llm-loop-breakerTwo-layer defense for Openclaw gateway: kills hallucinating LLM streams via entropy analysis, and watchdogs host CPU/RAM/Disk to prevent resource exhaustion.
Install via ClawdBot CLI:
clawdbot install hehkcaspar/llm-loop-breakerGrade Fair — based on market validation, documentation quality, package completeness, maintenance status, and authenticity signals.
Accesses system directories or attempts privilege escalation
/var/log/Calls external URL not in known-safe list
https://github.com/openclaw/clawhubAudited Apr 17, 2026 · audit v1.0
Generated Oct 6, 2026
A platform team running Openclaw as their LLM gateway deploys the Stream Guard to automatically abort streams that fall into repetitive hallucination loops. Entropy compression ratios above 10 trigger hard kills, preventing runaway token spend and downstream data corruption from garbage outputs.
A SaaS provider hosting dozens of LLM-powered apps on shared Openclaw workers uses the Python watchdog to prevent one tenant's runaway agent from exhausting CPU, RAM, or disk. Poison-pill and memory-leak rules kill offending process trees before they cascade into cluster-wide outages.
An IDE-integrated coding assistant backend deploys the guard to catch models that get stuck emitting repeated code snippets or whitespace. The soft-kill condition (compression ratio >6 + single-char dominance) stops degenerate generations while preserving valid long-form code output.
Teams running unattended overnight agent pipelines rely on the crash-loop detector and incident snapshots to recover from flapping gateway processes without human intervention. Journalctl and dmesg excerpts captured to the incidents folder enable morning root-cause analysis.
Deployments on resource-constrained Linux edge nodes use the watchdog's redline breach logging and memory-leak detection to keep inference gateways alive on limited hardware. The idempotent deploy script enables rapid rollout across fleets of edge devices.
Offer a hosted Openclaw gateway with Stream Guard pre-installed as a value-added reliability layer for enterprises that don't want to manage hallucination and resource-exhaustion defenses themselves. Position uptime and token-spend protection as the core selling point.
License the Stream Guard as a premium reliability module for existing Openclaw enterprise customers, bundled with incident snapshot analytics and compliance-friendly audit logs. Sell via annual seat or node-based contracts.
Distribute the guard open-source to drive adoption, then monetize a Pro dashboard that aggregates incident snapshots, entropy kill stats, and resource trends across fleets. Target DevOps teams needing observability into LLM reliability.
💬 Integration Tip
Run deploy.sh once per gateway rebuild and always restart the gateway afterward; verify by executing both bundled test suites before enabling in production. Tune the entropy thresholds for your specific model, since verbose models may legitimately exceed the default compression ratio during long structured outputs.
Scored Oct 6, 2026
通过网页抓取获取 GitHub 按日/周/月增长的热门仓库。当用户询问 GitHub 趋势、热门项目、本周热点或「什么在 GitHub 上 trending」时使用。可输出列表或 JSON,无需 API Key。
GitHub 操作技能 - 创建仓库、推送代码、管理 Release。全自动,无需用户干预。
Deep-dive analysis of GitHub projects. Use when the user mentions a GitHub repo/project name and wants to understand it — triggered by phrases like "帮我看看这个项目...
Git-based version control for OpenClaw system configuration changes. Provides two core capabilities: (1) SAVE - Create a checkpoint before sensitive operatio...
Use when preparing roadmap narratives, release notes, changelogs, or stakeholder updates tailored for executives, engineering teams, and customers.
PR Review Expert