The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/model_usage.py executes the codexbar CLI tool to retrieve usage data. The implementation uses subprocess.check_output with an argument list rather than a shell string, effectively preventing command injection. Additionally, the provider argument is restricted to a set of predefined choices (codex, claude) via argparse.
[EXTERNAL_DOWNLOADS]: The skill's metadata describes installing the codexbar tool using a Homebrew tap (steipete/tap/codexbar). This is a well-known developer source (Peter Steinberger) and is used for its intended purpose of providing the required binary for the skill to function.
[DATA_EXFILTRATION]: There are no network operations (such as requests, urllib, or curl) present in the skill's code. All data processing occurs locally, and results are printed to the standard output.
[SAFE]: The logic for processing JSON cost logs includes type checking and validation of input structures. It accesses local configuration directories for Codex and Claude as described in its documentation to provide the requested summaries.

model-usage