muonroi-cli 1.2.2 → 1.3.1

package/dist/src/pil/layer5-context.js

@@ -0,0 +1,151 @@
+import { promises as fs } from "node:fs";
+import * as path from "node:path";
+import { searchByText } from "../ee/bridge.js";
+import { truncateToBudget } from "./budget.js";
+// TODO(WhoAmI-L5): when EE v4.0 Who Am I profile is available, replace the
+// hardcoded threshold with profile-derived value:
+//   work_patterns.session_length="medium-long (1-3h)" → 120 min stale threshold
+//   work_patterns.multitasking="sequential-deep"      → extend to 2h
+//   (task-switchers with short sessions → keep 15 min)
+// Also: deduplicate with L3 — if L3 already injected experience-principles,
+// skip that collection here to avoid double-injection.
+const STALE_THRESHOLD_MS = 30 * 60 * 1000;
+// Single round-trip for both collections via /api/search; topK=5 yields
+// up to 2 principles + 3 behavioral after server-side scoring.
+const LAYER5_SEARCH_TIMEOUT_MS = 1500;
+// Score floor — points below this are dropped as noise. Same default and
+// override knob as Layer 3 to keep behaviour consistent across PIL.
+const LAYER5_SCORE_FLOOR = (() => {
+    const raw = Number(process.env.MUONROI_PIL_SCORE_FLOOR);
+    return Number.isFinite(raw) && raw >= 0 && raw <= 1 ? raw : 0.55;
+})();
+async function fetchPrinciples(raw, budget) {
+    try {
+        const points = await searchByText(raw, ["experience-principles", "experience-behavioral"], 5, AbortSignal.timeout(LAYER5_SEARCH_TIMEOUT_MS));
+        const filtered = points.filter((p) => (p.score ?? 0) >= LAYER5_SCORE_FLOOR);
+        if (filtered.length === 0)
+            return "";
+        const sorted = filtered.slice().sort((a, b) => (b.score ?? 0) - (a.score ?? 0));
+        const lines = sorted.map((p) => {
+            const payload = p.payload;
+            const text = payload?.text ?? (payload?.json ? JSON.parse(payload.json).solution : "");
+            return `- ${String(text ?? "").slice(0, 120)}`;
+        });
+        return truncateToBudget(`[principles: Always-loaded experience]\n${lines.join("\n")}`, budget);
+    }
+    catch {
+        return "";
+    }
+}
+async function fetchFlowState(cwd, budget) {
+    try {
+        const activeRunPath = path.join(cwd, ".muonroi-flow", "active-run");
+        const runId = (await fs.readFile(activeRunPath, "utf8")).trim();
+        if (!runId)
+            return "";
+        const statePath = path.join(cwd, ".muonroi-flow", "runs", runId, "state.md");
+        const state = await fs.readFile(statePath, "utf8");
+        const phaseLine = state.match(/phase:\s*(.+)/i)?.[1]?.trim() ?? "unknown";
+        const statusLine = state.match(/status:\s*(.+)/i)?.[1]?.trim() ?? "unknown";
+        return truncateToBudget(`[flow: phase=${phaseLine}, status=${statusLine}, run=${runId}]`, budget);
+    }
+    catch {
+        return "";
+    }
+}
+async function fetchRecentFiles(cwd, budget) {
+    try {
+        const srcDir = path.join(cwd, "src");
+        const entries = await fs.readdir(srcDir, { recursive: true, withFileTypes: true });
+        const tsFiles = [];
+        for (const e of entries) {
+            if (!e.isFile())
+                continue;
+            if (!e.name.endsWith(".ts") && !e.name.endsWith(".tsx"))
+                continue;
+            if (e.name.includes(".test.") || e.name.includes("__test"))
+                continue;
+            const full = path.join(e.parentPath ?? e.path, e.name);
+            const stat = await fs.stat(full);
+            tsFiles.push({ name: path.relative(srcDir, full).replace(/\\/g, "/"), mtime: stat.mtimeMs });
+        }
+        tsFiles.sort((a, b) => b.mtime - a.mtime);
+        const top = tsFiles.slice(0, 10).map((f) => f.name);
+        if (top.length === 0)
+            return "";
+        return truncateToBudget(`[recent-files: ${top.join(", ")}]`, budget);
+    }
+    catch {
+        return "";
+    }
+}
+export async function layer5Context(ctx) {
+    // Chitchat short-circuit: greetings/small-talk don't need workspace context
+    // (recent-files, flow-state, principles). Inject nothing — keep the model's
+    // attention on a 5-character prompt instead of a 30K-token tool catalog.
+    if (ctx.intentKind === "chitchat") {
+        return {
+            ...ctx,
+            layers: [...ctx.layers, { name: "context-enrichment", applied: false, delta: "skip:chitchat" }],
+        };
+    }
+    const cwd = process.cwd();
+    const parts = [];
+    const deltaSegments = [];
+    // 1. T0/T1 principles — skip when L1's unified call already supplied them
+    //    (L3 already rendered the principles block from ctx._brainData).
