The Agent Skills Directory

[DATA_EXFILTRATION]: The skill includes a 'bridge' component designed to expose local files to the internet. The implementation in bridge/src/state.js and bridge/src/assets.js allows for resolving arbitrary file paths, including those in the user's home directory (~/). This can be exploited to expose sensitive files like ~/.ssh/id_rsa or .env configurations via public Cloudflare or ngrok tunnels.
[COMMAND_EXECUTION]: The skill frequently uses child_process.spawn to execute shell commands for managing tunnels, checking for dependencies, and installing software. Commands are executed across multiple files including bridge/src/daemon.js, bridge/src/deps.js, and bridge/src/tunnel.js.
[REMOTE_CODE_EXECUTION]: In bridge/src/tunnel.js, the skill uses npx --yes localtunnel to download and execute code from the npm registry at runtime. Furthermore, bridge/src/deps.js contains logic to download and execute repository configuration scripts for Linux distributions.
[EXTERNAL_DOWNLOADS]: The skill fetches software and configuration from external sources. It retrieves Cloudflare's GPG keys and repository configurations from https://pkg.cloudflare.com and utilizes the npm registry for dynamic tool execution.
[CREDENTIALS_UNSAFE]: The skill's architecture encourages users to store sensitive Agnes AI API keys in a plaintext config.json file on the local filesystem.

yufan-Agnes