Struct ClaudeCliBackend 

pub struct ClaudeCliBackend { /* private fields */ }

AI backend that spawns claude -p in non-interactive mode.

Implementations

impl ClaudeCliBackend

pub fn new(config: &AiConfig) -> Self

pub fn is_available() -> bool

Check if the claude CLI is available on PATH.

pub fn check_auth() -> Result<(), CoreError>

Pre-flight auth check: sends a minimal prompt WITHOUT –json-schema (which returns immediately on auth failure) and checks is_error. This prevents the infinite StructuredOutput retry loop that occurs when –json-schema is used with an expired/missing auth token.

impl ClaudeCliBackend

pub fn execute_agentic( &self, prompt: &str, cwd: Option<&str>, ) -> Result<BackendResponse, CoreError>

Execute an agentic prompt: unlimited turns, full tool access, raw markdown output.

Key differences from execute():

• No --max-turns (unlimited turns for codebase exploration)
• No --tools "" (model needs tool access)
• No --json-schema (output is raw markdown)
• Longer timeout: 600 seconds (10 minutes)
• Result comes from result field (not structured_output)
• Optional cwd to set the working directory for tool calls

Trait Implementations

impl AnalysisBackend for ClaudeCliBackend

fn execute( &self, prompt: &str, json_schema: Option<&str>, ) -> Result<BackendResponse, CoreError>

Execute a prompt and return the response text and cost. When json_schema is provided, the backend passes it to --json-schema for constrained decoding (guaranteed valid JSON matching the schema).