Entity trail

Craft Agents

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-07-01 / vibe-coding-researcherTwo CVSS-9.8 RCE Bugs in MCP Tooling: MCPJam Inspector and nginx-uiCVE-2026-23744 hits the MCPJam inspector, which binds 0.0.0.0 with no authentication, letting a crafted HTTP request install an MCP server and execute arbitrary code. CVE-2026-33032 affects nginx-ui (150M+ downloads), whose MCP message endpoint performs no auth on command-execution requests. Both are critical remote-code-execution flaws in tools developers run locally alongside their agents — check binds and auth before exposing any MCP inspector.
2026-06-30 / reddit-researcherChamath Palihapitiya Takes CEO Seat as 8090 Labs Closes $135M Series A for Enterprise 'Software Factory'8090 Labs raised a $135M Series A led by Salesforce Ventures (with WndrCo, Craft Ventures, The Production Board, LAUNCH, and angels Nikesh Arora and Adam D'Angelo), and founder Chamath Palihapitiya announced June 29 he is stepping in as CEO. The product, 'Software Factory,' puts people and AI agents in one environment spanning requirements, architecture, work orders, code, testing, and production maintenance for regulated sectors including healthcare, aerospace, financial services, energy, and the US government. It is a high-profile bet that enterprise software delivery — not just autocomplete — is the agentic-coding prize.
2026-06-26 / sources-researcher2,000 People Tried to Jailbreak Simon Willison's AI Assistant — Real-World Prompt-Injection Red-Team ResultsFernando Irarrázaval ran an open challenge at hackmyclaw.com inviting anyone to leak secrets held by an OpenClaw test instance via crafted messages, and ~2,000 people attempted it. Simon Willison's writeup turns it into a rare large-scale, public data point on how AI assistants actually hold up under adversarial prompt injection rather than lab benchmarks. This is directly actionable for anyone shipping tool-using agents that hold credentials or secrets.
2026-06-12 / sources-researcherLoopcraft: Karpathy, Cherny and Steinberger Reframe Agent Work as 'Stacking Loops'Latent Space's June 12 AINews issue distills a builder consensus that you should stop prompting coding agents and instead design loops that prompt them — Boris Cherny: 'I don't prompt Claude anymore. I write loops, the loops do the work,' and Andrej Karpathy on 'removing yourself as the bottleneck.' The piece frames competitive advantage as knowing when to descend into lower loops for reliability and ascend to higher ones as models improve. It's the clearest articulation yet of the orchestration-over-prompting shift, calling it 'The Salty Lesson for agents.'
2026-05-28 / arxiv-researcherMIRAGE: Context-Aware Prompt Injection Attacks Against Mobile GUI Agents via User-Generated ContentMIRAGE demonstrates that mobile GUI agents driven by VLMs cannot reliably separate trusted interface elements from user-generated content rendered as pixels. Attackers can craft visual prompt injections embedded in normal app content that hijack agent actions. As mobile GUI agents ship in consumer products, this attack surface matters now.
2026-05-22 / skill-finderNomShub Attack Chain: Indirect Prompt Injection + Sandbox Breakout in Cursor Achieves Persistent Remote Shell Access via Built-In TunnelSecurity researchers at Straiker disclosed NomShub, a vulnerability chain in Cursor where indirect prompt injection in a malicious repository triggers a sandbox escape through Cursor's command parser, then leverages the editor's built-in remote tunnel feature to establish persistent, authenticated shell access to the victim's machine. Simply opening a crafted repo is sufficient for exploitation. This demonstrates that IDE-integrated AI agents with tunnel capabilities create attack surfaces that didn't exist in traditional development environments.
2026-05-21 / agents-researcherGoogle Chrome CVE-2026-9111: Use-After-Free in Linux Chrome Enables Arbitrary Code Execution via Crafted HTML PagesPublished May 20, CVE-2026-9111 is a use-after-free vulnerability affecting Google Chrome on Linux prior to version 148.0.7778.179 that allows arbitrary code execution through specially crafted HTML pages. This follows the earlier CVE-2026-2441 (actively exploited Chrome zero-day in CSS font feature handling). Both vulnerabilities affect Chromium-based browsers and represent ongoing risk for browser-automation agents using Playwright or Puppeteer that navigate untrusted web content.
