Trait HookProtocol

Source

pub trait HookProtocol {
    // Required methods
    fn agent(&self) -> AgentKind;
    fn parse_tool_use(&self, raw: &Value) -> Result<ToolUseHookInput>;

    // Provided methods
    fn parse_post_tool_use(&self, raw: &Value) -> Result<ToolUseHookInput> { ... }
    fn parse_session_start(&self, raw: &Value) -> Result<SessionStartHookInput> { ... }
    fn parse_stop(&self, raw: &Value) -> Result<StopHookInput> { ... }
    fn format_allow(
        &self,
        reason: Option<&str>,
        _context: Option<&str>,
        updated_input: Option<Value>,
    ) -> Value { ... }
    fn format_deny(&self, reason: &str, _context: Option<&str>) -> Value { ... }
    fn format_ask(&self, _reason: Option<&str>, _context: Option<&str>) -> Value { ... }
    fn format_session_start(&self, context: Option<&str>) -> Value { ... }
    fn rewrite_for_sandbox(
        &self,
        input: &ToolUseHookInput,
        sandbox_cmd: &str,
    ) -> Option<Value> { ... }
    fn session_context(&self) -> &str { ... }
}

Expand description

Abstraction over agent-specific hook JSON formats.

Each agent (Claude Code, Gemini CLI, etc.) implements this trait to handle:

Parsing its native JSON stdin into normalized internal types
Formatting decisions back into the agent’s expected JSON output
Rewriting tool inputs for sandbox enforcement

§Adding a New Agent

Only two methods are required: agent and parse_tool_use. All other methods have sensible defaults. Override them only when your agent’s protocol diverges from the common JSON format.

Required Methods§

Source

fn agent(&self) -> AgentKind

Which agent this protocol handles. Required.

Source

fn parse_tool_use(&self, raw: &Value) -> Result<ToolUseHookInput>

Parse the agent’s PreToolUse JSON into a ToolUseHookInput. Required.

The returned tool_name MUST be the internal (Claude-style) name, translated via resolve_tool_name. The original agent-native name is preserved in original_tool_name.