The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run to manage the bridge service, interact with Homebrew, and execute CLI-based model providers. These executions are scoped to system management and the skill's primary purpose. Security is maintained through manifest validation which restricts input to literal kebab-case flags, preventing arbitrary command injection.
[EXTERNAL_DOWNLOADS]: The skill automates the installation of the cliproxyapi dependency using Homebrew, a well-known and trusted package management service. This installation is triggered only during the explicit bootstrap workflow.
[DATA_EXPOSURE]: The script accesses sensitive local configuration files such as models.json and bridge secrets. It mitigates exposure risks by implementing a redact_text function that strips API keys, bearer tokens, and long hex strings from all generated reports and logs. It also enforces 0600 (owner-only) permissions on all credential-bearing files and backups.
[INDIRECT_PROMPT_INJECTION]: The skill has a data ingestion surface where it processes JSON responses from a local HTTP proxy to probe model capabilities. While this data influences the final application configuration, the risk is minimized by strict JSON parsing, lack of shell evaluation for data fields, and the requirement for models to pass specific text and streaming probes before registration.

workbuddy-cli-model-bridge