Struct DefaultGuardConfig 

pub struct DefaultGuardConfig {

    pub reasoning_only_max_strikes: usize,
    pub reasoning_only_nudge: String,
    pub reasoning_only_action: ReasoningOnlyAction,
    pub disable_thinking_nudge: String,
    pub empty_response_max_strikes: usize,
    pub empty_response_nudge: String,
    pub use_llm_judge: bool,
    pub judge_timeout_secs: u64,
    pub judge_skip_threshold: usize,
    pub judge_fail_open: bool,
    pub detect_short_response: bool,
    pub short_response_min_input: usize,
    pub short_response_max_output: usize,
    pub short_response_nudge: String,
    pub recent_user_count: usize,
}

Guard policies from agent-works (tool gating, reasoning-only enforcement). Default guard configuration

Fields

reasoning_only_max_strikes: usize

Maximum retries for reasoning-only responses

reasoning_only_nudge: String

Nudge message for reasoning-only responses

reasoning_only_action: ReasoningOnlyAction

Reasoning-only handling strategy

disable_thinking_nudge: String

Nudge message when disabling thinking

empty_response_max_strikes: usize

Maximum retries for empty responses

empty_response_nudge: String

Nudge message for empty responses

use_llm_judge: bool

Whether to use LLM judge for text-only after tools

judge_timeout_secs: u64

Timeout in seconds for the LLM judge call

judge_skip_threshold: usize

Skip judge if response is longer than this (likely complete)

judge_fail_open: bool

Whether to trust LLM when judge fails or times out.

• true: fail-open (trust the model, end loop)
• false: fail-closed (don’t trust, force continue)

detect_short_response: bool

Enable short-response detection (merged from CompletionGateMiddleware). When the user input is long but the model response is very short, treat it as potentially incomplete and nudge/judge accordingly.

short_response_min_input: usize

Minimum user input character count to trigger short-response detection.

short_response_max_output: usize

Maximum LLM output character count to be considered a short response.

short_response_nudge: String

Nudge message for short responses

recent_user_count: usize

Number of recent user messages to include in the judge prompt. Helps the judge understand context like “继续” after a multi-turn discussion.

Trait Implementations

impl Default for DefaultGuardConfig

fn default() -> DefaultGuardConfig

Returns the “default value” for a type.