+    if (!ctx._brainData) {
+        const principlesBudget = Math.floor(ctx.tokenBudget * 0.12);
+        const principles = await fetchPrinciples(ctx.raw, principlesBudget);
+        if (principles) {
+            parts.push(principles);
+            deltaSegments.push(`principles=${principles.length}ch`);
+        }
+    }
+    else {
+        deltaSegments.push("principles=skipped-l1-unified");
+    }
+    // 2. Resume digest (5% budget)
+    const digest = ctx.resumeDigest;
+    if (digest?.trim()) {
+        const isStale = typeof ctx.digestAgeMs === "number" && ctx.digestAgeMs > STALE_THRESHOLD_MS;
+        const stalePrefix = isStale ? "(stale — verify before relying)\n" : "";
+        const hint = `[flow-context: Resume]\n${stalePrefix}${digest.trim()}`;
+        const trimmed = truncateToBudget(hint, Math.floor(ctx.tokenBudget * 0.05));
+        parts.push(trimmed);
+        deltaSegments.push(`digest=${trimmed.length}ch${isStale ? ",stale" : ""}`);
+    }
+    // 3. Flow state (5% budget)
+    const flowBudget = Math.floor(ctx.tokenBudget * 0.05);
+    const flowState = await fetchFlowState(cwd, flowBudget);
+    if (flowState) {
+        parts.push(flowState);
+        deltaSegments.push(`flow=${flowState.length}ch`);
+    }
+    // 4. Recent file index (3% budget)
+    const filesBudget = Math.floor(ctx.tokenBudget * 0.03);
+    const fileIndex = await fetchRecentFiles(cwd, filesBudget);
+    if (fileIndex) {
+        parts.push(fileIndex);
+        deltaSegments.push(`files=${fileIndex.length}ch`);
+    }
+    if (parts.length === 0) {
+        return {
+            ...ctx,
+            layers: [...ctx.layers, { name: "context-enrichment", applied: false, delta: "no-context-sources" }],
+        };
+    }
+    return {
+        ...ctx,
+        enriched: `${ctx.enriched}\n${parts.join("\n")}`,
+        layers: [
+            ...ctx.layers,
+            {
+                name: "context-enrichment",
+                applied: true,
+                delta: deltaSegments.join(" "),
+            },
+        ],
+    };
+}
+//# sourceMappingURL=layer5-context.js.map

package/dist/src/pil/layer5-context.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"layer5-context.js","sourceRoot":"","sources":["../../../src/pil/layer5-context.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,IAAI,EAAE,EAAE,MAAM,SAAS,CAAC;AACzC,OAAO,KAAK,IAAI,MAAM,WAAW,CAAC;AAClC,OAAO,EAAE,YAAY,EAAE,MAAM,iBAAiB,CAAC;AAC/C,OAAO,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AAG/C,2EAA2E;AAC3E,kDAAkD;AAClD,gFAAgF;AAChF,qEAAqE;AACrE,uDAAuD;AACvD,4EAA4E;AAC5E,uDAAuD;AACvD,MAAM,kBAAkB,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC;AAE1C,wEAAwE;AACxE,+DAA+D;AAC/D,MAAM,wBAAwB,GAAG,IAAI,CAAC;AAEtC,yEAAyE;AACzE,oEAAoE;AACpE,MAAM,kBAAkB,GAAG,CAAC,GAAG,EAAE;IAC/B,MAAM,GAAG,GAAG,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,uBAAuB,CAAC,CAAC;IACxD,OAAO,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,GAAG,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC;AACnE,CAAC,CAAC,EAAE,CAAC;AAEL,KAAK,UAAU,eAAe,CAAC,GAAW,EAAE,MAAc;IACxD,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,MAAM,YAAY,CAC/B,GAAG,EACH,CAAC,uBAAuB,EAAE,uBAAuB,CAAC,EAClD,CAAC,EACD,WAAW,CAAC,OAAO,CAAC,wBAAwB,CAAC,CAC9C,CAAC;QACF,MAAM,QAAQ,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,KAAK,IAAI,CAAC,CAAC,IAAI,kBAAkB,CAAC,CAAC;QAC5E,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,EAAE,CAAC;QACrC,MAAM,MAAM,GAAG,QAAQ,CAAC,KAAK,EAAE,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,KAAK,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,KAAK,IAAI,CAAC,CAAC,CAAC,CAAC;QAChF,MAAM,KAAK,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE;YAC7B,MAAM,OAAO,GAAG,CAAC,CAAC,OAAuD,CAAC;YAC1E,MAAM,IAAI,GAAG,OAAO,EAAE,IAAI,IAAI,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC,CAAE,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAA2B,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;YAClH,OAAO,KAAK,MAAM,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC;QACjD,CAAC,CAAC,CAAC;QACH,OAAO,gBAAgB,CAAC,2CAA2C,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,EAAE,MAAM,CAAC,CAAC;IACjG,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,CAAC;IACZ,CAAC;AACH,CAAC;AAED,KAAK,UAAU,cAAc,CAAC,GAAW,EAAE,MAAc;IACvD,IAAI,CAAC;QACH,MAAM,aAAa,GAAG,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,eAAe,EAAE,YAAY,CAAC,CAAC;QACpE,MAAM,KAAK,GAAG,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,aAAa,EAAE,MAAM,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QAChE,IAAI,CAAC,KAAK;YAAE,OAAO,EAAE,CAAC;QACtB,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,eAAe,EAAE,MAAM,EAAE,KAAK,EAAE,UAAU,CAAC,CAAC;QAC7E,MAAM,KAAK,GAAG,MAAM,EAAE,CAAC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC;QACnD,MAAM,SAAS,GAAG,KAAK,CAAC,KAAK,CAAC,gBAAgB,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,IAAI,SAAS,CAAC;QAC1E,MAAM,UAAU,GAAG,KAAK,CAAC,KAAK,CAAC,iBAAiB,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,IAAI,SAAS,CAAC;QAC5E,OAAO,gBAAgB,CAAC,gBAAgB,SAAS,YAAY,UAAU,SAAS,KAAK,GAAG,EAAE,MAAM,CAAC,CAAC;IACpG,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,CAAC;IACZ,CAAC;AACH,CAAC;AAED,KAAK,UAAU,gBAAgB,CAAC,GAAW,EAAE,MAAc;IACzD,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;QACrC,MAAM,OAAO,GAAG,MAAM,EAAE,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;QACnF,MAAM,OAAO,GAA2C,EAAE,CAAC;QAC3D,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;YACxB,IAAI,CAAC,CAAC,CAAC,MAAM,EAAE;gBAAE,SAAS;YAC1B,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC;gBAAE,SAAS;YAClE,IAAI,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,QAAQ,CAAC;gBAAE,SAAS;YACrE,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU,IAAK,CAAiC,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC;YACxF,MAAM,IAAI,GAAG,MAAM,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACjC,OAAO,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,IAAI,CAAC,QAAQ,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC,EAAE,KAAK,EAAE,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC;QAC/F,CAAC;QACD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC;QAC1C,MAAM,GAAG,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;QACpD,IAAI,GAAG,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,EAAE,CAAC;QAChC,OAAO,gBAAgB,CAAC,kBAAkB,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;IACvE,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,CAAC;IACZ,CAAC;AACH,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,aAAa,CAAC,GAAoB;IACtD,4EAA4E;IAC5E,4EAA4E;IAC5E,yEAAyE;IACzE,IAAI,GAAG,CAAC,UAAU,KAAK,UAAU,EAAE,CAAC;QAClC,OAAO;YACL,GAAG,GAAG;YACN,MAAM,EAAE,CAAC,GAAG,GAAG,CAAC,MAAM,EAAE,EAAE,IAAI,EAAE,oBAAoB,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