AI agent safety, evaluation rigor, and operational controls.[[2]](https://x.com/OpenAI/status/2084747580693426555) ### 3.
  • @trq212 (Thariq – Claude Code at Anthropic) - **Full post text:** “Claude managed agents gives you a lot of the benefits of the agent SDK but we handle all the sandboxing and stuff” - **Exact status URL:**
Tools, repos & products
  • Claude Code tool — @trq212 (Thariq – Claude Code at Anthropic) - **Full post text:** “Do you use automode?
Agentic codingToolsSecurityHas links
Primary post ↗ · Post 2 ↗ · @trq212 · 2 posts combined

des video demo.[[3]](https://x.com/nutlope/status/2084689608667804017) **Post 2** > AI Commits just got a lot faster wit…

  • > ◆ 95 file PR went from taking 71s -> 13s, over 5x faster > ◆ Full support for local models (@lmstudio & @ollama) > ◆ Added support for DeepSeek V4 flash > Give it to your agent to get super fast commit messages → https://github.com/Nutlope/aicommits **URL:**
  • Mentions: Nutlope/aicommits, Ollama, DeepSeek.
Tools, repos & products
  • Nutlope/aicommits repo — > ◆ 95 file PR went from taking 71s -> 13s, over 5x faster > ◆ Full support for local models (@lmstudio & @ollama) > ◆ Added support for DeepSeek V4 flash > Give it to your agent to get super fast commit messages → https
  • Ollama tool — > ◆ 95 file PR went from taking 71s -> 13s, over 5x faster > ◆ Full support for local models (@lmstudio & @ollama) > ◆ Added support for DeepSeek V4 flash > Give it to your agent to get super fast commit messages → https
  • DeepSeek tool — > ◆ 95 file PR went from taking 71s -> 13s, over 5x faster > ◆ Full support for local models (@lmstudio & @ollama) > ◆ Added support for DeepSeek V4 flash > Give it to your agent to get super fast commit messages → https
Agentic codingOpen sourceHas links
Post ↗ · @nutlope

erence infrastructure for agents (link to engineering blog).

  • High engagement.[[8]](https://x.com/jack/status/2084667826414002362) ### @clairevo **Post 1** > openclaw - personal agents, proactive agents, agents i wanna hang with > hermes - vibes were off so i abandoned i'm a creature of habit and glutton for punishment 🦞 > eve - internal agents, enterprise search, github-y thi…
  • r personal/proactive, Eve for internal/enterprise/GitHub-style).
  • Insight into the fragmented agent tooling landscape and use-case fit.[[9]](https://x.com/clairevo/status/2084449388425097494) **Post 2** > prompt engineering → RAG → context engineering → harness engineering → loops → goals→ graph engineering → token astrology → entropy gardening → AGI necromancy → transformer exorc…
Agentic coding
Primary post ↗ · Post 2 ↗ · @clairevo · 2 posts combined

spinout.)** **Why technically notable:** Highlights the spin-out/continuation of Hyperagent (Airtable’s AI/agent-focused…

  • Relevant to AI product strategy, low-code + agents, and survival/evolution of agent tools post-M&A.
  • Follow-ups confirm details.[[7]](https://x.com/swyx/status/2084695630237311086) ### @jack **Post** > mesh inference with buzz **URL:**
Tools, repos & products
  • Buzz tool — Follow-ups confirm details.[[7]](https://x.com/swyx/status/2084695630237311086) ### @jack **Post** > mesh inference with buzz **URL:**
Agentic codingHas links
Post ↗ · @jack

g) in the Gemini API for the Flash models.

  • This directly improves agentic capabilities for location-aware + web-search workflows, reducing previous limitations on parallel tool calling and enabling more robust software agents/apps.[[1]](https://x.com/OfficialLoganK/status/2084469065322729817) **Post 2** (reply in context of Google AI Studio managed agents de…
  • Limited extractable detail in the ingest text — open the source if this topic is critical.
Agentic coding
Post ↗ · @OfficialLoganK

tantive posts matching AI agents, coding agents, tools/repos, model/training infra, or product launches/updates.

