The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/model_usage.py uses subprocess.check_output to run the codexbar CLI tool. The command is constructed securely using a list of arguments, and the input parameters (like the provider name) are constrained to a predefined set of safe choices, which effectively prevents shell command injection.
[EXTERNAL_DOWNLOADS]: The skill's metadata points to an external Homebrew repository (steipete/tap/codexbar) to install the required codexbar dependency. While this is a standard method for dependency management on macOS, users should be aware that it involves installing third-party software.
[INDIRECT_PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection as it ingests data from external sources and outputs it to the agent's context.
Ingestion points: scripts/model_usage.py reads JSON data from the output of a shell command or from local files specified by the user via the --input flag.
Boundary markers: Absent. The script prints summarized text directly to standard output, which is then seen by the agent without any delimiters or instructions to ignore embedded commands or malicious text within the logs.
Capability inventory: The skill possesses the capability to execute the codexbar binary and read files from the local file system (provided they are in JSON format).
Sanitization: No sanitization is performed on the data fields (such as model names) extracted from the logs before they are formatted into the final report for the agent.

model-usage