Struct PromptInjectionHook 

pub struct PromptInjectionHook { /* private fields */ }

Detects and sanitizes prompt injections in tool output.

Scans tool results in the on_after_tool_call phase; when any pattern in DEFAULT_INJECTION_PATTERNS (or a custom pattern) matches, replaces the whole result with a safe placeholder so malicious instructions never reach the next plan() prompt (blocking cross-turn pollution).

Example

use lc_agents::hooks::PromptInjectionHook;

let executor = AgentExecutor::new(agent, tools)
    .hook(PromptInjectionHook::new());

Implementations

impl PromptInjectionHook

pub fn new() -> Self

Creates the hook with the default injection patterns.

pub fn with_patterns(self, patterns: Vec<String>) -> Self

Replaces the default patterns with a custom list.

pub fn with_marker(self, marker: impl Into<String>) -> Self

Custom replacement placeholder; when it contains {}, it is filled with the matched pattern text.

pub fn detect(&self, text: &str) -> Option<&str>

Detects whether the text contains an injection pattern; returns the matched pattern (None if no match).

pub fn detected_count(&self) -> usize

Cumulative number of injections detected.

Trait Implementations

impl AgentHook for PromptInjectionHook

fn on_after_tool_call( &self, ctx: &mut ToolResultContext, ) -> Result<(), HookError>

Called after a tool call. Can modify the result.

fn on_before_completion(&self, _ctx: &mut CompletionContext) -> CompletionAction

Called before an LLM completion. Can modify messages or reject the call.

fn on_after_completion( &self, _ctx: &mut CompletionResult, ) -> Result<(), HookError>

Called after an LLM completion. Can modify the response.

fn on_before_tool_call(&self, _ctx: &mut ToolCallContext) -> ToolCallAction

Called before a tool call. Can approve, reject, modify, or skip.

fn on_stream_chunk(&self, chunk: &str) -> StreamAction

Called for each streaming token. Can filter, replace, or forward.

fn on_agent_start(&self, _input: &str) -> Result<(), HookError>

Called when the agent starts execution.

fn on_agent_end(&self, _output: &str) -> Result<(), HookError>

Called when the agent finishes execution.

fn on_error(&self, _error: &HookError) -> ErrorAction

Called when an error occurs. Can retry, ignore, or propagate.

impl Default for PromptInjectionHook

fn default() -> Self

Returns the “default value” for a type.