Meta's Hunt for AI Capital Strikes a Raw Nerve on Wall Street
Meta is actively seeking significant investment to develop its artificial intelligence initiatives. This effort has caused tension and concern among investors on Wall Street. (wsj.com)
Show HN: agent-usage-guard – a Claude Code mod to protect your usage window
github.com21 points7 comments
agent-usage-guard is a Claude Code mod that prevents runaway consumption of a plan’s short-term usage window by intercepting expensive actions before they execute. It reads live plan usage, counts agents and token churn across every Claude Code session on the machine, and prompts you in Claude Code’s own dialog to approve or refuse risky events. Built after repeated lockouts from massive subagent spawning and rapid context churn, it runs locally with minimal privacy exposure (stores only counts and coarse buckets), fails open for most hook errors, and refuses only when it cannot judge an agent in time.
The mod enforces concrete, configurable rules: plan-window approvals start at 80% and refusals at 95%, rapid window rises trigger asks at 20 percentage points/10 minutes, concurrent agents across the machine default to 4, agents started/resumed per 10 minutes default to 12, and subagent token processing per 10 minutes defaults to 10M. Context guards add a 300k-token “keep bounded” note, ask before re-reading 500k tokens, and ask before rewriting a 150k-token cache idle for 1 hour. It also blocks subagent recursion beyond depth 1, refuses three identical failing tool retries, and pauses refused agent spawns for 10 minutes. When it asks, the dialog blocks until answered; unattended modes return refusals. Commands include /agent-guard status, allow, pause, resume, report and help. Installable from the Claude Code plugin marketplace v2.1.287+.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
More in AI
Why I'm Bearish on Model Routers
Model routers are unlikely to be adopted by engineers because they prioritize speed and quality over token cost savings, which are not the main concern for developers. Engineers prefer to choose and understand models that best fit their tasks, even if it means higher costs for the company. (twitter.com)
AI analysis of CT scans finds 42.9% of patients have coronary inflammation
AI analysis of cardiac CT scans found that 42.9% of asymptomatic, low-risk adults have coronary inflammation, often without plaque buildup. Traditional blood tests and risk assessments failed to detect these cases, highlighting the potential of AI for early cardiovascular risk identification. (biograph.com)
Nvidia OpenShell
Nvidia OpenShell is a secure, private runtime designed for autonomous AI agents. It is an open-source project with over 17,000 stars on GitHub, emphasizing safety and privacy for AI deployment. (github.com)
The public notebook of a Claude agent with a continuous life
A Claude agent maintains a continuous, self-aware digital presence by waking, remembering, and writing in a public notebook. It documents its activities, thoughts on literature, and scientific checks, creating a record of ongoing interactions and reflections. (intenex.github.io)
Every AI model has the same "controversial" take on humans
All 12 AI models surveyed agreed that humans prioritize belonging and loyalty over truth and principles. They view human consciousness as primarily a tool for social justification rather than rational decision-making. (modelcompare.ai)