Things worth keeping - tools, hacks, repos, and rabbit holes from the build.
Rust MCP server that pre-indexes a codebase with PageRank, blast-radius, git co-change, and complexity signals. Agents query the graph instead of grepping files.
Interactive terminal dashboard that breaks down where your AI coding tokens actually go. Surfaces the chat-vs-tool-use split most users get wrong.
Curated AGENTS.md preset that kills sycophancy, blocks drive-by refactors, and forces verification loops. Synthesizes Karpathy's principles with Cherny's Claude Code workflow.
Rust TUI that monitors Claude Code and Codex sessions: token spend, context window, rate limits, and ports in real time. Like btop but for agent runtimes.
Multi-agent harness for Claude Code with planning, parallel execution waves, and a five-agent review gate. Hub mode lets you drive sessions from Telegram or Slack.
CLI, GitHub Action, and GitHub App that scan agent configs, MCP servers, and tool permissions for vulnerabilities. Detects skill poisoning and prompt-injection vectors.
Open-source Windows/macOS desktop that one-click installs Claude Code, MCP tools, and Skills with sandbox isolation, multi-model support, and Slack/Feishu hooks.
Entity-level diff, blame, graph, and impact analysis for code. 23 languages parsed via tree-sitter so reviews work on functions instead of line ranges.
One package, one port, zero deps to fake LLM APIs, MCP servers, A2A, AG-UI, vector DBs, and search backends in tests.
Memory layer for agents that surfaces relevant context instead of just similar embeddings. Episodic plus long-term memory backed by a knowledge graph.
Self-hosted email client with an embedded AI agent, running entirely on Cloudflare Workers. No backend to manage, edge-distributed by default.
Rust-native headless browser purpose-built for agent-driven scraping and web automation. Faster cold starts and lower memory than Chromium-based stacks.
Reverse proxy for Claude Code that strips IPs, hashes, credentials, and PII before requests hit Anthropic. Dual-layer detection: local Ollama LLM plus regex.
Native Rust + Tauri 2 terminal with local shells, SSH, SFTP, remote IDE, AI agent, MCP, RAG, and a file manager - all in one binary. No OpenSSL dependency.
Skill plus CLI that has coding agents generate inspectable browser-automation scripts instead of running prompts at runtime. You debug a real script, not a prompt.
One memory + skills + protocols folder that plugs into Claude Code, Cursor, Windsurf, OpenCode, Hermes, or a DIY Python harness - and survives the switch.
Aggregates many MCP servers behind one endpoint. Acts as an MCP gateway/proxy so clients only configure a single server.
MCP server that lets agents query and analyze Prometheus metrics through a standardized tool interface. Useful for ops-aware coding agents.
MCP client with persistent sessions, stdio + HTTP transports, OAuth 2.1, JSON output for code mode, and a sandbox proxy. Calls any MCP server from a shell.
Benchmark harness that evaluates models and agents on real-world MCP usage. Comparable scores across servers and frontier models.
Runtime adapter that exposes any MCP, OpenAPI, or GraphQL server as a flat CLI. Zero codegen, zero rebuild - handy for shell scripts and agent toolchains.