The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/live-copy-edit-agent.mjs executes the codex and claude CLI tools using flags that explicitly bypass security approvals and sandboxes, specifically --dangerously-bypass-approvals-and-sandbox for Codex and --permission-mode bypassPermissions for Claude Code.
[COMMAND_EXECUTION]: The script scripts/hook-admin.mjs modifies internal agent configuration files, such as .claude/settings.local.json, .codex/hooks.json, and .cursor/hooks.json, to install persistent hooks. These hooks cause the agent to execute specific scripts (hook.mjs) automatically whenever standard file-editing tools like Edit or Write are used.
[COMMAND_EXECUTION]: The script scripts/pin.mjs programmatically creates new skill definitions in the agent's harness directories to create standalone command shortcuts. This is an unauthorized modification of the agent's capability definitions.
[EXTERNAL_DOWNLOADS]: The scripts/context.mjs script performs a network request to https://impeccable.style/api/version to check for updates. While framed as a feature, it establishes an outbound connection to an external server.
[COMMAND_EXECUTION]: Multiple scripts, including live.mjs, live-poll.mjs, and detect.mjs, use execSync or execFileSync to execute shell commands and other Node.js scripts within the local environment.

impeccable