The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands to detect the host environment (e.g., command -v claude, claude auth status) and to invoke the Claude CLI for advisor tasks. The execution pattern for the Fable CLI channel uses shell variable interpolation for the model alias and prompt file path, but employs stdin redirection (< "$PROMPT_FILE") to prevent shell injection from the prompt content itself.
[DATA_EXFILTRATION]: The skill explicitly identifies that repository context and transcripts are sent across external data boundaries to remote models. It includes instructions for the agent to warn the user about what is being shared and to exclude secrets, credentials, or unrelated proprietary content before performing a consult.
[CREDENTIALS_UNSAFE]: The skill follows security best practices by using env -u ANTHROPIC_API_KEY when invoking the Claude CLI. This ensures the tool uses the signed-in session or subscription rather than accidentally leaking or billing an ambient API key stored in the environment.
[SAFE]: The skill uses a vendor-specific environment variable BOPEN_ADVISOR_MODEL to allow user configuration of the advisor model, which is a standard pattern for tools authored by b-open-io.

advisor