@caupulican/pi-adaptative 0.80.86 → 0.80.89

package/dist/core/model-router/route-judge.js

@@ -0,0 +1,128 @@
+import { runBoundedCompletion } from "../autonomy/bounded-completion.js";
+/**
+ * Routing-only judge: a bounded, tool-less completion (default: the medium model) that decides the
+ * final cheap/medium/expensive tier for a user prompt, refining the regex classifier's baseline.
+ * Core rule: planning is never cheap unless the judge explicitly deems the task trivial. The judge
+ * proposes; the existing pipeline (model resolution, auth, escalation, gates) still decides.
+ * Failure is honest: unparseable/timeout/unavailable falls back to the baseline with a visible
+ * reasonCode, never silently.
+ */
+/** Static across calls — callers pass cacheRetention "short" so only the variable tail is billed. */
+export const ROUTE_JUDGE_SYSTEM_PROMPT = [
+    "You are a routing judge for a coding agent. You only route; you never answer the task.",
+    "Pick which model tier should handle the user prompt:",
+    '- "cheap": trivial, mechanical, read-only lookups only.',
+    '- "medium": normal implementation, scoped edits, tests, and NON-trivial planning/design.',
+    '- "expensive": architecture, ambiguity, security/auth, destructive or release operations, high-impact changes.',
+    "Planning, design, and strategy prompts are NEVER cheap unless the task is genuinely trivial.",
+    "Respond with STRICT JSON only - no prose:",
+    '{"tier":"cheap"|"medium"|"expensive","risk":"read-only"|"scoped-write"|"high-impact"|"approval-required","trivial":true|false,"reason":"<short reason>"}',
+].join("\n");
+export const ROUTE_JUDGE_MAX_OUTPUT_TOKENS = 128;
+export const ROUTE_JUDGE_MAX_WALL_CLOCK_MS = 10_000;
+export function buildRouteJudgeUserPrompt(args) {
+    return [
+        `Baseline (regex) verdict: tier=${args.baseline.tier}, risk=${args.baseline.risk}, reason=${args.baseline.reasonCode}.`,
+        "User prompt:",
+        args.prompt.slice(0, 4000),
+    ].join("\n");
+}
+const JUDGE_TIERS = ["cheap", "medium", "expensive"];
+const JUDGE_RISKS = ["read-only", "scoped-write", "high-impact", "approval-required"];
+export function parseRouteJudgeVerdict(text) {
+    const trimmed = text.trim();
+    const candidates = [trimmed];
+    const fenced = /```(?:json)?\s*([\s\S]*?)```/.exec(trimmed);
+    if (fenced?.[1])
+        candidates.push(fenced[1].trim());
+    const start = trimmed.indexOf("{");
+    const end = trimmed.lastIndexOf("}");
+    if (start >= 0 && end > start)
+        candidates.push(trimmed.slice(start, end + 1));
+    for (const candidate of candidates) {
+        let parsed;
+        try {
+            parsed = JSON.parse(candidate);
+        }
+        catch {
+            continue;
+        }
+        if (!parsed || typeof parsed !== "object" || Array.isArray(parsed))
+            continue;
+        const record = parsed;
+        // The judge may never select the learning tier or anything outside the three foreground tiers.
+        if (typeof record.tier !== "string" || !JUDGE_TIERS.includes(record.tier))
+            continue;
+        const risk = typeof record.risk === "string" && JUDGE_RISKS.includes(record.risk)
+            ? record.risk
+            : undefined;
+        if (!risk)
+            continue;
+        return {
+            tier: record.tier,
+            risk,
+            trivial: record.trivial === true,
+            reason: typeof record.reason === "string" ? record.reason.slice(0, 200) : "",
+        };
+    }
+    return undefined;
+}
+/** Merge a judge verdict into the baseline decision (pure; never returns learning). */
+export function applyRouteJudgeVerdict(baseline, verdict) {
+    // Enforce the core rule in code, not just in the judge prompt: downgrading a non-cheap
+    // baseline to cheap requires an EXPLICIT trivial verdict. An untrusted judge saying
+    // {tier:"cheap", trivial:false} for an elevated prompt keeps the baseline tier.
