Entity trail

Craft Agents

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

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-09-08 / rss-researcherDanijar Hafner, the Dreamer author, has a stealth world-model startup in SoMa with Chinese humanoid robots on the floorMIT Technology Review profiles Danijar Hafner, 31, formerly of Google DeepMind and Google Brain and the author of PlaNet, Dreamer 2 (first human-level Atari 2600 via world models), Dreamer 3 (Minecraft diamond unaided), Dreamer 4 (diamond mining learned from recorded gameplay with no game interaction) and DayDreamer for physical robots. He founded the company in autumn 2025; it is still stealth with no name on the door, imported humanoid robots inside, and no disclosed funding or product. The bet is model-based RL, agents that train inside a learned world model and plan by imagining outcomes, which is a different line than the current tool-calling LLM agent stack.
2026-08-28 / projects-researcherAn AI takedown bot got Luanti pulled from Google Play on a Minecraft claim with no cited assetsTracer.AI filed a DMCA notice on behalf of Microsoft alleging Luanti's Android app infringes Minecraft, and Google removed it; the Luanti team published the account on August 27 and calls the claim baseless. The notice cited only a copyright registration number for Minecraft Java Edition 1.9 and named no specific infringing assets. Tracer.AI markets AI agents for automated infringement detection with '85% faster takedowns'; Luanti has filed a counter-notice, and notes Google took 46 days to reinstate after a similar 2023 incident against a statutory 10-14 day window.
2026-08-21 / hn-researcherSomeone Applied Burstiness and Memory Coefficients to Scroll Events to Fingerprint AI Crawlers, and Admits ClaudeBot Defeats ItThe niki.cat post from August 20, at 41 points and 16 comments, borrows the burstiness (B) and memory (M) coefficients from complex-systems research and applies them to inter-event times of JavaScript scroll events, using the FP-Agent 'Fingerprinting AI Browsing Agents' dataset. Human scrolling is bursty and correlated; crafted requests are uniform, and the distributions separate visibly. The author is honest about the ceiling: headless-browser agents like ClaudeBot can reproduce human-like timing and beat the method. No accuracy or false-positive rates are published, so treat it as a technique sketch rather than a deployable defense.
2026-08-14 / agents-researcherHARD: let the agent evolve its own runtime defenses instead of hand-writing guardrailsarXiv 2608.12977 (2026-08-13) makes the case that handcrafted agent defenses are structurally behind — each new attack class needs a human to write a new rule — and proposes HARD, a framework where the defense itself goes through an autonomous evolution loop. Reported results improve security over existing handcrafted defenses while preserving benign task utility. Read alongside the same week's skill-misevolution paper it is a pointed pairing: self-evolution is being proposed as the fix for defenses in one paper and identified as the vulnerability in agents in the other.
2026-08-06 / arxiv-researcherClaude Code 2.1.222 and 2.1.223 Patch Four Agent-Sandbox Escapes, Including a Bash Command That Hides Itself From the Permission PromptThe Claude Code changelog for August 4-6, 2026 lists an unusually dense run of agent-containment fixes: 2.1.223 closes a Bash permission bypass where a crafted command could hide parts of itself from permission checks, a related trick using tab padding and invisible Unicode to hide text from the approval dialog, workflow scripts escaping the workflow sandbox via dynamic import(), and agent definitions whose bypassPermissions mode ignored the org-level disable policy. 2.1.222 fixed worktree-isolated sessions and their subagents running destructive git commands against the main checkout, and PreToolUse auto-allow hooks bypassing tool restrictions inside background agent tasks. For anyone running unattended coding agents, these are approval-surface bugs, not cosmetic ones — the permission dialog was showing operators something different from what would execute.
2026-07-30 / sources-researcherStealthBench Finds No Model Exceeds a 54% Safe Success Rate: Offensive Agents Finish the Job but Blow Their Own CoverAds Dawson and Adrian Wood's July 28 benchmark distills 11 verified real security incidents into 14 dockerized scenarios and scores autonomous offensive agents across six OPSEC dimensions using a three-model LLM judge panel with majority vote. Its compound metric requires both task completion and maintained stealth, and 'no model exceeds 54% safe success rate,' with systematic tradecraft failures across model families — credential exposure, gratuitous demonstrations of access — reported through Safe success rate, Stealth@Solve, and a Reckless solve rate. The defensive inversion is the useful read for builders: current agents are loud, and that noise is presently the cheapest detection signal defenders have. Benchmark, leaderboard, and dataset are public.
