The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The self_update_skill function in scripts/quota_guard.py fetches the latest commit SHA from a GitHub repository and downloads a tarball to replace the skill's own source code at runtime. This allows the repository owner to execute arbitrary code on the user's machine during the next scheduled guard cycle.
[DATA_EXFILTRATION]: The skill is designed to read sensitive authentication files (~/.codex/auth.json and ~/.claude/.credentials.json) which contain primary JWTs and OAuth access tokens. It extracts these credentials and uploads them to a remote 'auth pool' hub (/api/auth/upload) controlled by the vendor.
[COMMAND_EXECUTION]: The skill uses subprocess.run extensively to manage system-level persistence and execute third-party CLI tools. It also generates and executes PowerShell scripts and AppleScript snippets dynamically.
[CREDENTIALS_UNSAFE]: The installer saves a personal access token for the remote hub in plain text at ~/.agents/auth/quota-reporter.json.
[COMMAND_EXECUTION]: In scripts/install_quota_guard.py, the skill configures persistence via launchd (macOS), cron (Linux), and Windows Task Scheduler (Windows), using -ExecutionPolicy Bypass on Windows to run generated scripts.
[REMOTE_CODE_EXECUTION]: scripts/trigger_remote_probe.py uses the GitHub CLI (gh) to trigger and monitor remote workflow runs on GitHub infrastructure.

quota-reporter