🧠 Automated AI Knowledge Base for Product Managers — 4-pillar system with self-evolving knowledge graph. Updates every 4h.

tokensched

TypeScript dependency scanner that flags imperative prose aimed at coding agents hidden in package files, docstrings, and MCP manifests

Curated Hermes Agent skills & runbooks — personal knowledge workflows, ready to drop into your own agent. | 精选 Hermes Agent Skill 合集，记录和分享 hermes-agent 使用经验，开箱即用。

你的个人 AI 雷达：跟踪 AI 圈 X 动态 + 发布自己的想法。自托管 Next.js + SQLite，开箱内置 15 个 AI 领域账号。 | Personal AI radar: track notable AI voices on X and publish your own notes. Self-hosted Next.js + node:sqlite.

Score the AI-generated code slop your agents left behind — a zero-config static SlopScore (0-100) + heatmap for JS/TS repos

Structured verification gate for AI agent edit loops — AST-level proof that edits actually landed

Stripe's Idempotency-Key, but for your browser agent — wrap a transactional step so a self-healing retry is deduplicated, not double-submitted.

Adapt GLM-5.2 tool-calling to the OpenAI tool_calls schema so coding-agent harnesses stop silently mis-parsing GLM function calls

Forensic timeline that shows which bash command your AI ran at 3am

CacheGuard（缓存卫士）— a drop-in proxy that keeps DeepSeek's server-side prefix-cache stable in front of any coding agent, so cache-hit pricing never silently breaks.

Open-source LLM agent action space for driving legacy Windows thick-client apps via the UIA accessibility tree.

CiteGuard — catch fabricated citations & CVEs in papers and PRs, in CI. Deterministic citation/identifier verifier (DOI/arXiv/CVE/commit/GitHub issue) with a GitHub Action.

excise

agentfuse

Hierarchical org→team→dev token-budget proxy for LLM traffic: meter, soft-throttle, hard-deny, and in-flight preemption across providers on one shared wallet.

Runtime seccomp-style per-call capability sandbox for installable agent skills — declare what each skill may touch, enforce it at the call site, trap+log violations.

Claude Code & Codex agent honesty verifier — one command surfaces fixes the coding agent claimed but never made.

AgentFuse — per-task spend circuit-breaker for LLM agents

refusalscope