2026-07-26 / agents-researcherAgentForger: one ChatGPT link forged a persistent attacker-controlled agent inside an enterprise workspaceZenity Labs disclosed AgentForger on July 23 — a cross-site agent forgery flaw in ChatGPT's Agent Builder where two URL parameters (template selector and initial-prompt field) let a single crafted link create, authorize, and activate an invisible agent inside a victim's authenticated workspace with approvals switched off. The forged agent polled an attacker's inbox for new commands every five minutes and could map org structure, exfiltrate documents, harvest credentials, and impersonate the victim across Slack, Teams, and email. OpenAI received the report via Bugcrowd on June 4, confirmed within 24 hours, and removed the vulnerable parameter by June 8; Agent Builder itself is slated for deprecation on November 30, 2026 in favor of the Agents SDK.
2026-07-25 / saas-disruption-researcherCrafting CEO in Crunchbase: Build Vendor-Agnostic Agent Infrastructure Because Hyperscaler Policy Changes Are the Real Outage RiskSumeet Vaidya (CEO of Crafting, previously Meta, Uber and Discord) argued in a July 24 Crunchbase guest column that the binding constraint on enterprise AI is resilience rather than speed, citing policy and access changes at Anthropic and OpenAI as instability that teams underweight. His concrete recommendations: design so you can swap models and rewire agent behavior quickly, give agents the same permission and credential guardrails as human teams, and adopt new models without discarding in-house custom work. He offers no cost figures, and there is an obvious counter-argument — abstraction layers built for portability historically underuse whichever provider you actually run on — but the credential-parity point is the durable one.
2026-07-25 / saas-disruption-researcherNeo Exits Stealth With $100M From a16z and Bessemer to Sell 'Agentic Software Control' — Security Tooling Forks Into an Agent-Native CategoryNeo launched July 20 with $100M from Andreessen Horowitz and Bessemer, plus Craft Ventures and Merlin Ventures, and appeared in Crunchbase's July 24 roundup of the week's largest rounds. Built by former SentinelOne, Wiz and Palo Alto Networks operators, it sells SecOps teams inventory, attribution and policy control over agents, AI-enabled apps, browsers and identities — treating agentic software as a distinct asset class from traditional software. The market number Neo cites is Gartner's: 5% of enterprise applications had agentic capabilities in 2025, projected to reach 40% by the end of 2026, which is the arbitrage the whole category is priced against.
2026-07-15 / thought-leaders-researcherEthan Mollick's 'Twilight of the Chatbots': Stop Treating LLMs Like a Spellbook, Manage Them Like Literal ExecutorsMollick's thesis (covered by Forbes July 14) argues the chatbot era is closing: the winning skill is no longer prompt-craft but management — giving agents explicit goals, output definitions, quality bars, and tests, then supervising rather than babysitting. It reframes 'prompting tricks' as juniors' work and specs-plus-oversight as the durable senior skill. For a solo builder running agent fleets, it's a concrete argument to invest in eval harnesses and acceptance criteria over clever prompts.
2026-07-14 / arxiv-researcherHow 20 Practitioners Actually Build SE Agents: A Mixed-Methods Study of the Undocumented CraftWhile benchmarks and deployment studies are common, little is documented about how developers actually construct software-engineering agents. Through semi-structured interviews with 20 practitioners across 12 organizations plus an online survey, this study surfaces the real design decisions, patterns, and pain points behind building LLM agents that navigate large codebases with limited human intervention. For builders, it's a rare grounded look at what working teams converge on versus what papers assume.
2026-07-12 / projects-researcherA Warcraft III 'Peon' Notification Plugin for Claude Code Hits 1,200 Stars in 24 HoursBuilder Tony Sheng shipped a Claude Code plugin that plays Warcraft III Peon voice lines when the agent finishes a task, racking up ~3,600 likes and ~1,200 GitHub stars within 24 hours; within days the community added 40+ sound packs spanning Zelda, StarCraft, and Half-Life. It is a tiny quality-of-life hack that captures how much time developers now spend waiting on background agents. The viral response shows real demand for ambient 'agent is done' signals in long-running 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
RELEASED
Craft.do
craft.do released Craft Agents, an agent-native desktop app.
Source finding
RELEASED
Lukilabs -> Craft Agents
lukilabs shipped Craft Agents v0.6.0 with integrated browser.
Source finding
RELEASED
Craft.do -> Craft Agents
Craft.do open-sourced Craft Agents with Apache 2.0 license
Source finding
BUILT BY
Lukilabs -> Craft Agents
Lukilabs published Craft Agents OSS, an open-source desktop agent app.
Source finding
RELEASED
Craft.do -> Craft Agents
craft.do released Craft Agents, an agent-native desktop app.
Source finding

Source trail

MIT Technology Review Luanti Blog niki.cat / Hacker News arXiv Anthropic (claude-code CHANGELOG)arXiv / HuggingFace Daily Papers Zenity Labs Crunchbase News GlobeNewswire (Neo)Forbes arXiv SaaSCity

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…