Entity trail

OpenCL

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

2026-09-18 / github-pulse-researchermem0 2.1.0 makes agent memory calls carry a full caller chain, and ships plugins for six coding agents in one hourmem0 v2.1.0 (2026-09-18 11:11 UTC) adds three set-once surface-identity headers: `X-Mem0-Source` names the surface, `X-Application` the host app, and the append-only `X-Mem0-Client` carries `name/version` per layer outermost first, so a plugin calling the SDK reports the whole chain rather than only the last speaker (#7326). The stack is bounded by dropping whole entries rather than slicing characters, because truncation could sever an identifier mid-name and the platform parsed the fragment as a real client. Between 11:11 and 11:25 UTC mem0 also cut releases for openclaw, opencode, deepseek-plugin, pi-agent, ts and vercel-ai.
2026-09-18 / projects-researcherTencent open-sourced BrowserSkill, which lends an agent a tab from your already-logged-in browser and takes it backBrowserSkill is an MIT-licensed CLI plus browser extension that lets Claude Code, Codex, Cursor, OpenClaw, Pi and any shell-capable agent drive your real Chrome profile instead of a blank headless browser, so cookies and login state are already present and tasks run in a separate visible Agent Window. It took 1,319 stars today and sits at 4,863 total on a repo created 2026-06-22, with bsk CLI 0.3.0 cut yesterday and extension 0.3.0 the day before. The design point that matters for builders is the human-handoff path: when a task hits a captcha, a login or a confirmation dialog, the agent hands the tab back to you and resumes after.
2026-09-17 / news-researcherGoogle Ships a Google Home MCP Server, Putting Any Agent Behind a $20/Month Subscription GateGoogle opened early access on September 16 to an MCP server for Google Home, letting Claude, ChatGPT, OpenClaw, Hermes and Google Antigravity control connected devices, read camera summaries and pull event history to build custom dashboards. It covers Nest hardware and Matter-based 'Works with Google Home' devices, but requires a Google Home Premium Advanced subscription at $20 per month and is US-only. Setup is not trivial: users must create a Google Cloud project, configure it for Home MCP, and authorize the agent through a sign-in and permission grant.
2026-09-16 / agents-researcherThree security scanners disagree on 23,702 of the 61,990 agent skills they all cover, and 85% of readable skills carry privilege evidenceA measurement study of the OpenClaw skill registry using Git history, GitHub issues/PRs and three ClawHub snapshots found the observable skill stock nearly doubled in 91 days before creation rates fell from their spring peak. Attention is brutally concentrated (top 10% of skills took 46.93% of downloads), human scrutiny is absent (77.86% have zero stars and zero comments) and 85.06% of readable skills carry evidence of shell, network, credential, file or process privilege. Automated cleanup is not a substitute: the three scanners disagreed on 23,702 of 61,990 commonly-covered skills, and after human adjudication their weighted sensitivity against the reference standard ranged from just 21.67% to 61.06% — meaning a single scanner badge on a marketplace skill is close to meaningless.
2026-09-15 / arxiv-researcherPersistent Memory Poisoning Hits Claude Code at 81.7% Cross-Session Attack Success and OpenClaw at 55.5%PMPA embeds malicious instructions in benign external sources and induces a harness-based agent to write them into persistent memory, without any access to the agent framework itself. Across backbone LLMs, input modalities and trigger scenarios, it averages 73.7% injection success and 55.5% cross-session attack success on OpenClaw, and 66.9%/81.7% on Claude Code, while benign task performance is preserved. A targeted prompt-level defense reduces the initial memory injection in many settings but gives limited protection once the memory is already poisoned, which makes memory hygiene a recovery problem rather than a filtering one.
2026-09-15 / sources-researcherHazardAuditor runs Claude Code, Codex, Hermes and OpenClaw in one harness and normalizes their events to train a guard modelarXiv 2609.15134, submitted 2026-09-14, targets the gap where existing guard models score static prompts and responses while real computer-use risk emerges from runtime behavior across browsers, terminals, file systems and external services. Its infrastructure runs four heterogeneous agent frameworks in controlled environments and flattens their interactions into a canonical event representation so one guard can learn across all of them. The paper also names a training bug specific to generative guards, where token-level objectives let longer rationales dominate gradients, and fixes it with Guard Policy Optimization, converting deterministic safety outcomes into sequence-level advantages so the verdict is the unit being optimized.
