Fetching from the wire…
Public story · 2026-07-30 · high
The Apache-licensed tool ships 52 rules across 11 threat categories but leaves every one in monitor-only mode by default.
Why now: Numbat tagged v0.1.1 on July 29, the day before this story ran.
Perplexity released numbat, an open-source detection layer for coding agents, on July 29, per the project's GitHub repo.
Anyone running Claude Code, Codex, OpenCode, or Pi unattended is exposed if something goes wrong. Numbat lets them reconstruct what the agent did after the fact, straight from artifacts already on disk.
None of numbat's rules block anything. Enforcement is off by default, per the repo, even though the rule set spans privilege escalation, secret exfiltration, and lateral movement checks.
Numbat watches through local hooks and OTLP/HTTP log exporters, normalizing everything into one event model that a CEL rule engine evaluates.
When a rule trips, it produces a case bundle with a SHA-256 manifest, evidence meant to hold up rather than a log line.
It's a single Go binary, no cgo, running on macOS, Linux, and Windows. Apache 2.0 licensed, it had 239 stars at v0.1.1.
Perplexity built the forensics half of an agent security product and shipped it while skipping enforcement entirely. That's a sign even the team behind the 52 rules doesn't trust automated blocking on a coding agent yet. Whether a future release adds a block mode is the thing to watch.
Each link below shares sources, entities, or timing with this story.
The agent-security topic holds 42 repos above 100 stars, four from large companies rather than startups: NVIDIA/SkillSpector (14,498 stars, scanning Claude Code/Codex/MCP skills for prompt injection), Tencent/AI-Infra-Guard (4,467, red-teaming with Many-Shot/PAIR/GOAT/ActorAtt...
open-design is a local-first macOS/Windows/Linux desktop app that turns your existing coding agent into a design engine, producing prototypes, landing pages, dashboards, slides, images, and video with real HTML/PDF/PPTX/MP4 export. It runs on 25 distinct local CLI executables...
+7,546 stars this week for an "Application Development Environment" running each concurrent agent in its own worktree. Works with any terminal CLI agent — Claude Code, Codex, OpenCode, Pi, Cursor, Copilot, Grok, 30+ others — across macOS, Windows, Linux, iOS via App Store/Test...
I check Product Hunt maybe once a week and usually regret it. Today's board is worth reading as market structure. The July 30 leaderboard: SKI at 277 upvotes (free voice input for Claude Code and Codex). AI Search Console at 249 (prompt analytics and citation mapping). Memmy A...
Google released Antigravity IDE Extensions, putting its agentic coding platform into VS Code (macOS, Linux, Windows), JetBrains IDEs from 2026.2.1 (IntelliJ, PyCharm, WebStorm, GoLand, CLion, Rider), Zed, and Visual Studio 2026 in preview. One Antigravity account works everywh...
One HTTP/SSE API that controls Claude Code, Codex, OpenCode, and Amp inside sandboxes. Solves the fragmentation problem: write one integration, swap agents with a config change. Session persistence to Postgres/ClickHouse. Apache 2.0. (GitHub)
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.