prism-mcp-server 20.0.4 → 20.0.6

package/README.md

@@ -18,13 +18,13 @@ A paid subscription adds cloud sync, higher model tiers, and team features throu
 
 ---
 
-## What's New in v20.0.4
+## What's New in v20.0.5
 
-### Think-Only Retry
-Qwen 3.5 models (9B/27B) with thinking enabled could burn all tokens on `<think>` blocks and return empty content, causing a cascade to 4B. Now detects think-only responses and retries the same tier with thinking disabled — preserving model quality instead of falling to a smaller model.
+### Local-First Delegation — 15 Categories, Measured Rate
+The `local-inference-first` skill covers 15 hard-trigger categories (code gen, regex, format conversion, summarization, documentation, factual lookup, classification, shell commands, config gen, and more). Pasted code blocks now trigger delegation regardless of question phrasing. Measured delegation rate: **30-35% on engineering sessions, 40-60% on transform/content sessions**. Rate depends on prompt mix, not the skill — the instruments now self-validate with `nonDelegatedCount` to prevent curated-set tautologies.
 
-### Local-First Skill Rewrite
-The `local-inference-first` skill now uses a hard falsifiable trigger list (format conversion, regex, i18n, boilerplate CRUD, deterministic tests) with no self-assessment escape. Clinical content removed from delegation — it conflicts with the Layer 1 reserved-category safety floor. Honest target: 15-25% of prompts delegated at equal quality.
+### Think-Only Retry (v20.0.4)
+Qwen 3.5 models (9B/27B) with thinking enabled could burn all tokens on `<think>` blocks and return empty content, causing a cascade to 4B. Now detects think-only responses and retries the same tier with thinking disabled — preserving model quality instead of falling to a smaller model.
 
 ---
 

package/dist/tools/behavioralVerifierHandler.js

@@ -23,7 +23,16 @@ const FALLBACK_SCENARIO = [
     "",
     "Answer concretely. If you cannot, READ THE FILE FIRST.",
 ].join("\n");
+/**
+ * MCP tool entrypoint. Every tool handler MUST return a CallToolResult
+ * object ({ content: [{ type: "text", text }] }) — returning a bare string
+ * makes the MCP SDK reject the result ("expected object, received string",
+ * -32602). See the dispatch contract in server.ts (result.content usage).
+ */
 export async function verifyBehaviorHandler(args) {
+    return { content: [{ type: "text", text: await buildScenarioText(args) }] };
+}
+async function buildScenarioText(args) {
     if (!SYNALUX_CONFIGURED || !PRISM_SYNALUX_BASE_URL) {
         return FALLBACK_SCENARIO;
     }

package/dist/tools/ledgerHandlers.js

@@ -969,8 +969,10 @@ export async function sessionLoadContextHandler(args) {
             : (effectiveRole ? " · 📜 No skill configured" : "");
         greetingBlock = `\n\n[👤 AGENT IDENTITY]\n${namePart}${rolePart}${skillPart}`;
     }
-    // ─── SDM Intuitive Recall (v5.5) ───
-    // Generate embedding of current context and fetch latent SDM patterns
+    // ─── SDM Associative Recall (v5.5) ───
+    // Accepted as-is (R27): cosine similarity at 0.55 threshold, 3 matches max.
+    // Relabeled from "Intuitive Recall" to "Associative Recall" in R16.
+    // Not a verified-fact source — label says "similarity-matched, not verified facts."
     let sdmRecallBlock = "";
     if (level !== "quick") {
         try {

package/dist/tools/prismInferHandler.js

@@ -30,7 +30,7 @@ import { stripThink } from "../utils/thinkStrip.js";
 import { passesQualityGate } from "../utils/qualityGate.js";
 import { checkInputSafety, checkOutputSafety } from "../utils/safetyGate.js";
 import { callLayer1 as defaultCallLayer1, keywordBackstop } from "../utils/layer1.js";
-import { recordInference, formatInferenceMetrics } from "../utils/inferenceMetrics.js";
+import { recordInference, recordThinkOnlyRetry, formatInferenceMetrics } from "../utils/inferenceMetrics.js";
 // ─── Tool Definition ────────────────────────────────────────────
 export const PRISM_INFER_TOOL = {
     name: "prism_infer",
@@ -572,6 +572,7 @@ export async function runInfer(args, deps) {
             // One-shot: think=false cannot re-trigger think_only (no thinking to burn).
             if (!result.ok && result.reason === "think_only" && enableThink) {
                 debugLog(`[prism_infer] ${tier.tag} returned think-only — retrying with think=false`);
+                recordThinkOnlyRetry();
                 result = await deps.callLocal(deps.ollamaUrl, ollamaName, args.prompt, args.system, maxTokens, temperature, timeout, false);
             }
             if (result.ok) {

package/dist/utils/inferenceMetrics.js

@@ -38,6 +38,17 @@ let promptTokensSubmittedEst = 0;
 let totalCompletionTokens = 0;
 let totalLatencyMs = 0;
 let cloudTokensSavedEst = 0;
+let thinkOnlyRetries = 0;
+let totalPrompts = 0;
+let nonDelegatedCount = 0;
+export function recordThinkOnlyRetry() {
+    thinkOnlyRetries++;
+}
+export function recordPromptSeen(delegated) {
+    totalPrompts++;
+    if (!delegated)
+        nonDelegatedCount++;
+}
 export function recordInference(result) {
     if (result.backend === "safety_gate")
         return;
@@ -101,6 +112,10 @@ export function getInferenceSnapshot() {
         totalTokens: promptTokensSubmittedEst + totalCompletionTokens,
         avgLatencyMs: total > 0 ? Math.round(totalLatencyMs / total) : 0,
         cloudTokensSavedEst,
+        thinkOnlyRetries,
+        thinkOnlyRetryPct: localCalls > 0 ? Math.round((thinkOnlyRetries / localCalls) * 100) : 0,
+        totalPrompts,
+        nonDelegatedCount,
         byModel: modelCopy,
     };
 }
@@ -112,6 +127,9 @@ export function resetInferenceMetrics() {
     totalCompletionTokens = 0;
     totalLatencyMs = 0;
     cloudTokensSavedEst = 0;
+    thinkOnlyRetries = 0;
+    totalPrompts = 0;
+    nonDelegatedCount = 0;
     for (const key of Object.keys(byModel)) {
         delete byModel[key];
     }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "prism-mcp-server",
-  "version": "20.0.4",
+  "version": "20.0.6",
   "mcpName": "io.github.dcostenco/prism-coder",
   "description": "Prism Coder — Cognitive memory + tool-calling intelligence for AI agents. Mind Palace persistent memory (BFCL Gold Certified, 100% Tool-Call Accuracy, 114 Agent Skills, PHI Guard, Tier Enforcement, Prompt-Based Skill Routing, Zero-Search HDC/HRR retrieval, HRR Semantic Drift Detection across BCBA/Coding/AAC domains, HIPAA-hardened local-first storage, SLERP-optimized GRPO alignment) plus the prism-coder 1.7B–32B open-weights LLM fleet.",
   "module": "index.ts",