Product Hunt
Arcjet took Launch of the Day on Product Hunt this week. It's a runtime security platform built specifically for AI agents: detect prompt injection, authorize what an agent's tool calls are allowed to do, redact sensitive data before it leaves your app, and block bots and abuse — all as real-time building blocks you call inline, before the action happens rather than after the damage. It's a sign of where the market's attention is moving: not "build an agent" but "keep the agent you already built from doing something you didn't authorize."
Jackalope also picked up traction in Product Hunt's AI coding agents category. It's a desktop workspace that runs multiple coding agents in parallel, each in its own isolated git worktree, so you can fan out a task across several agents, compare their diffs, and review the results in one place instead of juggling terminal tabs. It plugs into whichever agent CLI you already use rather than shipping its own.
Elsewhere
OpenAI shipped GPT-6 Sol and GPT-6 Luna on September 22nd, filling out the GPT-6 family alongside the earlier Astra release. Both are pitched as cheaper, more reliable options built with the same methods that improved Astra on professional work, factuality, coding, and computer use. Sol lands at $2 per million input tokens and $10 per million output; Luna is priced for volume at $0.10 in and $0.50 out — down from $0.20/$1.20 on the previous Luna. Sol and Luna are rolling into ChatGPT Work and Codex for most paid plans and the API immediately, with Luna also reaching the desktop app and Free/Go users. GitHub Copilot picked up both as selectable models the same day.
On GitHub, cloudflare/security-audit-skill has been sitting at the top of GitHub Trending since mid-September and has pulled in roughly 15,000 stars in the last week alone, north of 19,000 total. It's an open-source coding-agent skill that turns any capable agent into a security auditor — orchestrating isolated sub-agents through reconnaissance, coverage-led vulnerability hunting, candidate validation, and independently verified, machine-readable findings. Worth a look if you're already running agent-based code review and want the audit step to produce something more trustworthy than a wall of prose.
Worth your time
Arcjet and security-audit-skill are two sides of the same coin this week: agents are now trusted enough to run unsupervised, which means the tooling market has pivoted hard toward proving and constraining what they're allowed to do. If you're shipping anything agentic, that's the category to watch before the next one to build in.
On the model side, GPT-6 Luna's price cut is the more interesting number than Sol's release — a 4x drop in cost for a model this capable is the kind of move that resets what "good enough to ship at scale" costs.