usage-monitor
Fail
Audited by Gen Agent Trust Hub on Mar 13, 2026
Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill retrieves sensitive OAuth access tokens by querying the macOS Keychain using the 'security find-generic-password' command and by reading the file '~/.claude/.credentials.json'.
- [COMMAND_EXECUTION]: The script 'scripts/usage-check.sh' executes the 'security' system utility to access stored credentials and pipes results to Python for processing.
- [COMMAND_EXECUTION]: The Python script 'scripts/codex-usage-scrape.py' executes the 'codex' CLI tool via a subprocess to interface with an app server using the 'stdio://' protocol.
- [COMMAND_EXECUTION]: The skill uses 'python3 -c' and heredoc blocks to run dynamically constructed Python code for calculating budget targets and parsing JSON payloads.
- [EXTERNAL_DOWNLOADS]: The skill performs network requests to 'https://api.anthropic.com/api/oauth/usage' to fetch account utilization data.
- [DATA_EXFILTRATION]: Authentication tokens extracted from the local environment (keychain and files) are sent to the external Anthropic API to authorize usage queries.
Recommendations
- AI detected serious security threats
Audit Metadata