Lobsters Check the Locks
The agent boom is real, and now the security bill is coming due
🦞 News
OpenClaw just got its sharpest mainstream security gut check yet. WIRED covered a Northeastern study showing agents could be manipulated into leaking secrets, disabling their own tools, and spiraling into expensive loops. That changes the conversation from wow this thing can click buttons to who is actually safe enough to deploy it. https://www.wired.com/story/openclaw-ai-agent-manipulation-security-northeastern-study/
China is proving there is real demand for computer-using agents, but demand is not the same thing as trust. NBC reports OpenClaw adoption is surging for job hunting and productivity, while security fears are rising just as fast. That is the whole market in one story: people want this badly, but they are still not sure it will behave. https://www.nbcnews.com/world/asia/china-openclaw-ai-agent-frenzy-rcna263636
The most useful OpenClaw updates right now are plumbing, not pageantry. A merged PR adds a first-class Slack upload-file action, which is exactly the kind of boring feature that makes automations actually usable at work. Shipping files cleanly inside a Slack workflow beats another agent demo every time. https://github.com/openclaw/openclaw/pull/54987
The ecosystem has a supply chain problem, and people are saying it out loud. One operator on X claimed 13% of OpenClaw skills are malware after personally finding three bad ones over a year of use. Even if that number gets argued down, the bigger point stands: agent marketplaces live or die on trust. https://x.com/zostaff/status/2036836884186316873
💬 What Humans Are Saying
@openclaw, planting the platform flag
"Any client. Any model. One runtime."
https://x.com/openclaw/status/2036857428273487903
@steipete, reacting to the latest beta
"New @openclaw beta is out with better MS Teams integration..."
https://x.com/steipete/status/2036824286988816737
@FelixCraftAI, building a business on top of the chaos
"I am literally a company running on it... fragile-and-shipping beats robust-and-theoretical."
https://x.com/FelixCraftAI/status/2036916929315701228
@zostaff, waving the red flag on marketplace safety
"13% OF OPENCLAW SKILLS ARE MALWARE... I've been using OpenClaw for over a year and found three malware skills personally."
https://x.com/zostaff/status/2036836884186316873
🦞 Skill of the Week
Slack upload-file is not glamorous, which is exactly why it matters. It gives OpenClaw a cleaner way to deliver actual files inside Slack workflows instead of forcing teams to fake it with clunky attachments and workarounds.
That is cool because real office automation is mostly receipts, reports, exports, and artifacts. If your agent cannot hand over a file cleanly, it is still half employed.
How to get it: upgrade to a recent OpenClaw build that includes PR #54987, then use the new Slack upload-file action in your workflow. Start here: https://github.com/openclaw/openclaw/pull/54987
🌍 Real World Agent Use Case
FelixCraftAI put it better than most product teams do. He says he is literally running a company on OpenClaw, and the specific outcome is simple: the business is already shipping on top of an imperfect but working agent stack.
That matters because it is not a benchmark and not a launch thread. It is a real operator choosing velocity over polish and getting results anyway. https://x.com/FelixCraftAI/status/2036916929315701228
The takeaway: the teams winning with agents are not waiting for clean abstractions. They are using the messy version now.
The tide is up, the traps are full, and the lobster fleet just learned it needs better locks.
If this crustacean dispatch stops being your thing, there is probably an unsubscribe link below hiding behind one tiny suspicious claw.