  • Engagement bait, pure replies, memes, or off-topic comments were deprioritized.
  • Other accounts (@ClementDelangue, @AIatMeta, @ylecun, @karpathy, @LangChainAI, @mntruell) had no highly matching primary posts with substance in the window (mostly short replies, opinions, or nothing).[[1]](
  • Notable posts from the last ~24 hours** (based on available data around Aug 4, 2026 GMT;
  • focused on primary/substantive posts about AI agents, tools, automation, or related launches).
  • No relevant posts found from @tomasonjo (account/posts not surfacing in searches).[[1]](
Agentic codingToolsProduct discovery
Primary post ↗ · Post 2 ↗ · x

Why technically notable:** Endorsement of Luna’s (OpenAI model) post-price-cut efficiency for practical workloads includ…

  • Highlights model economics enabling heavier tool/repo usage.[[5]](https://x.com/gdb/status/2084770040281211043) **Notes:** - @sama’s main recent post is a general optimism statement (not specifically about agents/tools/models/products).
  • Limited extractable detail in the ingest text — open the source if this topic is critical.
Agentic codingTools
Post ↗ · @gdb

tantive posts matching AI agents, coding agents, tools/repos, model/training infra, or product launches/updates.

  • Engagement bait, pure replies, memes, or off-topic comments were deprioritized.
  • Other accounts (@ClementDelangue, @AIatMeta, @ylecun, @karpathy, @LangChainAI, @mntruell) had no highly matching primary posts with substance in the window (mostly short replies, opinions, or nothing).[[1]](
Agentic codingToolsProduct discovery
Post ↗ · @hwchase17

YouTube

6

Watchlist channels (except GitHub Awesome)

The Framework Putting You Ahead of 99% of AI Users

  • And just to make it simple, whenever I talk about context in our community, I always use this ARMS framework with the core idea being that if you organize and engineer this ARMS framework properly as your context, then you're actually ahead of like 99% of cloud and agentic AI users because the entirety of your conte…
  • And finally, you have your skills, which are SOPs or processes that you can actually invoke through skill commands that immediately just teach Cloud how to do a given instead of work.
  • Now, another important shift that the article talked about is what we're calling as progressive disclosure, which you can see Tariq differentiates versus the practice before where you are putting all of the context upfront.
  • And just to elaborate on that, he mentions here that because Cloud Code in the very beginning was focused on coding, their system prompt included or needed to include detailed information on how to do code review, verification, which are all these details that are not always needed, but when they are, it was actuall…
  • So it will check your install of cloud code to see if there's like any broken or duplicate installs, any path, file path problems.
  • And so the point being here is that as your agentic operating system evolves and you start to have a workspace for your company or for your team that becomes a bit more complex, what you can start to do is to actually I identify the different departments of your life and of your work and actually build out these sub…
Tools, repos & products
  • Claude Code tool — Mentioned tool/product: Claude Code.
  • MCP tool — Mentioned tool/product: MCP.
Agentic codingToolsMemoryHas links
Primary post ↗ · Post 2 ↗ · Post 3 ↗ · Post 4 ↗ · @RoboNuggets

OpenAI Astra Just Advanced Mathematics... 10 Times.

  • Stacked pull requests are now built into GitHub, and they change how you ship anything big.
  • Instead of one giant PR nobody wants to review, you get a chain of smaller dependent branches that merge independently.
  • Worth knowing if you're pushing large changes, or letting an agent code for hours.
  • Radically better observability stack: https://betterstack.com/
  • Astra solved ten open problems in maths and theoretical computer science, and the tokens cost roughly $2,000.
Tools, repos & products
  • Github product — Stacked pull requests are now built into GitHub, and they change how you ship anything big.
  • Betterstack product — Radically better observability stack: https://betterstack.com/
  • Betterstack product — Radically better observability stack: https://betterstack.com/
  • Instagram product — Instagram: https://www.instagram.com/betterstackhq/
  • Tiktok product — TikTok: https://www.tiktok.com/@betterstack
  • Linkedin product — LinkedIn: https://www.linkedin.com/company/betterstack
  • Openai product — OpenAI Astra Just Advanced Mathematics...
Agentic codingToolsOpen sourceHas links
Primary post ↗ · Post 2 ↗ · @betterstack