+    const tier = verdict.tier === "cheap" && baseline.tier !== "cheap" && !verdict.trivial ? baseline.tier : verdict.tier;
+    return {
+        ...baseline,
+        tier,
+        risk: verdict.risk,
+        confidence: Math.max(baseline.confidence, 0.75),
+        reasonCode: `judge_${tier}${verdict.trivial ? "_trivial" : ""}`,
+        reasons: [...baseline.reasons, `Route judge: ${verdict.reason || "no reason given"}`],
+    };
+}
+/**
+ * Run the judge over a baseline decision. The completion executor is injected (production:
+ * AgentSession.runIsolatedCompletion on the judge model). Never throws; every failure keeps the
+ * baseline with a visible fallbackReason.
+ */
+export async function runRouteJudge(args) {
+    const bounded = await runBoundedCompletion({
+        maxWallClockMs: args.maxWallClockMs ?? ROUTE_JUDGE_MAX_WALL_CLOCK_MS,
+        signal: args.signal,
+        execute: (signal) => args.complete({
+            systemPrompt: ROUTE_JUDGE_SYSTEM_PROMPT,
+            userPrompt: buildRouteJudgeUserPrompt({ prompt: args.prompt, baseline: args.baseline }),
+            signal,
+        }),
+    });
+    const costUsd = bounded.completion?.costUsd ?? 0;
+    if (bounded.failure || !bounded.completion) {
+        return {
+            decision: {
+                ...args.baseline,
+                reasons: [...args.baseline.reasons, "Route judge unavailable; baseline kept"],
+            },
+            fallbackReason: bounded.failure ? `judge_${bounded.failure.reasonCode}` : "judge_unavailable_fallback",
+            costUsd,
+        };
+    }
+    if (bounded.completion.stopReason === "error" || bounded.completion.stopReason === "aborted") {
+        return {
+            decision: { ...args.baseline, reasons: [...args.baseline.reasons, "Route judge errored; baseline kept"] },
+            fallbackReason: "judge_model_error",
+            costUsd,
+        };
+    }
+    const verdict = parseRouteJudgeVerdict(bounded.completion.text);
+    if (!verdict) {
+        return {
+            decision: { ...args.baseline, reasons: [...args.baseline.reasons, "Route judge unparseable; baseline kept"] },
+            fallbackReason: "judge_unparseable_fallback",
+            costUsd,
+        };
+    }
+    return { decision: applyRouteJudgeVerdict(args.baseline, verdict), verdict, costUsd };
+}
+//# sourceMappingURL=route-judge.js.map

package/dist/core/model-router/route-judge.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"route-judge.js","sourceRoot":"","sources":["../../../src/core/model-router/route-judge.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,oBAAoB,EAAE,MAAM,mCAAmC,CAAC;AAGzE;;;;;;;GAOG;AAEH,uGAAqG;AACrG,MAAM,CAAC,MAAM,yBAAyB,GAAG;IACxC,wFAAwF;IACxF,sDAAsD;IACtD,yDAAyD;IACzD,0FAA0F;IAC1F,gHAAgH;IAChH,8FAA8F;IAC9F,2CAA2C;IAC3C,0JAA0J;CAC1J,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAEb,MAAM,CAAC,MAAM,6BAA6B,GAAG,GAAG,CAAC;AACjD,MAAM,CAAC,MAAM,6BAA6B,GAAG,MAAM,CAAC;AASpD,MAAM,UAAU,yBAAyB,CAAC,IAAiD,EAAU;IACpG,OAAO;QACN,kCAAkC,IAAI,CAAC,QAAQ,CAAC,IAAI,UAAU,IAAI,CAAC,QAAQ,CAAC,IAAI,YAAY,IAAI,CAAC,QAAQ,CAAC,UAAU,GAAG;QACvH,cAAc;QACd,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC;KAC1B,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAAA,CACb;AAED,MAAM,WAAW,GAAsB,CAAC,OAAO,EAAE,QAAQ,EAAE,WAAW,CAAC,CAAC;AACxE,MAAM,WAAW,GAAsB,CAAC,WAAW,EAAE,cAAc,EAAE,aAAa,EAAE,mBAAmB,CAAC,CAAC;AAEzG,MAAM,UAAU,sBAAsB,CAAC,IAAY,EAAiC;IACnF,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;IAC5B,MAAM,UAAU,GAAa,CAAC,OAAO,CAAC,CAAC;IACvC,MAAM,MAAM,GAAG,8BAA8B,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IAC5D,IAAI,MAAM,EAAE,CAAC,CAAC,CAAC;QAAE,UAAU,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;IACnD,MAAM,KAAK,GAAG,OAAO,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IACnC,MAAM,GAAG,GAAG,OAAO,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC;IACrC,IAAI,KAAK,IAAI,CAAC,IAAI,GAAG,GAAG,KAAK;QAAE,UAAU,CAAC,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,KAAK,EAAE,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC;IAE9E,KAAK,MAAM,SAAS,IAAI,UAAU,EAAE,CAAC;QACpC,IAAI,MAAe,CAAC;QACpB,IAAI,CAAC;YACJ,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;QAChC,CAAC;QAAC,MAAM,CAAC;YACR,SAAS;QACV,CAAC;QACD,IAAI,CAAC,MAAM,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC;YAAE,SAAS;QAC7E,MAAM,MAAM,GAAG,MAAiC,CAAC;QACjD,+FAA+F;QAC/F,IAAI,OAAO,MAAM,CAAC,IAAI,KAAK,QAAQ,IAAI,CAAC,WAAW,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC;YAAE,SAAS;QACpF,MAAM,IAAI,GACT,OAAO,MAAM,CAAC,IAAI,KAAK,QAAQ,IAAI,WAAW,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC;YACnE,CAAC,CAAE,MAAM,CAAC,IAA8B;YACxC,CAAC,CAAC,SAAS,CAAC;QACd,IAAI,CAAC,IAAI;YAAE,SAAS;QACpB,OAAO;YACN,IAAI,EAAE,MAAM,CAAC,IAAiC;YAC9C,IAAI;YACJ,OAAO,EAAE,MAAM,CAAC,OAAO,KAAK,IAAI;YAChC,MAAM,EAAE,OAAO,MAAM,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE;SAC5E,CAAC;IACH,CAAC;IACD,OAAO,SAAS,CAAC;AAAA,CACjB;AAED,uFAAuF;AACvF,MAAM,UAAU,sBAAsB,CAAC,QAAuB,EAAE,OAA0B,EAAiB;IAC1G,uFAAuF;IACvF,oFAAoF;IACpF,gFAAgF;IAChF,MAAM,IAAI,GACT,OAAO,CAAC,IAAI,KAAK,OAAO,IAAI,QAAQ,CAAC,IAAI,KAAK,OAAO,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC;IAC1G,OAAO;QACN,GAAG,QAAQ;QACX,IAAI;QACJ,IAAI,EAAE,OAAO,CAAC,IAAI;QAClB,UAAU,EAAE,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,UAAU,EAAE,IAAI,CAAC;QAC/C,UAAU,EAAE,SAAS,IAAI,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,EAAE;QAC/D,OAAO,EAAE,CAAC,GAAG,QAAQ,CAAC,OAAO,EAAE,gBAAgB,OAAO,CAAC,MAAM,IAAI,iBAAiB,EAAE,CAAC;KACrF,CAAC;AAAA,CACF;AAUD;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,aAAa,CAAC,IAUnC,EAAgC;IAChC,MAAM,OAAO,GAAG,MAAM,oBAAoB,CAAC;QAC1C,cAAc,EAAE,IAAI,CAAC,cAAc,IAAI,6BAA6B;QACpE,MAAM,EAAE,IAAI,CAAC,MAAM;QACnB,OAAO,EAAE,CAAC,MAAM,EAAE,EAAE,CACnB,IAAI,CAAC,QAAQ,CAAC;YACb,YAAY,EAAE,yBAAyB;YACvC,UAAU,EAAE,yBAAyB,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,MAAM,EAAE,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,CAAC;YACvF,MAAM;SACN,CAAC;KACH,CAAC,CAAC;IACH,MAAM,OAAO,GAAG,OAAO,CAAC,UAAU,EAAE,OAAO,IAAI,CAAC,CAAC;IAEjD,IAAI,OAAO,CAAC,OAAO,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC;QAC5C,OAAO;YACN,QAAQ,EAAE;gBACT,GAAG,IAAI,CAAC,QAAQ;gBAChB,OAAO,EAAE,CAAC,GAAG,IAAI,CAAC,QAAQ,CAAC,OAAO,EAAE,wCAAwC,CAAC;aAC7E;YACD,cAAc,EAAE,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,OAAO,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,4BAA4B;YACtG