Merge pull request #14457 from arpanghoshal/add-ctrlrun

Add CTRLRun to Security 🤖🤖🤖

README.md

@@ -3873,6 +3873,7 @@ Tools for conducting research, surveys, interviews, and data collection.
 - [Jiangw2718i/frisk](https://github.com/Jiangw2718i/frisk) [![Jiangw2718i/frisk MCP server](https://glama.ai/mcp/servers/Jiangw2718i/frisk/badges/score.svg)](https://glama.ai/mcp/servers/Jiangw2718i/frisk) 📇 🏠 ☁️ - Screen the counterparty of an x402 payment before an agent pays it. One tool, `screen_payment`, runs deterministic checks — address sanity, dynamic-`payTo` swap detection, transport safety and your own spend policy — and returns `allow`/`review`/`block` with the reasons for the verdict. Advisory: it never holds funds, and your code decides. Runs entirely on your machine by default with no key, no account and no network call (MIT, 5 packages installed); set `FRISK_API_KEY` to answer the same tool from the hosted reputation graph instead. `npx frisk-mcp`
 
 - [epistemedeus/skillguard](https://github.com/epistemedeus/skillguard) [![epistemedeus/skillguard MCP server](https://glama.ai/mcp/servers/epistemedeus/skillguard/badges/score.svg)](https://glama.ai/mcp/servers/epistemedeus/skillguard) 📇 🏠 - Static security scanner for Claude Code skills, plugins, and MCP servers — vet a dependency for malware (secret/env exfiltration, install-time hooks, prompt injection, committed binaries) **before** you install it. Static-only: it clones and reads files, never executes the scanned code. `npx -y github:epistemedeus/skillguard mcp`
+- [CTRLRun/ctrlrun](https://github.com/CTRLRun/ctrlrun) [![CTRLRun/ctrlrun MCP server](https://glama.ai/mcp/servers/CTRLRun/ctrlrun/badges/score.svg)](https://glama.ai/mcp/servers/CTRLRun/ctrlrun) 🐍 🏠 - Execution safety for AI agent actions. `ctrlrun mcp-operator` exposes the approval queue as tools (`list_pending_approvals`, `approve`, `deny`, `resolve`, plus `receipts`, `effects`, `stats` and `inspect_action`), so the human who has to answer a held action answers it from the assistant they are already in; read tools answer without a credential and write tools refuse without one that names a person. Separately, the gateway proxies any MCP server so every tool call is checked against a YAML policy before it runs: allow, hold for a human, or deny. Approvals are single-use and bound to the hash of the exact action, one logical effect runs at most once across processes and hosts, and an unknown outcome is recorded AMBIGUOUS rather than FAILED so a blind retry is refused. Every attempt leaves a hash-chained JSON receipt. Apache-2.0. `pip install ctrlrun`
 
 - [Drumworks/ssid-mcp](https://github.com/Drumworks/ssid-mcp) [![Drumworks/ssid-mcp MCP server](https://glama.ai/mcp/servers/Drumworks/ssid-mcp/badges/score.svg)](https://glama.ai/mcp/servers/Drumworks/ssid-mcp) 📇 ☁️ - MAC-address (OUI) vendor lookup and router default-login directory for AI agents. Identify a device manufacturer from its MAC address, detect randomized/private (locally-administered) addresses instead of reporting "unknown", or fetch a router's default login IP and admin credentials — every router field cited to the manufacturer's own documentation. Free tier, no signup. `npx -y ssid-mcp`