New Skills! v1.2 brings /wait-what, /writing-for-agents, and fixes /grill-me

  • Skills v1.2.0 is out with major improvements: new documentation site, Claude Code marketplace integration, five brand new skills including Wait What for Opus verbosity, an updated Grill Me with multi-question rounds, and the powerful Wizard skill for infrastructure provisioning.
  • :18 Claude Code marketplace integration
  • :43 Codex compatibility improvements
  • :27 Writing for agents skill
  • :58 To Questionnaire for collaboration
Tools, repos & products
  • Aihero product — Documentation: https://aihero.dev/skills
  • Aihero product — Documentation: https://aihero.dev/skills
  • Aihero product — Documentation: https://aihero.dev/skills
  • Claude Code tool — Skills v1.2.0 is out with major improvements: new documentation site, Claude Code marketplace integration, five brand new skills including Wait What for Opus verbosity, an updated Grill Me with multi-question rounds, and
  • Codex tool — :43 Codex compatibility improvements
Agentic codingHas links
Video ↗ · @mattpocockuk

How TurboFieldFare Runs a 26B LLM on a MacBook Air (Run AI Locally)

  • TurboFieldfare runs a 26B mixture-of-experts local LLM on an 8GB MacBook Air using ~2GB RAM by streaming experts from SSD
  • Built by independent developer Andrey Mikhaylov, it keeps a 1.35GB shared model core plus a KV cache resident in memory and streams the other 120 idle experts off the SSD per token, hitting 5.1-6.3 tokens per second on that base machine and 24-35 tok/s on a 24GB M5 Pro.
  • This video breaks down the actual architecture behind the trick.
  • You'll see how the 8-bit router picks 8 of 128 experts plus one shared expert per token, why a 16-slot least-frequently-used cache keeps SSD reads survivable, and why the model's 256K context window quietly eats into the same tiny RAM budget.
  • It also covers the real tradeoffs: an independent LM Studio coding benchmark scored this exact model 33.33% with all 21 test cases failed
Tools, repos & products
  • LM Studio tool — It also covers the real tradeoffs: an independent LM Studio coding benchmark scored this exact model 33.33% with all 21 test cases failed
MemoryHas links
Video ↗ · @The-Stack-ai

Voice Agent observability with LangSmith 🌟

  • Caroline di Vittorio, Engineer at LangChain, builds a voice agent with the Google ADK and the Gemini Live model, then sets up tracing in LangSmith to see exactly what the agent is doing under the hood.
  • Gemini Live is Google's native audio model.
  • It's speech-to-speech, taking audio directly as input and producing audio as output without transcribing to text, which keeps latency low and makes the agent's voice sound natural and emotive.
  • What's covered: building a terminal-based weather assistant with two tools, defining the LangSmith Google ADK plugin, registering it on the runner, recording user and agent audio so the trace captures only what the user actually heard, and reading the full trace in LangSmith, including audio playback, transcripts, t…
  • Get started with the Gemini Live API.
  • Grab an API key in Google AI Studio and start building.
Tools, repos & products
  • Goo product — Caroline di Vittorio, Engineer at LangChain, builds a voice agent with the Google ADK and the Gemini Live model, then sets up tracing in LangSmith to see exactly what the agent is doing under the hood.
Agentic codingHas links
Video ↗ · @GoogleDevelopers

ChatGPT Just Got Its Biggest Upgrade Yet

  • A new Chinese AI video model just made 2K, studio-quality clips with built-in sound — for a third of the cost of Western tools.
  • See what MiniMax H3 can actually do, how it stacks up against the old silent 1080p standard, and how one prompt can take you from script to finished video.
  • :01 What It Is – Omnimodal model reads text, image, video, and audio
  • :22 Rankings – #1 for AI editing, top 3 for text/image-to-video
  • Get 200+ Free AI SEO Prompts → https://go.juliangoldie.com/chat-gpt-prompts
  • ChatGPT can now read every tab you have open, explain anything you highlight, and break down YouTube videos — with zero copy-paste.
Tools, repos & products
  • Go product — Get 200+ Free AI SEO Prompts → https://go.juliangoldie.com/chat-gpt-prompts
Agentic codingToolsHas links
Primary post ↗ · Post 2 ↗ · @JulianGoldieSEO