,OAAO;SACP,CAAC;IACH,CAAC;IACD,IAAI,OAAO,CAAC,UAAU,CAAC,UAAU,KAAK,OAAO,IAAI,OAAO,CAAC,UAAU,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;QAC9F,OAAO;YACN,QAAQ,EAAE,EAAE,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,CAAC,GAAG,IAAI,CAAC,QAAQ,CAAC,OAAO,EAAE,oCAAoC,CAAC,EAAE;YACzG,cAAc,EAAE,mBAAmB;YACnC,OAAO;SACP,CAAC;IACH,CAAC;IAED,MAAM,OAAO,GAAG,sBAAsB,CAAC,OAAO,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC;IAChE,IAAI,CAAC,OAAO,EAAE,CAAC;QACd,OAAO;YACN,QAAQ,EAAE,EAAE,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,CAAC,GAAG,IAAI,CAAC,QAAQ,CAAC,OAAO,EAAE,wCAAwC,CAAC,EAAE;YAC7G,cAAc,EAAE,4BAA4B;YAC5C,OAAO;SACP,CAAC;IACH,CAAC;IAED,OAAO,EAAE,QAAQ,EAAE,sBAAsB,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,CAAC,EAAE,OAAO,EAAE,OAAO,EAAE,CAAC;AAAA,CACtF","sourcesContent":["import { runBoundedCompletion } from \"../autonomy/bounded-completion.ts\";\nimport type { RouteDecision } from \"../autonomy/contracts.ts\";\n\n/**\n * Routing-only judge: a bounded, tool-less completion (default: the medium model) that decides the\n * final cheap/medium/expensive tier for a user prompt, refining the regex classifier's baseline.\n * Core rule: planning is never cheap unless the judge explicitly deems the task trivial. The judge\n * proposes; the existing pipeline (model resolution, auth, escalation, gates) still decides.\n * Failure is honest: unparseable/timeout/unavailable falls back to the baseline with a visible\n * reasonCode, never silently.\n */\n\n/** Static across calls — callers pass cacheRetention \"short\" so only the variable tail is billed. */\nexport const ROUTE_JUDGE_SYSTEM_PROMPT = [\n\t\"You are a routing judge for a coding agent. You only route; you never answer the task.\",\n\t\"Pick which model tier should handle the user prompt:\",\n\t'- \"cheap\": trivial, mechanical, read-only lookups only.',\n\t'- \"medium\": normal implementation, scoped edits, tests, and NON-trivial planning/design.',\n\t'- \"expensive\": architecture, ambiguity, security/auth, destructive or release operations, high-impact changes.',\n\t\"Planning, design, and strategy prompts are NEVER cheap unless the task is genuinely trivial.\",\n\t\"Respond with STRICT JSON only - no prose:\",\n\t'{\"tier\":\"cheap\"|\"medium\"|\"expensive\",\"risk\":\"read-only\"|\"scoped-write\"|\"high-impact\"|\"approval-required\",\"trivial\":true|false,\"reason\":\"<short reason>\"}',\n].join(\"\\n\");\n\nexport const ROUTE_JUDGE_MAX_OUTPUT_TOKENS = 128;\nexport const ROUTE_JUDGE_MAX_WALL_CLOCK_MS = 10_000;\n\nexport interface RouteJudgeVerdict {\n\ttier: \"cheap\" | \"medium\" | \"expensive\";\n\trisk: RouteDecision[\"risk\"];\n\ttrivial: boolean;\n\treason: string;\n}\n\nexport function buildRouteJudgeUserPrompt(args: { prompt: string; baseline: RouteDecision }): string {\n\treturn [\n\t\t`Baseline (regex) verdict: tier=${args.baseline.tier}, risk=${args.baseline.risk}, reason=${args.baseline.reasonCode}.`,\n\t\t\"User prompt:\",\n\t\targs.prompt.slice(0, 4000),\n\t].join(\"\\n\");\n}\n\nconst JUDGE_TIERS: readonly string[] = [\"cheap\", \"medium\", \"expensive\"];\nconst JUDGE_RISKS: readonly string[] = [\"read-only\", \"scoped-write\", \"high-impact\", \"approval-required\"];\n\nexport function parseRouteJudgeVerdict(text: string): RouteJudgeVerdict | undefined {\n\tconst trimmed = text.trim();\n\tconst candidates: string[] = [trimmed];\n\tconst fenced = /```(?:json)?