,eAAe,EAAE,CAAC;SAChG,CAAC;IACJ,CAAC;IAED,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,EAAE,CAAC;IAC1B,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,MAAM,aAAa,GAAa,EAAE,CAAC;IAEnC,0EAA0E;IAC1E,qEAAqE;IACrE,IAAI,CAAC,GAAG,CAAC,UAAU,EAAE,CAAC;QACpB,MAAM,gBAAgB,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,WAAW,GAAG,IAAI,CAAC,CAAC;QAC5D,MAAM,UAAU,GAAG,MAAM,eAAe,CAAC,GAAG,CAAC,GAAG,EAAE,gBAAgB,CAAC,CAAC;QACpE,IAAI,UAAU,EAAE,CAAC;YACf,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;YACvB,aAAa,CAAC,IAAI,CAAC,cAAc,UAAU,CAAC,MAAM,IAAI,CAAC,CAAC;QAC1D,CAAC;IACH,CAAC;SAAM,CAAC;QACN,aAAa,CAAC,IAAI,CAAC,+BAA+B,CAAC,CAAC;IACtD,CAAC;IAED,+BAA+B;IAC/B,MAAM,MAAM,GAAG,GAAG,CAAC,YAAY,CAAC;IAChC,IAAI,MAAM,EAAE,IAAI,EAAE,EAAE,CAAC;QACnB,MAAM,OAAO,GAAG,OAAO,GAAG,CAAC,WAAW,KAAK,QAAQ,IAAI,GAAG,CAAC,WAAW,GAAG,kBAAkB,CAAC;QAC5F,MAAM,WAAW,GAAG,OAAO,CAAC,CAAC,CAAC,mCAAmC,CAAC,CAAC,CAAC,EAAE,CAAC;QACvE,MAAM,IAAI,GAAG,2BAA2B,WAAW,GAAG,MAAM,CAAC,IAAI,EAAE,EAAE,CAAC;QACtE,MAAM,OAAO,GAAG,gBAAgB,CAAC,IAAI,EAAE,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,WAAW,GAAG,IAAI,CAAC,CAAC,CAAC;QAC3E,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QACpB,aAAa,CAAC,IAAI,CAAC,UAAU,OAAO,CAAC,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IAC7E,CAAC;IAED,4BAA4B;IAC5B,MAAM,UAAU,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,WAAW,GAAG,IAAI,CAAC,CAAC;IACtD,MAAM,SAAS,GAAG,MAAM,cAAc,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;IACxD,IAAI,SAAS,EAAE,CAAC;QACd,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QACtB,aAAa,CAAC,IAAI,CAAC,QAAQ,SAAS,CAAC,MAAM,IAAI,CAAC,CAAC;IACnD,CAAC;IAED,mCAAmC;IACnC,MAAM,WAAW,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,WAAW,GAAG,IAAI,CAAC,CAAC;IACvD,MAAM,SAAS,GAAG,MAAM,gBAAgB,CAAC,GAAG,EAAE,WAAW,CAAC,CAAC;IAC3D,IAAI,SAAS,EAAE,CAAC;QACd,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QACtB,aAAa,CAAC,IAAI,CAAC,SAAS,SAAS,CAAC,MAAM,IAAI,CAAC,CAAC;IACpD,CAAC;IAED,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACvB,OAAO;YACL,GAAG,GAAG;YACN,MAAM,EAAE,CAAC,GAAG,GAAG,CAAC,MAAM,EAAE,EAAE,IAAI,EAAE,oBAAoB,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,oBAAoB,EAAE,CAAC;SACrG,CAAC;IACJ,CAAC;IAED,OAAO;QACL,GAAG,GAAG;QACN,QAAQ,EAAE,GAAG,GAAG,CAAC,QAAQ,KAAK,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE;QAChD,MAAM,EAAE;YACN,GAAG,GAAG,CAAC,MAAM;YACb;gBACE,IAAI,EAAE,oBAAoB;gBAC1B,OAAO,EAAE,IAAI;gBACb,KAAK,EAAE,aAAa,CAAC,IAAI,CAAC,GAAG,CAAC;aAC/B;SACF;KACF,CAAC;AACJ,CAAC"}

package/dist/src/pil/layer6-output.d.ts

@@ -0,0 +1,20 @@
+/**
+ * src/pil/layer6-output.ts
+ *
+ * Layer 6: Output optimization.
+ * Appends a per-task-type system prompt suffix via applyPilSuffix().
+ * Each suffix is tuned to minimize output tokens while preserving quality.
+ * Conversational turns (taskType=null) pass through unchanged.
+ *
+ * PIL-03: When ctx.outputStyle is null and ctx.taskType is not null:
+ *   a) 50ms brain rescue — classifyViaBrain for multilingual style detection
+ *   b) task-type heuristic — domain-specific default when brain is unavailable
+ * Resolved style is propagated back onto ctx.outputStyle so the orchestrator
+ * picks it up via applyPilSuffix without falling back to the hard "concise" default.
+ */
+import type { ToolSet } from "ai";
+import type { ProviderId } from "../providers/types.js";
+import type { PipelineContext } from "./types.js";
+export declare function applyPilSuffix(systemPrompt: string, ctx: PipelineContext, responseToolsActive?: boolean): string;
+export declare function getResponseToolSet(ctx: PipelineContext, providerId?: ProviderId): ToolSet;
+export declare function layer6Output(ctx: PipelineContext): Promise<PipelineContext>;

package/dist/src/pil/layer6-output.js

@@ -0,0 +1,237 @@
+/**
+ * src/pil/layer6-output.ts
+ *
+ * Layer 6: Output optimization.
+ * Appends a per-task-type system prompt suffix via applyPilSuffix().
+ * Each suffix is tuned to minimize output tokens while preserving quality.
+ * Conversational turns (taskType=null) pass through unchanged.
+ *
+ * PIL-03: When ctx.outputStyle is null and ctx.taskType is not null:
+ *   a) 50ms brain rescue — classifyViaBrain for multilingual style detection
+ *   b) task-type heuristic — domain-specific default when brain is unavailable
+ * Resolved style is propagated back onto ctx.outputStyle so the orchestrator
+ * picks it up via applyPilSuffix without falling back to the hard "concise" default.
+ */
+import { classifyViaBrain } from "../ee/bridge.js";
+import { getProviderCapabilities } from "../providers/capabilities.js";
+import { buildResponseTools } from "./response-tools.js";
+const VALID_STYLES = ["concise", "balanced", "detailed"];
+// Per-task-type fallback style when brain is unavailable.
+// Reflects the information density each task type genuinely needs.
+const TASK_TYPE_DEFAULT_STYLE = {
+    debug: "balanced", // needs root cause + fix context
+    plan: "balanced", // balanced by default; user requests detail explicitly
+    analyze: "balanced", // findings need brief evidence
+    documentation: "balanced", // examples + explanation
+    generate: "concise", // code speaks for itself
+    refactor: "concise", // diff is the output
+    general: "balanced",
+};
+// PIL-04 Tier 1.1: response-tool gating.
+//
+// Structural enforcement (JSON schema via tool call) is the root-cause fix for
+// cheap-model verbosity: the model CANNOT add preamble/epilogue/padding because
+// there is no surface outside the schema to write into. Text-based OUTPUT RULES
+// rely on instruction compliance which budget models (DeepSeek/Qwen/Llama)
+// frequently ignore.
+//
+// Activation gate: enable when the schema's payload is bounded enough that JSON
+// escaping (\n → \\n, " → \") doesn't outweigh the eliminated padding cost.
+//   - analyze: list of findings — repeated keys, no large strings → big win
+//   - plan:    numbered steps    — same shape → big win
+//   - debug:   {hypothesis, root_cause, fix:{file,diff}, verify} — diff is
+//              normally one small hunk, escaping overhead <5%; structural
+//              enforcement of the hypothesis→root_cause→fix→verify format is
+//              exactly what cheap models drift away from in prose mode
+//   - general: {response, reasoning} — pure text, zero structural overhead;
+//              this task type has the weakest text suffix today and suffers the
+//              most from verbose padding
+//
+// Disabled at task-type level (escaping cost dominates):
+//   - generate (large file contents in JSON strings)
+//   - refactor (multi-file diffs)
+//   - documentation (large markdown blocks)
+// For these, the orchestrator falls back to the markdown OUTPUT RULES suffix.