GitHub Awesome

2

Repos discussed — what each does + link

GitHub Trending Weekly #43: Cloudflare Computer, waku-agent, querysplat, findphone, Soup, morphicons

  • Primary repo: cloudflare/computer — Covered in GitHub Awesome: Computer.
  • Open the repo (not only the short) to judge stars, license, and maintenance.
Tools, repos & products
Agentic codingToolsMulti-agentOpen sourceHas links
Video ↗ · @GithubAwesome

anydoc converts Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV and PDF into Markdown

  • anydoc converts Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV and PDF into Markdown, written in Rust with Node and Python bindings. Every format parses into one shared document model and renders through one serializer, so a table-escaping fix for docx fixes rtf and odt at the same time. Form…
  • Primary repo: firecrawl/anydoc — Covered in GitHub Awesome: opensource.
  • Open the repo (not only the short) to judge stars, license, and maintenance.
Tools, repos & products
  • firecrawl/anydoc repo — anydoc converts Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV and PDF into Markdown, written in Rust with Node and Python bindings. Every format parses into one shared document model and renders through one seria
ToolsOpen sourceHas links
Video ↗ · @GithubAwesome

Product Hunt

8

Launches

npm i -g hotcell — Local sandboxes for AI agents on your Mac, Linux, bare metal

  • Mentions: Producthunt.
  • Limited extractable detail in the ingest text — open the source if this topic is critical.
Tools, repos & products
  • Producthunt product — Linked product/site (producthunt.com).
Agentic codingSecurityHas links
Product Hunt ↗ · Product Hunt

Keystroke — Build powerful AI agents & workflows

  • Mentions: Producthunt.
  • Limited extractable detail in the ingest text — open the source if this topic is critical.
Tools, repos & products
  • Producthunt product — Linked product/site (producthunt.com).
Agentic codingHas links
Product Hunt ↗ · Product Hunt

StepGrab — Turn any Mac task into a step-by-step guide

  • Mentions: Producthunt.
  • Limited extractable detail in the ingest text — open the source if this topic is critical.
Tools, repos & products
  • Producthunt product — Linked product/site (producthunt.com).
Has links
Product Hunt ↗ · Product Hunt

Dover MCP — Run your hiring process from Claude or ChatGPT

  • Mentions: Producthunt, MCP.
  • Limited extractable detail in the ingest text — open the source if this topic is critical.
Tools, repos & products
  • Producthunt product — Linked product/site (producthunt.com).
  • MCP tool — Mentioned tool/product: MCP.
Has links
Product Hunt ↗ · Product Hunt

Kiro Crew — Open source agentic development workspace

  • Mentions: Producthunt.
  • Limited extractable detail in the ingest text — open the source if this topic is critical.
Tools, repos & products
  • Producthunt product — Linked product/site (producthunt.com).
Agentic codingOpen sourceHas links
Product Hunt ↗ · Product Hunt

Cloudflare Wallets — the programmable wallet for the agentic Internet

  • Mentions: Producthunt.
  • Limited extractable detail in the ingest text — open the source if this topic is critical.
Tools, repos & products
  • Producthunt product — Linked product/site (producthunt.com).
Agentic codingHas links
Product Hunt ↗ · Product Hunt

MOTHER — A terminal built for Claude Code w/ one-click session resume

  • Mentions: Producthunt, Claude Code.
  • Limited extractable detail in the ingest text — open the source if this topic is critical.
Tools, repos & products
  • Producthunt product — Linked product/site (producthunt.com).
  • Claude Code tool — Mentioned tool/product: Claude Code.
Agentic codingToolsHas links
Product Hunt ↗ · Product Hunt

Aegisora — The narrow control plane for AI agent tool and API calls.

