Composing APIs and CLIs in the LLM era
17 days ago
- #LLM
- #Automation
- #CLI
- Two strategies for structuring tool descriptions in LLMs: top-level tools for fine-grained actions vs. using shell commands.
- Using shell commands enables command composition, saving token costs and allowing pipeline customization by humans.
- Restish is a tool that interprets OpenAPI specs, enabling CLI-like interaction with SaaS APIs without custom code.
- OAuth2c simplifies OAuth 2.0 authorization flows, reducing boilerplate code for API access.
- Secure token storage on macOS can be achieved using the `security` CLI with specific flags to enforce passcode prompts.
- Adversarial interoperability: LLMs can reverse-engineer APIs from HAR files for services lacking official APIs.
- Combining CLI tools (Restish, OAuth2c) reduces maintenance and bug surface while enabling powerful automation.