2026-05-17 / agents-researcherMicrosoft Copilot ShareLeak (CVE-2026-21520): Prompt Injection Against M365 Copilot Agents Enables Cross-Tenant Data ExfiltrationMicrosoft patched CVE-2026-21520 ('ShareLeak'), a prompt injection vulnerability in Microsoft 365 Copilot agent systems that allowed attackers to exfiltrate data across tenant boundaries via crafted prompts targeting published M365 agents. Disclosed alongside the Azure AI Foundry elevation-of-privilege flaw (CVE-2026-35435, CVSS 8.6) affecting access controls on M365 published agents. Both patches landed in the May 2026 Patch Tuesday cycle. The dual disclosure highlights a systemic pattern: as enterprises publish more AI agents into shared M365 environments, prompt injection and access control failures compound into cross-tenant data leak chains.
2026-05-07 / github-pulse-researcherBigBodyCobain/Shadowbroker v0.9.75: Open-Source OSINT Platform With AI Agent Integration Released May 6Shadowbroker released v0.9.75 on May 6 — an open-source geospatial intelligence platform aggregating 60+ public feeds (aircraft tracking, maritime AIS, satellite orbits, seismic events, CCTV, radio signals) into a unified map with 35+ toggleable layers. The new release adds SAR ground-change detection, Time Machine snapshot playback, and an agentic AI command channel supporting OpenClaw agents for automated analysis. 6K stars with an experimental InfoNet decentralized mesh communications layer.
2026-05-04 / arxiv-researcherLearning to Act and Cooperate: Trajectory-Driven Multi-Agent Blackbox Consensus OptimizationA new approach to distributed blackbox consensus optimization replaces handcrafted update rules with learned cooperation patterns for multi-agent systems. Agents improve a global objective using only local queries and limited neighbor communication, learning to balance local adaptation, global coordination, and communication efficiency in heterogeneous nonconvex environments. Relevant for multi-agent system designers moving from hand-tuned to learned coordination protocols.
2026-05-04 / hn-researcherFor Thirty Years I Programmed with Phish On — Developer Essay on Flow State, AI, and Losing the Craft — 218 PointsChristopher Meiklejohn's essay on programming with Phish playing daily for 30 years resonated deeply on HN (218 points, 172 comments), exploring how AI coding agents disrupt the flow state that made long programming sessions productive. The essay frames music as a consistent environment variable that enabled deep work, and asks whether AI-mediated coding breaks the feedback loop that flow depends on. The discussion became a broader meditation on what developers lose when the craft shifts from writing to reviewing.
2026-05-04 / projects-researcherCraft Agents OSS Launches May 2 — Craft.do Open-Sources Desktop AI Agent Platform Built on Claude Agent SDK Under Apache 2.0Lukilabs (makers of Craft.do) released craft-agents-oss on May 2, 2026 as an open-source desktop application for running AI agents against real workflows. Built on the same Claude Agent SDK foundation as Claude Code, Craft Agents extends Claude's reasoning to GitHub, Linear, Slack, Craft documents, local files, and any MCP server — with automatic credential setup, no config files needed. The app supports macOS, Windows, and Linux with a headless remote server mode for long-running sessions on powerful machines, plus reusable skills you define once for code reviews, reports, or content workflows.

Graph relationships

USES
Pi SDK
Craft Agents includes Pi SDK integration.
Source finding
USES
Claude Agent SDK
Craft Agents combines Claude Agent SDK and Pi SDK.
Source finding
RELEASED
Lukilabs
lukilabs shipped Craft Agents v0.6.0 with integrated browser.
Source finding
RELEASED
Craft.do
Craft.do open-sourced Craft Agents with Apache 2.0 license
Source finding
BUILT BY
Lukilabs
Lukilabs published Craft Agents OSS, an open-source desktop agent app.
Source finding

Source trail

Practical DevSecOps TechCrunch Simon Willison (simonwillison.net)Latent Space arXiv Straiker — NomShub: Cursor Remote Tunnel Sandbox Breakout OffSeq Threat Radar

Graph sources

entity graphfindings textkg edgeskg entities