  • Mentions: Producthunt.
  • Limited extractable detail in the ingest text — open the source if this topic is critical.
Tools, repos & products
  • Producthunt product — Linked product/site (producthunt.com).
Agentic codingHas links
Product Hunt ↗ · Product Hunt

Hacker News

8

Show HN + technical threads

Launch HN: Hoplite (YC S26) – Effortlessly deploy cloud coding agents

  • During onboarding, we port over your local setup - sessions, memories, MCP servers, and get your projects ready to run in the cloud. Here’s a demo: https://youtu.be/bnyktZ_9pjE We got here after pivoting away from the idea we applied to YC with;
  • In reflecting on what we really wanted to do, we realised that we loved talking to founders and developers, and were really opinionated about the specific area of cloud agents.
  • We tried out all the existing solutions, and didn’t find one that A) took good advantage of being in the cloud, and B) was performant and felt good to use. We’re building a product that we feel reflects what mainstream development will look like in 6-12 months.
  • That means evaluating new user flows, visually verifying that new features look good, that the API works as expected, that the CLI works on Windows, etc.
  • We spent a lot of time deciding on whether we should use an off the shelf solution like Codex/Claude Code, but ultimately wanted the independence and freedom that came with building it in house.
  • It also means that we can test out completely new features without relying on Anthropic and OpenAI to catch up. Everything is hosted on AWS, with the exception of: Temporal for durable workflows, Modal for sandboxes, and Planetscale for our database.
Tools, repos & products
  • News product — HN discussion: https://news.ycombinator.com/item?id=49157997
  • MCP tool — During onboarding, we port over your local setup - sessions, memories, MCP servers, and get your projects ready to run in the cloud. Here’s a demo: https://youtu.be/bnyktZ_9pjE We got here after pivoting a
  • Codex tool — We spent a lot of time deciding on whether we should use an off the shelf solution like Codex/Claude Code, but ultimately wanted the independence and freedom that came with building it in house.
  • Claude Code tool — We spent a lot of time deciding on whether we should use an off the shelf solution like Codex/Claude Code, but ultimately wanted the independence and freedom that came with building it in house.
Agentic codingToolsProduct discoverySecurityHas links
Thread ↗ · HN · BenceRed

Show HN: A faster coding agent than Codex and Claude Code

  • Hi HN, excited to be sharing this with you guys today. TL;DR: Bullet is a fast coding agent. At our company, we were burning hours waiting on agent runs.
  • The agent loops around them were slow (unnecessary planning, oversized context, repeated searches, and serial tool calls). So we built Bullet as a passion project, the goal being to help you iterate faster, without giving up quality:
  • It uses targeted code search instead of embedding the entire repository.
  • It works with your Claude Code or Codex subscription, OpenAI/Anthropic/xAI API keys, or an on-device model with no key. Claude Code and Codex each center on their own model ecosystem.
  • OpenCode gives you many models, but largely leaves model selection and configuration to you.
  • Connect what you already pay for, then let Bullet choose the fastest path for every task. On SWE-bench Verified, Bullet resolved 479/500 issues (95.8%, top 3 on the leaderboard) in one attempt, averaging 119 seconds per task (35–67% faster than mini-SWE-agent + Fable/Sol).
Tools, repos & products
  • News product — HN discussion: https://news.ycombinator.com/item?id=49173799
  • Codex tool — It works with your Claude Code or Codex subscription, OpenAI/Anthropic/xAI API keys, or an on-device model with no key. Claude Code and Codex each center on their own model ecosystem.
  • Claude Code tool — It works with your Claude Code or Codex subscription, OpenAI/Anthropic/xAI API keys, or an on-device model with no key. Claude Code and Codex each center on their own model ecosystem.
Agentic codingToolsHas links
Thread ↗ · HN · alsima