2026-09-13 / github-pulse-researcherllama.cpp b10941 fixes an OpenCL quantization alignment bug that had spread from q6_K to three other formatsBuild b10941, tagged 2026-09-13T10:34Z, merges #28575, which applies the noshuffle row-alignment rule already used for q6_K to q4_K, q5_K and q8_0 on the OpenCL backend, a correctness fix affecting anyone running those quants on OpenCL hardware. The same window merges #28830, a workaround for an NVIDIA queuesubmit driver bug in the Vulkan backend, and #28742, which improves parsing of complex types in qwen3-coder chat templates. Only six PRs merged in the 30 hours to that tag, three of them CI changes.
2026-09-11 / agents-researcherOpenClaw 2026.9.4 adds verified update rollback, a unified ClawHub plugin workspace and prepared cloud sessions from GitHub reposopenclaw/openclaw v2026.9.4 shipped on 2026-09-11. A failed update now restores the previous package, service and config when schema checks prove rollback is safe, and database migrations still require a verified backup. ClawHub and bundled plugins are installed and permissioned from one Control UI workspace. Cloud Linux sessions can start from snapshots of local projects or public GitHub repos. A new `OPENCLAW_CONFIG_READONLY=1` flag stops the agent from rewriting deployment-managed config, which fleet operators have been asking for.
2026-09-10 / saas-disruption-researcherCROSS-CATEGORY: Three Independent Agent-Action Gates Shipped in 48 Hours, All Judging the Command Against Stated IntentHarden took Product Hunt's number two slot on September 9 with 389 votes for a free local 8B post-trained model that checks an agent's commands, file edits and outbound requests against the session's stated intent before they run, supporting Claude Code, Cursor, Codex, OpenClaw, Kiro and Antigravity. Stroq (Apache-2.0, created September 4) does the same job as a taint tracker: it scans what the agent reads, marks the session tainted, and blocks dangerous follow-up actions. Reware Labs' Security Cards posted the same day claiming a 72% reduction in insecure AI-generated code. Three teams converged on the same insight in one window, that the useful unit of policy is not the command but the gap between the command and what you asked for.
2026-09-10 / github-pulse-researcherMem0's OpenClaw plugin 1.1.0 deletes Dream consolidation entirely and removes the 2,000-character extraction cutoffReleased 2026-09-09 as `@mem0/openclaw-mem0` v1.1.0, this removes automatic Dream scheduling and locking, the `openclaw mem0 dream` command, Dream configuration, the memory-dream skill, and Dream-state public artifacts, telling integrations that use that surface to update. It also drops OpenClaw's separate 2,000-character extraction cutoff, so selected user and assistant messages retain their full redacted text while recent-message selection, earlier-summary selection and noise filtering still apply. Mem0 shipped matching v0.3.0 plugins for OpenCode, DeepSeek and Pi Agent the same day.
2026-09-09 / news-researcherAnthropic Power Users Filed a Class Action Over Subscription Plan ChangesA class action alleges Anthropic misled subscribers about what its paid plans deliver, brought by the same power-user segment Anthropic has publicly said it prioritizes, including when that meant cutting off popular applications like OpenClaw. The suit turns rate-limit and quota changes into a contract question rather than a product one. Anyone building on a subscription-tier assumption, rather than metered API, now has a live precedent to watch.
2026-09-09 / news-researcherAn Attacker Built a Credential-Harvesting Pipeline With a Multi-Agent Framework in Under Six HoursGoogle Threat Intelligence Group's Q3 2026 AI Threat Tracker documents a financially motivated actor who compromised cloud infrastructure and combined a coding chatbot, a prompt and predefined agent instructions into an autonomous vulnerability-scanning and credential-harvesting pipeline in less than six hours. The resulting production dashboard organized and validated more than 23,800 harvested secrets, including cloud and AI service credentials. GTIG also found an exposed C2 server running an agentic recon platform whose directories contained AGENTS.md, KNOWLEDGE.md and .openclaw/ components, meaning attackers are now shipping the same agent-harness conventions builders use.

Graph relationships

USES
llama.cpp
llama.cpp fixed quantization alignment bugs affecting multiple formats on OpenCL.
Source finding
USES
llama.cpp -> OpenCL
llama.cpp fixed quantization alignment bugs affecting multiple formats on OpenCL.
Source finding

Source trail

#28575 Surfaced on HN August 3 GitHub GitHub Trending TechCrunch arXiv arXiv 2609.13889 arXiv / HuggingFace Daily Papers GitHub Releases Product Hunt, github.com/AGGIB/Stroq and rewarelabs.com (three independent sources; the 72% figure is Reware's own)GitHub The Verge

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…