\\s*([\\s\\S]*?)```/.exec(trimmed);\n\tif (fenced?.[1]) candidates.push(fenced[1].trim());\n\tconst start = trimmed.indexOf(\"{\");\n\tconst end = trimmed.lastIndexOf(\"}\");\n\tif (start >= 0 && end > start) candidates.push(trimmed.slice(start, end + 1));\n\n\tfor (const candidate of candidates) {\n\t\tlet parsed: unknown;\n\t\ttry {\n\t\t\tparsed = JSON.parse(candidate);\n\t\t} catch {\n\t\t\tcontinue;\n\t\t}\n\t\tif (!parsed || typeof parsed !== \"object\" || Array.isArray(parsed)) continue;\n\t\tconst record = parsed as Record<string, unknown>;\n\t\t// The judge may never select the learning tier or anything outside the three foreground tiers.\n\t\tif (typeof record.tier !== \"string\" || !JUDGE_TIERS.includes(record.tier)) continue;\n\t\tconst risk =\n\t\t\ttypeof record.risk === \"string\" && JUDGE_RISKS.includes(record.risk)\n\t\t\t\t? (record.risk as RouteDecision[\"risk\"])\n\t\t\t\t: undefined;\n\t\tif (!risk) continue;\n\t\treturn {\n\t\t\ttier: record.tier as RouteJudgeVerdict[\"tier\"],\n\t\t\trisk,\n\t\t\ttrivial: record.trivial === true,\n\t\t\treason: typeof record.reason === \"string\" ? record.reason.slice(0, 200) : \"\",\n\t\t};\n\t}\n\treturn undefined;\n}\n\n/** Merge a judge verdict into the baseline decision (pure; never returns learning). */\nexport function applyRouteJudgeVerdict(baseline: RouteDecision, verdict: RouteJudgeVerdict): RouteDecision {\n\t// Enforce the core rule in code, not just in the judge prompt: downgrading a non-cheap\n\t// baseline to cheap requires an EXPLICIT trivial verdict. An untrusted judge saying\n\t// {tier:\"cheap\", trivial:false} for an elevated prompt keeps the baseline tier.\n\tconst tier =\n\t\tverdict.tier === \"cheap\" && baseline.tier !== \"cheap\" && !verdict.trivial ? baseline.tier : verdict.tier;\n\treturn {\n\t\t...baseline,\n\t\ttier,\n\t\trisk: verdict.risk,\n\t\tconfidence: Math.max(baseline.confidence, 0.75),\n\t\treasonCode: `judge_${tier}${verdict.trivial ? \"_trivial\" : \"\"}`,\n\t\treasons: [...baseline.reasons, `Route judge: ${verdict.reason || \"no reason given\"}`],\n\t};\n}\n\nexport interface RouteJudgeRunResult {\n\tdecision: RouteDecision;\n\tverdict?: RouteJudgeVerdict;\n\t/** Set when the judge could not decide and the baseline was kept. */\n\tfallbackReason?: string;\n\tcostUsd: number;\n}\n\n/**\n * Run the judge over a baseline decision. The completion executor is injected (production:\n * AgentSession.runIsolatedCompletion on the judge model). Never throws; every failure keeps the\n * baseline with a visible fallbackReason.