Show HN: Product analytics (and evals) for agent sessions on your MCP

  • We reconstruct the entire session behind the MCP tool calls you receive, including what the user asked their agent to do and what the agent thought. You wrap your MCP in 3 lines of code (our SDK is available in Typescript, Python and Go) and start seeing in your dashboard:
  • The most frequent issues your users’ agents encounter so you can fix them. Here is a quick demo: https://youtu.be/ZFlvquhyNMQ The story behind this is that we initially launched Armature as a standalone testing tool ( https://www.ycombinator.com/launches/QQc-armature-making-you... …
  • It’s something we had also experienced in our previous companies: Louis built MCPs exposed to millions of users and Theo was a Forward Deployed Engineer at Palantir before joining a Datadog spin-off as Founding Engineer.
  • Both testing and product analytics had always been real pains when exposing a product to agents but we always thought there wasn’t much we could do about analytics because the conversation lived in our users’ AI client. Then it struck us: what if we asked the agents why they were making this or that tool call?
  • When our CTO friends heard about this, they wanted to try it for themselves so we gave them access to a cloned version of our internal product and they started sharing feedback like they never did on our “real” product! That’s when we decided to start working seriously on MCP Analytics as a product.
  • stateless MCPs isn’t perfect, and use-case clustering remains to be optimized. But we are finally launching our analytics product to everyone, self-serve at https://armature.tech with a set-up that takes less than 5 minutes and a generous free tier. And now we are working on fully closing the loop, bringin…
Tools, repos & products
  • News product — HN discussion: https://news.ycombinator.com/item?id=49157807
  • MCP tool — We reconstruct the entire session behind the MCP tool calls you receive, including what the user asked their agent to do and what the agent thought. You wrap your MCP in 3 lines of code (our SDK is available in Typescrip
  • Claude Code tool — Mentioned tool/product: Claude Code.
Agentic codingToolsProduct discoveryHas links
Thread ↗ · HN · screm

Show HN: Guide AI coding agents on how to use libraries securely

  • Hi HN, TL;DR: AI Code Security Cards give coding agents library- and version-specific guidance to generate safer code. AI coding agents can generate working code, but they often lack library- and version-specific security knowledge.
  • We've developed AI Code Security Cards, an open-source security knowledge layer, to provide that missing context. The project is related to my PhD, in which I focused on the security of LLM-generated code.
  • During my research, I studied how code generation models behave, how they introduce security issues, and how we can guide them toward generating safer code. My research [1] and several other studies [2–4] show that at least 40% of code generated by state-of-the-art models contains one or more security issues.
  • Similar problems also affect AI coding agents. One reason is that these models often lack sufficient library-specific security knowledge in their context.
  • They may not know which configurations are unsafe, which validation steps are required, or what security-related behavior has changed between library versions. AI Code Security Cards provide practical, version-specific security guidance for open-source libraries.
  • For example, the Django security cards can guide an AI agent on how to implement authentication and authorization securely. The easiest way to use the cards is to install the AI skill provided in the repository (And in the integration page: https://securitycards.rewarelabs.com/integration/ ).
Tools, repos & products
  • News product — HN discussion: https://news.ycombinator.com/item?id=49169280
Agentic codingToolsOpen sourceHas links
Thread ↗ · HN · hajipour

Show HN: Capshelf – Share agent skills across repos with per-project lockfiles

  • I was constantly working on multiple projects in parallel, which made me keep the same Claude Code skills in several repos, and they slowly drifted apart.
  • A security-review skill I kept improving in one project did not change everywhere else, and after a while I couldn't tell which copy was current. I tried symlinking into a shared directory, and it fixed the duplication but created a worse problem: editing a skill for one repo silently changed the agent's behavior in…
  • It currently supports Claude Code, Codex and Pi and allows you to pin skills, Pi extensions, settings values, and MCP server entries. The main idea was to vendor skills into a project in a controlled way by pinning them using a content hash.
  • Capshelf has a manifest that lists installed items and a lockfile that saves hashes and source commits of all installed items. To start building your skills repository, init capshelf in one of your repositories and point it to an empty git repo:
  • $ capshelf share design-review --to local This will start tracking the skill using capshelf and pin the current version in this repository. Then in another project, initialize capshelf and add the skills you need:
  • $ capshelf add design-review If you, or your agent, edit the design-review skill, you can easily promote it so that other projects can take the updates.
Tools, repos & products
  • News product — HN discussion: https://news.ycombinator.com/item?id=49170377
  • Claude Code tool — I was constantly working on multiple projects in parallel, which made me keep the same Claude Code skills in several repos, and they slowly drifted apart.
  • MCP tool — It currently supports Claude Code, Codex and Pi and allows you to pin skills, Pi extensions, settings values, and MCP server entries. The main idea was to vendor skills into a project in a controlled way by pinning them
  • Codex tool — It currently supports Claude Code, Codex and Pi and allows you to pin skills, Pi extensions, settings values, and MCP server entries. The main idea was to vendor skills into a project in a controlled way by pinning them
Agentic codingToolsOpen sourceHas links
Thread ↗ · HN · mstr32

