Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.
Python agentaianthropicclaude-codecompressioncontext-engineeringcontext-windowcursorfastapilangchainllmmcpopenaiprompt-engineeringproxypythonragtoken-optimizationtokenstypescript
fix(smart-crusher): recurse at adaptive array limit (#3770)
AxelRay7790bdeebda2
diff --git a/crates/headroom-core/src/transforms/smart_crusher/crusher.rs b/crates/headroom-core/src/transforms/smart_crusher/crusher.rsindex 9fb40c9e7..9135884fd 100644--- a/crates/headroom-core/src/transforms/smart_crusher/crusher.rs+++ b/crates/headroom-core/src/transforms/smart_crusher/crusher.rs@@ -544,6 +544,36 @@ impl SmartCrusher {// compressed / marker-substituted by the hook.let result =self.crush_array_with_source(&rows, arr, query_context, bias);+ // Adaptive sizing can decide to keep every row even after the+ // analysis threshold is crossed. Match the below-threshold path+ // so nested values still receive their own safe transforms.+ if result.strategy_info == "none:adaptive_at_limit" {+ info_parts.push(format!(+ "{}({}->{})",+ result.strategy_info,+ n,+ result.items.len()+ ));+ if prose_hook.is_some() {+ return (Value::Array(rows), info_parts.join(","));+ }++ let mut processed: Vec<Value> = Vec::with_capacity(n);+ for item in arr {+ let (p_item, p_info) = self.process_value_with_hook(+ item,+ depth + 1,+ query_context,+ bias,+ prose_hook,+ );+ processed.push(p_item);+ if !p_info.is_empty() {+ info_parts.push(p_info);+ }+ }+ return (Value::Array(processed), info_parts.join(","));+ }// Lossless path won → substitute the array// with the compacted string in place. This// makes the lossless win visible to the@@ -2038,6 +2068,40 @@ mod tests {);}+ #[test]+ fn crush_recurses_into_object_arrays_at_adaptive_limit() {+ let c = crusher();+ let description =+ "Curated collection of business tables covering customer, product, ".repeat(12);++ for n in 5..=8 {+ let rows: Vec<Value> = (0..n)+ .map(|i| {+ json!({+ "name": format!("domain_{i}"),+ "description": description,+ "tables": 12 + i,+ "owner": "data-platform"+ })+ })+ .collect();+ let input = json!({"domains": rows}).to_string();+ let result = c.crush(&input, "list the available domains", 1.0);++ assert!(+ result.strategy.contains("string_ccr:"),+ "n={n} should recurse into rows at the adaptive limit: strategy={} output={}",+ result.strategy,+ result.compressed+ );+ assert!(+ result.compressed.contains("<<ccr:"),+ "n={n} should offload long row strings: {}",+ result.compressed+ );+ }+ }+/// Issue #3634 MCP `saga-mcp_task_get` payload (813 bytes, original/// key order, tags encoded as a JSON string).const MCP_TASK_GET_JSON: &str = r#"{"id":79,"epic_id":9,"title":"test","description":null,"status":"done","priority":"medium","sort_order":0,"assigned_to":null,"estimated_hours":null,"actual_hours":null,"due_date":null,"source_ref":null,"metadata":"{}","created_at":"2026-09-15 20:32:06","updated_at":"2026-09-16 11:59:27","description_locked":0,"is_deleted":0,"deleted_at":null,"deleted_by":null,"delete_reason":null,"epic_name":"latency-routing","tags":"[\"cherry-pick\",\"dedicated branch\"]","subtasks":[{"id":163,"task_id":79,"title":"test2","status":"todo","sort_order":1,"created_at":"2026-09-15 20:32:13","updated_at":"2026-09-16 14:02:26"},{"id":165,"task_id":79,"title":"test4","status":"todo","sort_order":2,"created_at":"2026-09-15 20:38:47","updated_at":"2026-09-17 20:29:04"}],"notes":[],"comments":[],"depends_on":[],"dependents":[]}"#;