The Agent Skills Directory

[COMMAND_EXECUTION]: The PluginContext API provides a ctx.exec(command, opts?) method that allows plugins to execute arbitrary shell commands on the host system. While described as gated by policy, this provides a direct path for command execution that could be abused by a malicious plugin.
[REMOTE_CODE_EXECUTION]: The skill documentation describes a web-intel plugin that spawns sidecar daemons using ctx.spawnProcess to run npx -y wigolo serve. Spawning processes from dynamically loaded plugins constitutes a remote code execution vector if the plugin source or configuration is compromised.
[REMOTE_CODE_EXECUTION]: The script scripts/test-plugin.mjs contains a hardcoded absolute file path (/Users/olu/work/RTerm/packages/backend/src/services/plugin/pluginRegistry.ts) to import the PluginRegistry. This pattern is fragile and suggests the tool is tailored for a specific local environment, which can be a security risk if it forces an agent to interact with unauthorized or unexpected local file paths.
[COMMAND_EXECUTION]: The scripts/scaffold-plugin.mjs script uses process.env.USER and executes directory creation/file writing based on user-supplied command line arguments without significant sanitization, which could lead to path traversal if the --out or --name arguments are manipulated.

rterm-plugin