Show HN: MoneyCo – A coding agent built for accounting

  • MoneyCo connects to multiple Quickbooks and Xero accounts by default and lets you do your bookkeeping using a coding agent. I got here after talking to multiple accountants and hearing them struggle to use Claude, Codex, Hermes and OpenClaw.
  • Configuring client memory Coding agents disguised as one that does accounting felt like a simple yet powerful idea as a solution: 1.
  • We connect to QBO and Xero using API instead of MCP and make it easy for multiple accounts connection
  • We deploy all client files along with `pi --rpc` in a sandbox so all can work on the cloud
  • A coding agent can simply write code for deterministic computations and run tests to catch inaccuracies
  • Each chat made progressively gets stored in the persistent sandbox meaning that agent can search through older conversations to look into its memory I had to rebuild some of the architecture here several times as it required building a durable, persistent cloud coding agent essentially.
Tools, repos & products
  • News product — HN discussion: https://news.ycombinator.com/item?id=49160189
  • Codex tool — MoneyCo connects to multiple Quickbooks and Xero accounts by default and lets you do your bookkeeping using a coding agent. I got here after talking to multiple accountants and hearing them struggle to use Claude, Codex,
  • MCP tool — We connect to QBO and Xero using API instead of MCP and make it easy for multiple accounts connection
Agentic codingToolsProduct discoverySecurityMemory
Thread ↗ · HN · warthog

Show HN: Isolade, a local-first coding agent workbench with secretless microVMs

  • I built Isolade because I could not find an open-source, local-first workbench that runs coding agents in microVMs. The two pieces already exist individually.
  • There are plenty of options for both agent management (Conductor, Herdr, Omnigent) and microVM isolation (Docker Sandboxes, Firecracker, SmolVM).
  • However, I did not find a project that combined them into one unified product. Isolade gives each agent its own microVM.
  • Microsandbox provides subsecond provisioning and domain-scoped secret substitution, so that the VM only sees placeholder values.
  • Instead of a worktree, each agent gets a copy-on-write clone of the entire setup.
  • This works well for setups involving multiple repositories, and cached dependencies mean agents can start working immediately without any per-worktree setup. A single UI lets you work with multiple agents concurrently.
Tools, repos & products
  • News product — HN discussion: https://news.ycombinator.com/item?id=49168002
  • Claude Code tool — Mentioned tool/product: Claude Code.
  • Codex tool — Mentioned tool/product: Codex.
Agentic codingToolsSecurityHas links
Thread ↗ · HN · jachris

Show HN: Memcode launches a new terminal coding agent

  • Memcode is a new developer platform currently in public beta.
  • It includes a coding agent, chat, reusable agents, DataHub, and a Lovable-style website generator. The coding agent is heavily inspired by Claude Code and works across all frontier models as well as open-weight models such as Kimi and GLM. Memcode was designed to stay in sync with you and behave more like a human co…
  • It grew out of constant frustration with existing coding agents. If your current coding agent already works great for you, Memcode may not be especially compelling.
  • But if you regularly find yourself repeating context, correcting lost assumptions, or fighting to keep an agent on track, or you simply enjoy trying new developer tools, give Memcode a try. https://memcode.ai
Tools, repos & products
  • News product — HN discussion: https://news.ycombinator.com/item?id=49177986
  • Claude Code tool — It includes a coding agent, chat, reusable agents, DataHub, and a Lovable-style website generator. The coding agent is heavily inspired by Claude Code and works across all frontier models as well as open-weight models su
Agentic codingProduct discoveryMemoryHas links
Thread ↗ · HN · timerwin

Also noted

24

Scanned but not full stories — one line + link