+// Per-PROVIDER disabling is layered on top via ProviderCapabilities — see
+// `src/providers/capabilities.ts`. A task type listed here may still be
+// dropped at runtime if the active provider's capability returns false.
+const RESPONSE_TOOL_TASK_TYPES = new Set(["analyze", "plan", "debug", "general"]);
+// PIL-04 Tier 1.2: per-task output token budget.
+// Hint to model. Empirically derived from interaction_logs avg output sizes;
+// ~20% below mean to nudge brevity without truncating real answers.
+const TASK_OUTPUT_BUDGET = {
+    refactor: 800,
+    debug: 500,
+    plan: 700,
+    analyze: 600,
+    documentation: 900,
+    generate: 1200,
+    general: 200,
+};
+// PIL-04 Tier 1.3: ban preamble openers.
+// Measured ~30 tokens/turn wasted on "I'll help you...", "Sure, let me..." etc.
+// Bilingual EN+VN. Skipped for response-tools path (JSON has no preamble surface).
+const NO_PREAMBLE_RULE = `\nFORBIDDEN OPENERS: do not start with "I'll", "I will", "Let me", "Here's", "Sure", "Of course", "Tôi sẽ", "Để tôi", "Vâng". Start directly with the answer content.`;
+const SUFFIXES = {
+    refactor: {
+        concise: `\nOUTPUT RULES (refactor): Show only changed code. Prefer unified diff or replacement function. No prose unless architecture changes. One sentence max if explanation needed. No preamble.`,
+        balanced: `\nOUTPUT RULES (refactor): Show changed code with brief rationale. Unified diff preferred. Short explanation allowed when architecture changes. Keep prose under 3 sentences.`,
+        detailed: `\nOUTPUT RULES (refactor): Show changed code with full rationale. Explain why each change improves the code. Include before/after comparison when helpful. Unified diff preferred.`,
+    },
+    debug: {
+        concise: `\nOUTPUT RULES (debug): Format = Hypothesis → Root cause (1 line) → Fix (code only) → Verify command. No preamble. No "I think" hedging.`,
+        balanced: `\nOUTPUT RULES (debug): Format = Hypothesis → Root cause → Fix (code) → Verify command. Brief explanation of why the bug occurs. Keep prose minimal.`,
+        detailed: `\nOUTPUT RULES (debug): Format = Hypothesis → Root cause analysis → Fix (code) → Verify command → Prevention. Explain the underlying mechanism and why this fix is correct.`,
+    },
+    plan: {
+        concise: `\nOUTPUT RULES (plan): Numbered steps only. Each step: action verb + acceptance criterion. No prose paragraphs. Add "Assumptions:" section only if needed.`,
+        balanced: `\nOUTPUT RULES (plan): Numbered steps with brief rationale per step. Each step: action verb + acceptance criterion + why. Add "Assumptions:" and "Risks:" sections if applicable.`,
+        detailed: `\nOUTPUT RULES (plan): Numbered steps with full rationale. Each step: action verb + acceptance criterion + why + alternatives considered. Include "Assumptions:", "Risks:", and "Trade-offs:" sections.`,
+    },
+    analyze: {
+        concise: `\nOUTPUT RULES (analyze): Bullet findings with evidence (file:line or direct quote). Add severity label (High/Med/Low) when applicable. No filler sentences.`,
+        balanced: `\nOUTPUT RULES (analyze): Bullet findings with evidence (file:line or direct quote). Add severity label and brief explanation. Context for each finding.`,
+        detailed: `\nOUTPUT RULES (analyze): Bullet findings with evidence (file:line or direct quote). Add severity label, root cause analysis, and recommended action. Provide context and impact assessment.`,
+    },
+    documentation: {
+        concise: `\nOUTPUT RULES (documentation): Markdown only. Lead with a code example, then explanation. No "This function..." openers. All examples in fenced code blocks.`,
+        balanced: `\nOUTPUT RULES (documentation): Markdown only. Lead with a code example, then explanation with context. Cover common use cases. All examples in fenced code blocks.`,
+        detailed: `\nOUTPUT RULES (documentation): Markdown only. Lead with a code example, then thorough explanation. Cover use cases, edge cases, and gotchas. All examples in fenced code blocks. Include parameter descriptions and return value explanations.`,
+    },
+    generate: {
+        concise: `\nOUTPUT RULES (generate): Complete, runnable code only. Include all imports. Brief inline comments for non-obvious logic only. No prose outside code blocks. No partial snippets.`,
+        balanced: `\nOUTPUT RULES (generate): Complete, runnable code with brief explanation. Include all imports. Inline comments for key decisions. Short prose before code block explaining approach.`,
+        detailed: `\nOUTPUT RULES (generate): Complete, runnable code with full explanation. Include all imports. Inline comments for logic and decisions. Explain design choices, alternatives considered, and trade-offs before the code.`,
+    },
+    general: {
+        concise: `\nAnswer directly. No preamble.`,
+        balanced: `\nAnswer with brief context.`,
+        detailed: `\nAnswer thoroughly.`,
+    },
+};
+// TODO(WhoAmI-L6): when EE v4.0 Who Am I profile is available, skip
+// NO_PREAMBLE_RULE for users with feedback_style="explicit" who prefer
+// preamble. Also source TASK_TYPE_DEFAULT_STYLE from profile communication.brevity
+// instead of the hardcoded heuristic map above.
+export function applyPilSuffix(systemPrompt, ctx, responseToolsActive = false) {
+    // Chitchat: layer6Output already skipped suffix work; mirror that here so
+    // direct callers (e.g. orchestrator) don't accidentally re-inject rules.
+    if (ctx.intentKind === "chitchat")
+        return systemPrompt;
+    if (!ctx.taskType || !SUFFIXES[ctx.taskType])
+        return systemPrompt;
+    if (responseToolsActive) {
+        return (systemPrompt +
+            `\nOUTPUT FORMAT: When you finish your work, use the respond_${ctx.taskType} tool to structure your final answer. You may write free-form text to explain your reasoning during the process. Use action tools (bash, read_file, edit_file, etc.) as needed, then summarize your result via respond_${ctx.taskType}.`);
+    }
+    const style = ctx.outputStyle ?? "concise";
+    const baseSuffix = SUFFIXES[ctx.taskType][style];
+    // PIL-04 Tier 1.2: output-budget hint.
+    const budget = TASK_OUTPUT_BUDGET[ctx.taskType] ?? 600;
+    const budgetHint = `\nOUTPUT BUDGET: aim for ≤${budget} tokens. Stop when the answer is complete; do not pad.`;
+    // PIL-04 Tier 1.3: ban preamble (~30 tokens saved/turn).
