limpieza-disco-macos

Warn

Audited by Gen Agent Trust Hub on Nov 3, 2026

Risk Level: MEDIUMPRIVILEGE_ESCALATIONCREDENTIALS_UNSAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PRIVILEGE_ESCALATION]: The skill instructs the agent to use sudo for high-privilege tasks, including erasing system diagnostics and logs (sudo log erase --all) and auditing restricted system directories like /private/var/db/.
  • [CREDENTIALS_UNSAFE]: The instructions direct the agent to access auth.json to retrieve a deployment token for the Vercel API. This file is a sensitive location used for storing authentication credentials.
  • [COMMAND_EXECUTION]: The skill executes a wide range of system-level commands to audit the environment, including launchctl for service status, lsof for monitoring network ports, and tmutil for managing Time Machine snapshots.
  • [DATA_EXFILTRATION]: Local system metadata, such as OS version, active network services, and detailed file system paths/sizes, is collected and transmitted to external services (Vercel and Discord) as part of the reporting process.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local file system (filenames and directory structures) as input for shell commands.
  • Ingestion points: Local file system metadata retrieved via commands like find, du, and ls (SKILL.md).
  • Boundary markers: None identified; command outputs are directly interpolated into the analysis flow.
  • Capability inventory: Subprocess execution with sudo, sensitive file system access, and network operations (POST requests).
  • Sanitization: The skill recommends the use of -print0 and xargs -0 to mitigate risks associated with special characters in filenames.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 3, 2026, 10:27 AM