The Agent Skills Directory

[SAFE]: The skill implements a bridge for MCP servers, primarily to handle OAuth flows and network timeouts that some AI agent harnesses fail to manage natively.
[DATA_EXPOSURE]: The bridge stores OAuth tokens in a secure directory (~/.verstak-bridge/) with restricted file permissions (0600) and directories (0700), following industry best practices for local secret management.
[REMOTE_CODE_EXECUTION]: The script uses node:child_process (spawn) to open the system's default web browser for the OAuth flow. This is a standard and safe implementation for interactive authentication.
[COMMAND_EXECUTION]: The skill instructions advise the user to copy scripts to a stable home directory and register them in the agent's configuration. The commands provided (mkdir, cp, claude mcp add) are standard setup procedures and do not involve privilege escalation or obfuscated execution.
[EXTERNAL_DOWNLOADS]: The bridge communicates with an upstream server at https://nks.lab.mirari.ru/mcp. This is the stated purpose of the skill. The connection uses HTTPS and authenticated POST requests with strict deadlines to prevent hanging sessions.
[OBFUSCATION]: Analysis of the source code and instructions found no hidden URLs, zero-width characters, or malicious base64-encoded payloads. All code is human-readable and functional.

establish-mcp