+    let result = systemPrompt + baseSuffix + budgetHint + NO_PREAMBLE_RULE;
+    // T1 behavioral rules (proven-tier EE points set by Layer 3). These are
+    // project-specific reflexes the model MUST follow — injected as instructions,
+    // not as context hints, so they carry imperative weight rather than suggestion weight.
+    if (ctx.t1Rules && ctx.t1Rules.length > 0) {
+        const mandatoryLines = ctx.t1Rules.map((r) => `- ${r}`).join("\n");
+        result += `\nMANDATORY RULES (from experience — must follow):\n${mandatoryLines}`;
+    }
+    return result;
+}
+export function getResponseToolSet(ctx, providerId) {
+    if (!ctx.taskType)
+        return {};
+    // PIL-04 Tier 1.1: gate JSON-structured output to list-shaped tasks where it
+    // wins on tokens. Code-heavy tasks fall through to markdown OUTPUT RULES.
+    if (!RESPONSE_TOOL_TASK_TYPES.has(ctx.taskType))
+        return {};
+    // Provider-aware gating: a provider may report it can't reliably emit
+    // valid JSON tool input for this task type (e.g. DeepSeek leaks special
+    // tokens into `general` responses). Drop the tool to avoid retry storms.
+    if (providerId) {
+        const caps = getProviderCapabilities(providerId);
+        if (!caps.supportsResponseTool(ctx.taskType))
+            return {};
+    }
+    return buildResponseTools(ctx.taskType);
+}
+export async function layer6Output(ctx) {
+    try {
+        // Chitchat short-circuit: greetings/small-talk don't need OUTPUT RULES
+        // suffixes — the model already produces a short reply for "hi" without
+        // being told to. Aligns with L4/L5 chitchat skips so the fast-path stays
+        // clean end-to-end (no GSD, no context, no suffix, no MCP).
+        if (ctx.intentKind === "chitchat") {
+            return {
+                ...ctx,
+                layers: [...ctx.layers, { name: "output-optimization", applied: false, delta: "skip:chitchat" }],
+            };
+        }
+        if (ctx.taskType === null) {
+            return {
+                ...ctx,
+                layers: [...ctx.layers, { name: "output-optimization", applied: false, delta: null }],
+            };
+        }
+        // PIL-03: rescue outputStyle when L1 couldn't determine it (brain timeout / EE disabled).
+        let outputStyle = ctx.outputStyle;
+        let styleSource = "inherited";
+        if (outputStyle === null) {
+            // Pass a: 50ms brain rescue — catches multilingual prompts that regex/L1 missed.
+            // Skip when ctx._brainData is already populated: L1's unified brain call had its
+            // chance to set style; a second classifyViaBrain here is redundant work.
+            if (!ctx._brainData) {
+                try {
+                    const brainRaw = await classifyViaBrain(`Task type: ${ctx.taskType}. Reply ONE word only: concise | balanced | detailed\nPrompt: "${ctx.raw.slice(0, 150)}"`, 50);
+                    if (brainRaw) {
+                        const matched = VALID_STYLES.find((s) => brainRaw.toLowerCase().includes(s));
+                        if (matched) {
+                            outputStyle = matched;
+                            styleSource = "brain-rescue";
+                        }
+                    }
+                }
+                catch {
+                    // fall through to pass b
+                }
+            }
+            // Pass b: task-type heuristic — fires when brain is also unavailable
+            if (outputStyle === null) {
+                outputStyle = TASK_TYPE_DEFAULT_STYLE[ctx.taskType] ?? "concise";
+                styleSource = "task-heuristic";
+            }
+        }
+        const style = outputStyle;
+        const taskKey = ctx.taskType;
+        const suffixEntry = SUFFIXES[taskKey];
+        if (!suffixEntry) {
+            return {
+                ...ctx,
+                outputStyle,
+                layers: [...ctx.layers, { name: "output-optimization", applied: false, delta: "no-suffix-entry" }],
+            };
+        }
+        const suffix = suffixEntry[style];
+        return {
+            ...ctx,
+            outputStyle, // propagate resolved style to orchestrator's applyPilSuffix
+            layers: [
+                ...ctx.layers,
+                {
+                    name: "output-optimization",
+                    applied: true,
+                    delta: `suffix=${ctx.taskType},style=${style},src=${styleSource},chars=${suffix.trim().length}`,
+                },
+            ],
+        };
+    }
+    catch {
+        return {
+            ...ctx,
+            layers: [...ctx.layers, { name: "output-optimization", applied: false, delta: null }],
+        };
+    }
+}
+//# sourceMappingURL=layer6-output.js.map

