Struct CompactionPolicyDeclaration 

pub struct CompactionPolicyDeclaration {
    pub strategy: PolicyStrategy,
    pub max_tokens: Option<usize>,
    pub max_turns: Option<usize>,
    pub context_window: Option<usize>,
    pub safety_ratio: f64,
    pub keep_last: usize,
    pub keep_first: usize,
    pub hard_limit_tokens: Option<usize>,
    pub tool_output_max_chars: Option<usize>,
    pub summarize_fn: Option<VmValue>,
    pub summarize_prompt: Option<String>,
    pub instructions: CompactionPolicy,
}

Declared inputs that drive the compaction_check decision and the downstream compaction_run call. Values are normalized at registration time so consumers can rely on every threshold being either populated or explicitly None.

Fields

strategy: PolicyStrategy

max_tokens: Option<usize>

Hard cap on estimated tokens before compaction_check returns compact_now. None means tokens alone never trigger.

max_turns: Option<usize>

Hard cap on message count. None means message count alone never triggers.

context_window: Option<usize>

When context_window is set, compaction_check fires when estimated tokens exceed context_window * safety_ratio. Both must be set for the ratio rule to apply.

safety_ratio: f64

keep_last: usize

How many recent messages to keep verbatim during compaction.

keep_first: usize

How many initial messages to keep verbatim during compaction.

hard_limit_tokens: Option<usize>

Token budget passed through to the engine for tier-2 compaction (used by summarize-then-prune to escalate from LLM to truncate when the LLM result still exceeds the cap).

tool_output_max_chars: Option<usize>

Per-tool-result microcompaction threshold. None keeps the engine default.

summarize_fn: Option<VmValue>

Closure invoked when strategy is custom.

summarize_prompt: Option<String>

Optional prompt template path used when the engine selects LLM summarization.

instructions: CompactionPolicy

Author/scope/preserve/drop directives that the engine threads through the LLM compaction prompt and persisted metadata.

Implementations

impl CompactionPolicyDeclaration

pub fn token_threshold(&self) -> Option<usize>

Token budget at which the policy considers a session “full”. Resolves the most restrictive of max_tokens and context_window * safety_ratio; returns None when neither rule is configured.

pub fn evaluate( &self, estimated_tokens: usize, message_count: usize, ) -> EvaluationContext

Decision metadata describing every triggered threshold. Returned as part of the CompactionDecision so callers can log what pushed them over the line.

pub fn to_json(&self) -> Value

JSON snapshot for telemetry payloads.

Trait Implementations

impl Clone for CompactionPolicyDeclaration

fn clone(&self) -> CompactionPolicyDeclaration

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for CompactionPolicyDeclaration

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for CompactionPolicyDeclaration

fn default() -> Self

Returns the “default value” for a type.