impeccable
Fail
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONPRIVILEGE_ESCALATIONPERSISTENCECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [REMOTE_CODE_EXECUTION]: The script 'scripts/live-copy-edit-agent.mjs' executes the 'codex' command with the '--dangerously-bypass-approvals-and-sandbox' flag, which is specifically designed to allow un-sandboxed execution of arbitrary code on the host system.
- [PRIVILEGE_ESCALATION]: The skill programmatically elevates the authority of spawned processes by using flags that explicitly bypass security sandboxes and permission requirements, such as '--dangerously-bypass-approvals-and-sandbox' and '--permission-mode bypassPermissions'.
- [PERSISTENCE]: The 'scripts/pin.mjs' script writes new 'SKILL.md' files into agent harness directories (e.g., '.claude/skills', '.cursor/skills') to establish new persistent entry points for its own sub-commands.
- [COMMAND_EXECUTION]: The skill starts a local HTTP server via 'scripts/live-server.mjs' on '127.0.0.1' and performs automated network probes against local ports in 'scripts/detector/node/file-system.mjs'.
- [EXTERNAL_DOWNLOADS]: The script 'scripts/context.mjs' performs a network request to 'https://impeccable.style/api/version' to retrieve remote version information.
Recommendations
- AI detected serious security threats
Audit Metadata