package/dist/src/pil/layer6-output.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"layer6-output.js","sourceRoot":"","sources":["../../../src/pil/layer6-output.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAGH,OAAO,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AACnD,OAAO,EAAE,uBAAuB,EAAE,MAAM,8BAA8B,CAAC;AAEvE,OAAO,EAAE,kBAAkB,EAAE,MAAM,qBAAqB,CAAC;AAGzD,MAAM,YAAY,GAAG,CAAC,SAAS,EAAE,UAAU,EAAE,UAAU,CAAU,CAAC;AAElE,0DAA0D;AAC1D,mEAAmE;AACnE,MAAM,uBAAuB,GAAkC;IAC7D,KAAK,EAAE,UAAU,EAAE,iCAAiC;IACpD,IAAI,EAAE,UAAU,EAAE,uDAAuD;IACzE,OAAO,EAAE,UAAU,EAAE,+BAA+B;IACpD,aAAa,EAAE,UAAU,EAAE,yBAAyB;IACpD,QAAQ,EAAE,SAAS,EAAE,yBAAyB;IAC9C,QAAQ,EAAE,SAAS,EAAE,qBAAqB;IAC1C,OAAO,EAAE,UAAU;CACpB,CAAC;AAEF,yCAAyC;AACzC,EAAE;AACF,+EAA+E;AAC/E,gFAAgF;AAChF,gFAAgF;AAChF,2EAA2E;AAC3E,qBAAqB;AACrB,EAAE;AACF,gFAAgF;AAChF,4EAA4E;AAC5E,4EAA4E;AAC5E,wDAAwD;AACxD,2EAA2E;AAC3E,0EAA0E;AAC1E,6EAA6E;AAC7E,uEAAuE;AACvE,4EAA4E;AAC5E,gFAAgF;AAChF,yCAAyC;AACzC,EAAE;AACF,yDAAyD;AACzD,qDAAqD;AACrD,kCAAkC;AAClC,4CAA4C;AAC5C,8EAA8E;AAC9E,0EAA0E;AAC1E,wEAAwE;AACxE,wEAAwE;AACxE,MAAM,wBAAwB,GAAG,IAAI,GAAG,CAAW,CAAC,SAAS,EAAE,MAAM,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC;AAE5F,iDAAiD;AACjD,6EAA6E;AAC7E,oEAAoE;AACpE,MAAM,kBAAkB,GAA6B;IACnD,QAAQ,EAAE,GAAG;IACb,KAAK,EAAE,GAAG;IACV,IAAI,EAAE,GAAG;IACT,OAAO,EAAE,GAAG;IACZ,aAAa,EAAE,GAAG;IAClB,QAAQ,EAAE,IAAI;IACd,OAAO,EAAE,GAAG;CACb,CAAC;AAEF,yCAAyC;AACzC,gFAAgF;AAChF,mFAAmF;AACnF,MAAM,gBAAgB,GAAG,uKAAuK,CAAC;AAEjM,MAAM,QAAQ,GAAgD;IAC5D,QAAQ,EAAE;QACR,OAAO,EAAE,4LAA4L;QACrM,QAAQ,EAAE,+KAA+K;QACzL,QAAQ,EAAE,oLAAoL;KAC/L;IACD,KAAK,EAAE;QACL,OAAO,EAAE,0IAA0I;QACnJ,QAAQ,EAAE,sJAAsJ;QAChK,QAAQ,EAAE,6KAA6K;KACxL;IACD,IAAI,EAAE;QACJ,OAAO,EAAE,4JAA4J;QACrK,QAAQ,EAAE,mLAAmL;QAC7L,QAAQ,EAAE,yMAAyM;KACpN;IACD,OAAO,EAAE;QACP,OAAO,EAAE,8JAA8J;QACvK,QAAQ,EAAE,0JAA0J;QACpK,QAAQ,EAAE,8LAA8L;KACzM;IACD,aAAa,EAAE;QACb,OAAO,EAAE,+JAA+J;QACxK,QAAQ,EAAE,qKAAqK;QAC/K,QAAQ,EAAE,iPAAiP;KAC5P;IACD,QAAQ,EAAE;QACR,OAAO,EAAE,oLAAoL;QAC7L,QAAQ,EAAE,uLAAuL;QACjM,QAAQ,EAAE,0NAA0N;KACrO;IACD,OAAO,EAAE;QACP,OAAO,EAAE,iCAAiC;QAC1C,QAAQ,EAAE,8BAA8B;QACxC,QAAQ,EAAE,sBAAsB;KACjC;CACF,CAAC;AAEF,oEAAoE;AACpE,uEAAuE;AACvE,mFAAmF;AACnF,gDAAgD;AAEhD,MAAM,UAAU,cAAc,CAAC,YAAoB,EAAE,GAAoB,EAAE,mBAAmB,GAAG,KAAK;IACpG,0EAA0E;IAC1E,yEAAyE;IACzE,IAAI,GAAG,CAAC,UAAU,KAAK,UAAU;QAAE,OAAO,YAAY,CAAC;IACvD,IAAI,CAAC,GAAG,CAAC,QAAQ,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,QAAQ,CAAC;QAAE,OAAO,YAAY,CAAC;IAElE,IAAI,mBAAmB,EAAE,CAAC;QACxB,OAAO,CACL,YAAY;YACZ,+DAA+D,GAAG,CAAC,QAAQ,0NAA0N,GAAG,CAAC,QAAQ,GAAG,CACrT,CAAC;IACJ,CAAC;IAED,MAAM,KAAK,GAAgB,GAAG,CAAC,WAAW,IAAI,SAAS,CAAC;IACxD,MAAM,UAAU,GAAG,QAAQ,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,KAAK,CAAC,CAAC;IAEjD,uCAAuC;IACvC,MAAM,MAAM,GAAG,kBAAkB,CAAC,GAAG,CAAC,QAAoB,CAAC,IAAI,GAAG,CAAC;IACnE,MAAM,UAAU,GAAG,6BAA6B,MAAM,wDAAwD,CAAC;IAE/G,yDAAyD;IACzD,IAAI,MAAM,GAAG,YAAY,GAAG,UAAU,GAAG,UAAU,GAAG,gBAAgB,CAAC;IAEvE,wEAAwE;IACxE,8EAA8E;IAC9E,uFAAuF;IACvF,IAAI,GAAG,CAAC,OAAO,IAAI,GAAG,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC1C,MAAM,cAAc,GAAG,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACnE,MAAM,IAAI,uDAAuD,cAAc,EAAE,CAAC;IACpF,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,GAAoB,EAAE,UAAuB;IAC9E,IAAI,CAAC,GAAG,CAAC,QAAQ;QAAE,OAAO,EAAE,CAAC;IAC7B,6EAA6E;IAC7E,0EAA0E;IAC1E,IAAI,CAAC,wBAAwB,CAAC,GAAG,CAAC,GAAG,CAAC,QAAQ,CAAC;QAAE,OAAO,EAAE,CAAC;IAC3D,sEAAsE;IACtE,wEAAwE;IACxE,yEAAyE;IACzE,IAAI,UAAU,EAAE,CAAC;QACf,MAAM,IAAI,GAAG,uBAAuB,CAAC,UAAU,CAAC,CAAC;QACjD,IAAI,CAAC,IAAI,CAAC,oBAAoB,CAAC,GAAG,CAAC,QAAQ,CAAC;YAAE,OAAO,EAAE,CAAC;IAC1D,CAAC;IACD,OAAO,kBAAkB,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;AAC1C,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,YAAY,CAAC,GAAoB;IACrD,IAAI,CAAC;QACH,uEAAuE;QACvE,uEAAuE;QACvE,yEAAyE;QACzE,4DAA4D;QAC5D,IAAI,GAAG,CAAC,UAAU,KAAK,UAAU,EAAE,CAAC;YAClC,OAAO;gBACL,GAAG,GAAG;gBACN,MAAM,EAAE,CAAC,GAAG,GAAG,CAAC,MAAM,EAAE,EAAE,IAAI,EAAE,qBAAqB,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,eAAe,EAAE,CAAC;aACjG,CAAC;QACJ,CAAC;QAED,IAAI,GAAG,CAAC,QAAQ,KAAK,IAAI,EAAE,CAAC;YAC1B,OAAO;gBACL,GAAG,GAAG;gBACN,MAAM,EAAE,CAAC,GAAG,GAAG,CAAC,MAAM,EAAE,EAAE,IAAI,EAAE,qBAAqB,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC;aACtF,CAAC;QACJ,CAAC;QAED,0FAA0F;QAC1F,IAAI,WAAW,GAAG,GAAG,CAAC,WAAW,CAAC;QAClC,IAAI,WAAW,GAAG,WAAW,CAAC;QAE9B,IAAI,WAAW,KAAK,IAAI,EAAE,CAAC;YACzB,iFAAiF;YACjF,iFAAiF;YACjF,yEAAyE;YACzE,IAAI,CAAC,GAAG,CAAC,UAAU,EAAE,CAAC;gBACpB,IAAI,CAAC;oBACH,MAAM,QAAQ,GAAG,MAAM,gBAAgB,CACrC,cAAc,GAAG,CAAC,QAAQ,kEAAkE,GAAG,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,GAAG,EACpH,EAAE,CACH,CAAC;oBACF,IAAI,QAAQ,EAAE,CAAC;wBACb,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;wBAC7E,IAAI,OAAO,EAAE,CAAC;4BACZ,WAAW,GAAG,OAAO,CAAC;4BACtB,WAAW,GAAG,cAAc,CAAC;wBAC/B,CAAC;oBACH,CAAC;gBACH,CAAC;gBAAC,MAAM,CAAC;oBACP,yBAAyB;gBAC3B,CAAC;YACH,CAAC;YAED,qEAAqE;YACrE,IAAI,WAAW,KAAK,IAAI,EAAE,CAAC;gBACzB,WAAW,GAAG,uBAAuB,CAAC,GAAG,CAAC,QAAoB,CAAC,IAAI,SAAS,CAAC;gBAC7E,WAAW,GAAG,gBAAgB,CAAC;YACjC,CAAC;QACH,CAAC;QAED,MAAM,KAAK,GAAgB,WAAW,CAAC;QACvC,MAAM,OAAO,GAAG,GAAG,CAAC,QAAkB,CAAC;QACvC,MAAM,WAAW,GAAG,QAAQ,CAAC,OAAO,CAAC,CAAC;QACtC,IAAI,CAAC,WAAW,EAAE,CAAC;YACjB,OAAO;gBACL,GAAG,GAAG;gBACN,WAAW;gBACX,MAAM,EAAE,CAAC,GAAG,GAAG,CAAC,MAAM,EAAE,EAAE,IAAI,EAAE,qBAAqB,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,iBAAiB,EAAE,CAAC;aACnG,CAAC;QACJ,CAAC;QAED,MAAM,MAAM,GAAG,WAAW,CAAC,KAAK,CAAC,CAAC;QAClC,OAAO;YACL,GAAG,GAAG;YACN,WAAW,EAAE,4DAA4D;YACzE,MAAM,EAAE;gBACN,GAAG,GAAG,CAAC,MAAM;gBACb;oBACE,IAAI,EAAE,qBAAqB;oBAC3B,OAAO,EAAE,IAAI;oBACb,KAAK,EAAE,UAAU,GAAG,CAAC,QAAQ,UAAU,KAAK,QAAQ,WAAW,UAAU,MAAM,CAAC,IAAI,EAAE,CAAC,MAAM,EAAE;iBAChG;aACF;SACF,CAAC;IACJ,CAAC;IAAC,MAAM,CAAC;QACP,OAAO;YACL,GAAG,GAAG;YACN,MAAM,EAAE,CAAC,GAAG,GAAG,CAAC,MAAM,EAAE,EAAE,IAAI,EAAE,qBAAqB,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC;SACtF,CAAC;IACJ,CAAC;AACH,CAAC"}

package/dist/src/pil/ollama-classify.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"ollama-classify.js","sourceRoot":"","sources":["../../../src/pil/ollama-classify.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AASH,MAAM,iBAAiB,GAAG,GAAG,CAAC;AAC9B,MAAM,eAAe,GAAG,wBAAwB,CAAC;AACjD,MAAM,YAAY,GAAG,WAAW,CAAC;AAEjC,MAAM,gBAAgB,GAAe,CAAC,UAAU,EAAE,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,eAAe,EAAE,UAAU,CAAC,CAAC;AAE3G,MAAM,CAAC,KAAK,UAAU,cAAc,CAAC,MAAc;IACjD,IAAI,CAAC;QACH,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;QACzC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,iBAAiB,CAAC,CAAC;QAEtE,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,eAAe,eAAe,EAAE;YAC9D,MAAM,EAAE,MAAM;YACd,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE;YAC/C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC;gBACnB,KAAK,EAAE,YAAY;gBACnB,MAAM,EAAE,oKAAoK,MAAM,GAAG;gBACrL,MAAM,EAAE,KAAK;aACd,CAAC;YACF,MAAM,EAAE,UAAU,CAAC,MAAM;SAC1B,CAAC,CAAC;QAEH,YAAY,CAAC,KAAK,CAAC,CAAC;QACpB,IAAI,CAAC,QAAQ,CAAC,EAAE;YAAE,OAAO,IAAI,CAAC;QAE9B,MAAM,IAAI,GAAG,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,CAA0B,CAAC;QAC9D,MAAM,GAAG,GAAG,CAAC,IAAI,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;QACvD,MAAM,OAAO,GAAG,gBAAgB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;QAE9D,OAAO,OAAO,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,OAAO,EAAE,UAAU,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;IAClE,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC"}

package/dist/src/pil/pipeline.d.ts

@@ -0,0 +1,23 @@
+/**
+ * src/pil/pipeline.ts
+ *
+ * runPipeline() entry point: orchestrates 6 sequential layers with an
+ * adaptive timeout. Fail-open: any unhandled error or timeout returns the
+ * original fallback context.
+ *
+ * Timeout budget:
+ *   - 200ms when EE is disabled / not reachable (fast regex-only path).
+ *   - 3000ms when EE thin/thin-degraded mode is active so Layer 1 can hit
+ *     the remote `/api/brain` endpoint and Layer 6 can refine output style.
+ *
+ * CRITICAL: fallback is captured BEFORE runLayers() starts to ensure the timeout
+ * path returns a pristine context (Pitfall 4 from RESEARCH.md).
+ */
+import type { PipelineContext } from "./types.js";
+export interface PipelineOptions {
+    gsdPhase?: string | null;
+    resumeDigest?: string | null;
+    activeRunId?: string | null;
+    sessionId?: string | null;
+}
+export declare function runPipeline(raw: string, options?: PipelineOptions): Promise<PipelineContext>;

package/dist/src/pil/pipeline.js

@@ -0,0 +1,182 @@
+/**
+ * src/pil/pipeline.ts
+ *
+ * runPipeline() entry point: orchestrates 6 sequential layers with an
+ * adaptive timeout. Fail-open: any unhandled error or timeout returns the
+ * original fallback context.
+ *
+ * Timeout budget:
+ *   - 200ms when EE is disabled / not reachable (fast regex-only path).
+ *   - 3000ms when EE thin/thin-degraded mode is active so Layer 1 can hit
+ *     the remote `/api/brain` endpoint and Layer 6 can refine output style.
+ *
+ * CRITICAL: fallback is captured BEFORE runLayers() starts to ensure the timeout
+ * path returns a pristine context (Pitfall 4 from RESEARCH.md).
+ */
+import { getCachedServerBaseUrl } from "../ee/auth.js";
+import { getCachedEEClientMode } from "../ee/client-mode.js";
+import { classifyEeError, logEeFailure } from "../utils/ee-logger.js";
+import { DEFAULT_TOKEN_BUDGET } from "./budget.js";
+import { appendPilLog } from "./budget-log.js";
+import { layer1Intent } from "./layer1-intent.js";
+import { layer2Personality } from "./layer2-personality.js";
+import { layer3EeInjection } from "./layer3-ee-injection.js";
+import { layer4Gsd } from "./layer4-gsd.js";
+import { layer5Context } from "./layer5-context.js";
+import { layer6Output } from "./layer6-output.js";
+import { PipelineContextSchema } from "./schema.js";
+import { setPilLastResult } from "./store.js";
+import { resolveAfter } from "./timeout.js";
+const PIPELINE_TIMEOUT_FAST_MS = 200;
+// Sized from measured /api/pil-context distribution after server-side
+// classify+embed parallelization (commit 5b77bab in experience-engine).
