@pentoshi/clai 3.11.27 → 3.11.29

package/dist/agent/runner.d.ts

@@ -58,6 +58,7 @@ export interface AgentRunOptions {
     previousTurn?: PreviousTurnSignal | undefined;
     /** User-declared model window for this provider/model/session. */
     contextLimitTokens?: number | undefined;
+    getContextLimitTokens?: (provider: ProviderId | undefined, model: string | undefined) => number | undefined;
 }
 export declare function runAgentTurn(prompt: string, options?: AgentRunOptions): Promise<import("./turn-outcome.js").TurnOutcome>;
 /** Compatibility boundary for callers that consume visible assistant text. */

package/dist/agent/runner.js

@@ -458,6 +458,9 @@ export async function runAgentTurn(prompt, options = {}) {
         let provider = initialProvider;
         await ensureProviderConfigured(provider);
         let model = initialModel;
+        const currentContextLimitTokens = () => options.getContextLimitTokens
+            ? options.getContextLimitTokens(provider, model)
+            : options.contextLimitTokens;
         // Some Groq free-tier models have a per-request/per-minute input budget
         // below the normal agent prompt alone. Select a purpose-built compact
         // instruction set before the request is made, rather than treating the
@@ -2931,11 +2934,12 @@ export async function runAgentTurn(prompt, options = {}) {
         }
         async function maybeAutoCompact(reason, force = false) {
             const beforeTokens = estimateNextRequestTokens(messages);
+            const contextLimitTokens = currentContextLimitTokens();
             const compactTrigger = autoCompactTriggerTokens(getReliabilityPolicy(), {
                 provider,
                 model,
-                ...(options.contextLimitTokens
-                    ? { contextLimitTokens: options.contextLimitTokens }
+                ...(contextLimitTokens !== undefined
+                    ? { contextLimitTokens }
                     : {}),
             });
             if (!force && beforeTokens < compactTrigger)
@@ -3196,6 +3200,7 @@ export async function runAgentTurn(prompt, options = {}) {
                             writeNotice("info", notice, chalk.dim(`  ℹ ${notice}\n`));
                         }
                     }
+                    const contextLimitTokens = currentContextLimitTokens();
                     await auditLog("agent.turn", {
                         provider,
                         model,
@@ -3204,7 +3209,13 @@ export async function runAgentTurn(prompt, options = {}) {
                         step,
                         // Metadata-only composition metrics (no prompt/tool text).
                         ...contextBreakdownAuditPayload(contextBreakdown),
-                        compactTriggerTokens: autoCompactTriggerTokens(),
+                        compactTriggerTokens: autoCompactTriggerTokens(getReliabilityPolicy(), {
+                            provider,
+                            model,
+                            ...(contextLimitTokens !== undefined
+                                ? { contextLimitTokens }
+                                : {}),
+                        }),
                         maxTokensBudget: resolveStepMaxTokens({
                             nativeToolsActive,
                             toolsAttached,