verbatim-audit-notify
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill uses several shell scripts to perform local operations. These include
launchctlfor managing daily timers on macOS,securityfor accessing the macOS Keychain, and standard utilities likecurl,jq,sed, andawkfor data processing and network requests to the official Pushover API. - [EXTERNAL_DOWNLOADS]: The skill communicates with
api.pushover.netto send notifications and check account quotas. This is the primary intended function of the skill. It also fetches credentials from 1Password via theopCLI if configured by the user. - [CREDENTIALS_UNSAFE]: The skill implements a secure credential loading pattern (self-custody-first). It attempts to load Pushover tokens from environment variables, then the macOS login-Keychain, and finally from a 1Password vault. It explicitly avoids passing secrets as command-line arguments to prevent exposure in process lists, using
curlconfiguration files on stdin instead. - [DATA_EXFILTRATION]: While the skill sends data to Pushover, it is restricted to the notification title and message (limited to 1024 characters). The full "verbatim" payload is stored only in a local JSONL audit file (
~/.local/state/pushover/) and is not sent over the network.
Audit Metadata