What if your AI agent could actually check out at Walmart?
AgentMall gives agents one API and MCP server to search, fill a cart and check out. Its site lists Walmart, Amazon and Target. The team says checkout works about 70% of the time so far.
getagentmall.com
6
What if your AI agent could build your slide deck for you?
Decks is an open-source Slidev framework with agent guidance built in, plus charts and a CLI to scaffold, build and export decks. Apache 2.0. Early project.
github.com/steamfield/decks
2
10
Former Claude Code engineer Thariq Shihipar shares a skill he says people at Anthropic used a lot: /eli5. Ask it to explain anything, and it answers with an HTML page of big pictures and few words.
Video: @trq212
a skill people at Anthropic have been using a lot recently: ELI5
/eli5 <what you want explained>
"explain like I'm someone who knows nothing about this topic, using a HTML artifact with big pictures and few words"
1
20
What if running five coding agents didn't mean five terminals?
Mergepond is a Mac Git client that gives each agent its own worktree and shows them side by side. Free for local repos. The agent board is part of Pro.
mergepond.dev
7
Claude Code creator Boris Cherny says he makes his slide decks in Claude Slides now. No more googling "iphone frame svg."
Watch his short demo.
Video: @bcherny
Don't sleep on Claude Slides. This is how I make slide decks now -- no more googling "iphone frame svg" like the old days
Try it: claude.ai/artifacts
14
A Google DeepMind engineer shows how to get a coding agent to test a voice feature on its own.
Philipp Schmid's steps: generate speech with the Gemini API CLI, play it into Chrome as a fake mic, let the agent click the mic, check the transcript.
Video: @_philschmid
Did you know you can use the Gemini API CLI together with your coding agent to test multimodal features like voice and audio? Here is how it works:
1. Use Gemini API CLI to generate speak
2. Use Chrome with --use-file-for-fake-audio-capture=voice.wav%noloop
3. Click the mic button via Playwright or browser-use so getUserMedia streams the WAV from 0.0s
4. Verify the transcript and UI state in the DOM
Prompt below 👇
19
What if your AI could flag citations it made up?
Citeparse reads a reference list on your machine, then looks each one up on Crossref. Anything it can't match gets flagged to check by hand. A miss is not proof of a fake. Free, open source (Apache 2.0).
github.com/un-seen/citeparse
1
19
Claude Code creator Boris Cherny says he doesn't know Lean well. With Opus 5.5 and a few prompts, he got 16 PRs tackling bugs and race conditions in the Claude Agent SDK.
Watch what he asks it to do.
Video: @bcherny
I used Opus 5.5 to formally verify the Claude Agent SDK using Lean. A couple short prompts = 16 PRs fixing various bugs and race conditions. Video attached.
TLA+ also works well. I sometimes combine Lean and TLA+ to look for issues around data flow, concurrency, and state mgmt.
I don't know either language well, but Claude is excellent at both. This approach is super useful for formally modeling your code and finding bugs that a human probably wouldn't have spotted.
Is formal verification the future of coding (or at least, bug finding)?
33
What if the AI tools you build were made to be thrown away?
Disposable is a site for sharing small AI-made tools, skills and prompts, so others can rebuild them their way with their own agent. Open source. Early project.
github.com/x-ify/disposable
1
40
What if your issue tracker could ship its own pull requests?
Mastra Factory is an open-source setup that turns issues into plans, code and reviewed PRs, following the approvals you set. Runs on Mastra platform by default, or self-host.
factory.mastra.ai
9
What if any website could become a tool your AI agent can call?
Machine Internet watches the JSON requests a site makes and turns them into MCP tools. Open source, Apache 2.0. The author says it worked on about 1 in 3 of 51 sites tested.
github.com/chetanty/machine_…
12
Your old prompts may be making Claude worse.
Anthropic's Lance Martin reports tuning model, effort and prompt cut cost 90% on a ~50-ticket support benchmark while improving the score. Talk at AI Engineer.
youtube.com/watch?v=ys6YQ2yf…
13
What if your AI agent saved the steps that worked as a reusable tool?
Coral is a self-hosted, open-source gateway (Apache 2.0). Agents write TypeScript programs, save good ones as custom tools, and each call is checked against your policy first.
github.com/withcoral/coral
18
What if your AI agent kept a history you could branch like git?
Pizza is an open-source desktop and terminal agent. Every action is an event in a log, so you can fork a conversation from any earlier point.
github.com/tomsun28/pizza
17
What if a whole coding agent was one 9 MB file?
uya-agent is a static Linux binary: no Node, no Python. It has a terminal UI, a browser mode and a permission sandbox. Free, MIT license. Early project.
github.com/uya-lang/uya-agen…
19
Tired of clicking through your app to reach the state you want to test?
Google Cloud Tech shows how reusable agent skills can build a hidden edit mode, with a personal trainer app as the demo.
youtube.com/watch?v=1fJVOyLm…
23
How do you let a coding agent ship code without handing it your Git credentials?
GitSeed (Apache 2.0, self-hosted) runs agents in isolated containers and a separate step does the push. Works with Claude Code, Codex, Gemini. New project.
github.com/gitseed-dev/gitse…
4
3
80
Your AI agent gets slower and forgets more as the chat grows.
Google Cloud Tech's agent clinic shows a fix: trim old context as you go and add long-term memory with Vertex AI Memory Bank.
youtube.com/watch?v=RvJ6UpR2…
1
27
What if one agent could work in Slack, Teams, Discord and iMessage?
Vercel's eve 1.0 is out: an open-source framework for production agents. You pick the model and decide which actions need approval.
vercel.com/blog/eve-is-now-g…
19
What if every coding-agent task needed plan, build, review, test and human sign-off? Tencent's LoopForge runs Claude Code, Codex and other agent CLIs in Docker sandboxes with approval gates.
Self-hosted, MIT license, per its README.
github.com/Tencent/LoopForge
25