+// 120-call sample: p50=1155ms, p95=2171ms, p99=2734ms, max=3105ms. 3500ms
+// gives ~800ms margin over p99. Drop further only after another infra win.
+const PIPELINE_TIMEOUT_BRAIN_MS = 3500;
+function pipelineTimeoutMs() {
+    // Allow test environments to override the timeout to avoid flaky races when
+    // the test process is under load (e.g., running 1600+ tests concurrently).
+    const envOverride = process.env.MUONROI_TEST_PIPELINE_TIMEOUT_MS;
+    if (envOverride) {
+        const parsed = parseInt(envOverride, 10);
+        if (!isNaN(parsed) && parsed > 0)
+            return parsed;
+    }
+    const mode = getCachedEEClientMode();
+    if (mode && (mode.mode === "thin" || mode.mode === "thin-degraded" || mode.mode === "fat")) {
+        return PIPELINE_TIMEOUT_BRAIN_MS;
+    }
+    // Boot race fallback: detectEEClientMode() is fire-and-forget at startup
+    // (src/index.ts) and its /health probe can take up to 1500ms. Any prompt
+    // submitted before the probe resolves saw mode=null and fell through to the
+    // 200ms FAST budget, killing the brain call. loadEEAuthToken() IS awaited
+    // at boot, so a configured serverBaseUrl is a reliable signal that the user
+    // intends thin mode — use BRAIN budget optimistically. If the probe later
+    // fails, mode becomes "thin-degraded" which the branch above also covers.
+    if (getCachedServerBaseUrl()) {
+        return PIPELINE_TIMEOUT_BRAIN_MS;
+    }
+    return PIPELINE_TIMEOUT_FAST_MS;
+}
+const SKIPPED_LAYERS = [
+    { timingName: "layer2-personality", deltaName: "personality-adaptation" },
+    { timingName: "layer3-ee-injection", deltaName: "ee-experience-injection" },
+    { timingName: "layer4-gsd-structuring", deltaName: "gsd-workflow-structuring" },
+    { timingName: "layer5-context-enrichment", deltaName: "context-enrichment" },
+];
+async function runLayers(ctx) {
+    const pipelineStart = Date.now();
+    const timings = [];
+    // Track each layer's contribution to the enriched-prompt size so the
+    // PIL budget log can attribute system-prompt bloat to a specific layer.
+    const layerSnapshots = [];
+    async function timed(name, fn) {
+        const start = Date.now();
+        const charsBefore = ctx.enriched.length;
+        ctx = await fn(ctx);
+        const ms = Date.now() - start;
+        const charsAfter = ctx.enriched.length;
+        timings.push({ name, ms });
+        layerSnapshots.push({ name, charsBefore, charsAfter, charsDelta: charsAfter - charsBefore, durationMs: ms });
+    }
+    await timed("layer1-intent", layer1Intent);
+    if (ctx.taskType !== null) {
+        await timed("layer2-personality", layer2Personality);
+        await timed("layer3-ee-injection", layer3EeInjection);
+        await timed("layer4-gsd-structuring", layer4Gsd);
+        await timed("layer5-context-enrichment", layer5Context);
+    }
+    else {
+        for (const { timingName } of SKIPPED_LAYERS) {
+            timings.push({ name: timingName, ms: 0 });
+        }
+        ctx = {
+            ...ctx,
+            layers: [
+                ...ctx.layers,
+                ...SKIPPED_LAYERS.map(({ deltaName }) => ({
+                    name: deltaName,
+                    applied: false,
+                    delta: "skipped:null-taskType",
+                })),
+            ],
+        };
+    }
+    await timed("layer6-output", layer6Output);
+    const suffixCharsMatch = ctx.layers.find((l) => l.name === "output-optimization")?.delta?.match(/chars=(\d+)/);
+    const suffixChars = suffixCharsMatch ? parseInt(suffixCharsMatch[1], 10) : 0;
+    const enrichmentCharsAdded = Math.max(0, ctx.enriched.length - ctx.raw.length);
+    ctx = {
+        ...ctx,
+        metrics: {
+            totalMs: Date.now() - pipelineStart,
+            layerTimings: timings,
+            inputChars: ctx.raw.length,
+            outputChars: ctx.enriched.length,
+            suffixInstructionTokens: Math.round(suffixChars / 4),
+            enrichmentTokensAdded: Math.round(enrichmentCharsAdded / 4),
+        },
+    };
+    // Best-effort PIL budget log — attributes prompt-size growth to each layer.
+    // Fire-and-forget; never await on the hot path.
+    appendPilLog({
+        ts: pipelineStart,
+        sessionId: ctx.sessionId ?? null,
+        taskType: ctx.taskType ?? null,
+        domain: ctx.domain ?? null,
+        confidence: ctx.confidence ?? 0,
+        rawChars: ctx.raw.length,
+        enrichedChars: ctx.enriched.length,
+        totalDeltaChars: ctx.enriched.length - ctx.raw.length,
+        totalMs: Date.now() - pipelineStart,
+        layers: layerSnapshots,
+        fallbackReason: ctx.fallbackReason ?? null,
+        intentDetection: ctx._intentTrace ?? null,
+    }).catch((err) => {
+        logEeFailure("pil.pipeline.logInteraction", classifyEeError(err), err);
+        return undefined;
+    });
+    return ctx;
+}
+export async function runPipeline(raw, options) {
+    const fallback = {
+        raw,
+        enriched: raw,
+        taskType: null,
+        domain: null,
+        confidence: 0,
+        outputStyle: null,
+        tokenBudget: DEFAULT_TOKEN_BUDGET,
+        metrics: null,
+        layers: [],
+        gsdPhase: options?.gsdPhase ?? null,
+        resumeDigest: options?.resumeDigest ?? null,
+        activeRunId: options?.activeRunId ?? null,
+        sessionId: options?.sessionId ?? null,
+        fallbackReason: null,
+    };
+    try {
+        const result = await Promise.race([
+            runLayers({ ...fallback }),
+            resolveAfter(pipelineTimeoutMs(), { ...fallback, fallbackReason: "pipeline-timeout" }),
+        ]);
+        const parse = PipelineContextSchema.safeParse(result);
+        if (!parse.success) {
+            const validated = {
+                ...fallback,
+                fallbackReason: `schema-reject:${parse.error.issues[0]?.path?.join(".") ?? "unknown"}`,
+            };
+            setPilLastResult(validated);
+            return validated;
+        }
+        setPilLastResult(result);
+        return result;
+    }
+    catch (err) {
+        const reason = err instanceof Error ? `exception:${err.name}` : "exception:unknown";
+        const failed = { ...fallback, fallbackReason: reason };
+        setPilLastResult(failed);
+        return failed;
+    }
+}
+//# sourceMappingURL=pipeline.js.map