Struct PlanningAgent 

pub struct PlanningAgent { /* private fields */ }

Wraps any Agent to enforce read-only tool access for planning.

Filters tools via prepare_tools to only allow read-only operations. Sets plan_mode: true in context so tools can check and adapt behavior.

Implementations

impl PlanningAgent

pub fn new(inner: Box<dyn Agent>) -> Self

pub fn with_allowed_tools(self, tools: Vec<String>) -> Self

Override the set of allowed tools (replaces default READ_ONLY_TOOLS).

pub fn allow_tool(self, name: impl Into<String>) -> Self

Add extra tools to the allowed set (e.g. custom read-only tools).

Trait Implementations

impl Agent for PlanningAgent

fn decide<'life0, 'life1, 'life2, 'async_trait>( &'life0 self, messages: &'life1 [Message], tools: &'life2 ToolRegistry, ) -> Pin<Box<dyn Future<Output = Result<Decision, AgentError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait, 'life2: 'async_trait,

Given messages and available tools, decide what to do next.

fn prepare_tools( &self, _ctx: &AgentContext, tools: &ToolRegistry, ) -> Vec<String>

Hook: filter or reorder tools before each step. Returns tool names to include. Default: all tools.

fn prepare_context(&self, ctx: &mut AgentContext, messages: &[Message])

Hook: modify context before each step. Default: no-op.

fn after_action(&self, ctx: &mut AgentContext, tool_name: &str, output: &str)

Hook: called after tool execution with the tool name and output. Can modify context or messages. Default: no-op.

fn decide_stateful<'life0, 'life1, 'life2, 'life3, 'async_trait>( &'life0 self, messages: &'life1 [Message], tools: &'life2 ToolRegistry, _previous_response_id: Option<&'life3 str>, ) -> Pin<Box<dyn Future<Output = Result<(Decision, Option<String>), AgentError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait, 'life2: 'async_trait, 'life3: 'async_trait,

Stateful decide — returns response_id for multi-turn chaining.