The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Configures the installation of the codexbar CLI utility from a third-party Homebrew tap (steipete/tap/codexbar).
[COMMAND_EXECUTION]: Executes the codexbar CLI tool using subprocess.check_output. Command arguments for the provider are constrained to a fixed set of options ('codex', 'claude'), effectively preventing command injection.
[COMMAND_EXECUTION]: The model_usage.py script includes an --input argument that allows for reading arbitrary local files. The script validates that the input is valid JSON before processing, but this provides a generic file reading capability to the agent.
[DATA_EXFILTRATION]: The skill uses tools that access local session logs (e.g., ~/.codex/sessions/) containing conversation history to extract usage metadata. While sensitive, this access is limited to the purpose of calculating local costs and no network exfiltration was detected.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by processing external data from log files or user-provided input.
Ingestion points: scripts/model_usage.py reads data from files via --input or from sys.stdin.
Boundary markers: No specific delimiters or safety instructions are used to separate ingested data from agent commands.
Capability inventory: The skill can execute subprocesses and read filesystem data.
Sanitization: Input is parsed as JSON, ensuring structural integrity, but the content is not sanitized to prevent malicious instructions embedded within the logs from influencing the agent.

model-usage