\n */\nexport async function runRouteJudge(args: {\n\tprompt: string;\n\tbaseline: RouteDecision;\n\tcomplete: (input: { systemPrompt: string; userPrompt: string; signal?: AbortSignal }) => Promise<{\n\t\ttext: string;\n\t\tcostUsd: number;\n\t\tstopReason: string;\n\t}>;\n\tsignal?: AbortSignal;\n\tmaxWallClockMs?: number;\n}): Promise<RouteJudgeRunResult> {\n\tconst bounded = await runBoundedCompletion({\n\t\tmaxWallClockMs: args.maxWallClockMs ?? ROUTE_JUDGE_MAX_WALL_CLOCK_MS,\n\t\tsignal: args.signal,\n\t\texecute: (signal) =>\n\t\t\targs.complete({\n\t\t\t\tsystemPrompt: ROUTE_JUDGE_SYSTEM_PROMPT,\n\t\t\t\tuserPrompt: buildRouteJudgeUserPrompt({ prompt: args.prompt, baseline: args.baseline }),\n\t\t\t\tsignal,\n\t\t\t}),\n\t});\n\tconst costUsd = bounded.completion?.costUsd ?? 0;\n\n\tif (bounded.failure || !bounded.completion) {\n\t\treturn {\n\t\t\tdecision: {\n\t\t\t\t...args.baseline,\n\t\t\t\treasons: [...args.baseline.reasons, \"Route judge unavailable; baseline kept\"],\n\t\t\t},\n\t\t\tfallbackReason: bounded.failure ? `judge_${bounded.failure.reasonCode}` : \"judge_unavailable_fallback\",\n\t\t\tcostUsd,\n\t\t};\n\t}\n\tif (bounded.completion.stopReason === \"error\" || bounded.completion.stopReason === \"aborted\") {\n\t\treturn {\n\t\t\tdecision: { ...args.baseline, reasons: [...args.baseline.reasons, \"Route judge errored; baseline kept\"] },\n\t\t\tfallbackReason: \"judge_model_error\",\n\t\t\tcostUsd,\n\t\t};\n\t}\n\n\tconst verdict = parseRouteJudgeVerdict(bounded.completion.text);\n\tif (!verdict) {\n\t\treturn {\n\t\t\tdecision: { ...args.baseline, reasons: [...args.baseline.reasons, \"Route judge unparseable; baseline kept\"] },\n\t\t\tfallbackReason: \"judge_unparseable_fallback\",\n\t\t\tcostUsd,\n\t\t};\n\t}\n\n\treturn { decision: applyRouteJudgeVerdict(args.baseline, verdict), verdict, costUsd };\n}\n"]}

package/dist/core/model-router/status.d.ts

@@ -1,17 +1,20 @@
+import type { RouteDecision } from "../autonomy/contracts.ts";
 import type { SessionEntry } from "../session-manager.ts";
 import type { ModelRouterIntent } from "./intent-classifier.ts";
 export declare const MODEL_ROUTER_DECISION_CUSTOM_TYPE = "model_router_decision";
 export type ModelRouterStatusSettings = {
     enabled: boolean;
     cheapModel?: string;
+    mediumModel?: string;
     expensiveModel?: string;
     learningModel?: string;
 };
 export type ModelRouterDecisionStatus = {
-    intent: ModelRouterIntent;
+    route: RouteDecision;
     routedModel: string;
     outcome: "routed" | "escalated" | "failed";
     retryModel?: string;
+    intent?: ModelRouterIntent;
 };
 export declare function getRecentModelRouterDecisions(entries: SessionEntry[], limit?: number): ModelRouterDecisionStatus[];
 export declare function formatModelRouterStatus(settings: ModelRouterStatusSettings, lastDecision?: ModelRouterDecisionStatus, formatLabel?: (label: string) => string, recentDecisions?: ModelRouterDecisionStatus[], lastSkipReason?: string, latestIntent?: ModelRouterIntent): string;

package/dist/core/model-router/status.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"status.d.ts","sourceRoot":"","sources":["../../../src/core/model-router/status.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,uBAAuB,CAAC;AAC1D,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,wBAAwB,CAAC;AAEhE,eAAO,MAAM,iCAAiC,0BAA0B,CAAC;AAEzE,MAAM,MAAM,yBAAyB,GAAG;IACvC,OAAO,EAAE,OAAO,CAAC;IACjB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,aAAa,CAAC,EAAE,MAAM,CAAC;CACvB,CAAC;AAEF,MAAM,MAAM,yBAAyB,GAAG;IACvC,MAAM,EAAE,iBAAiB,CAAC;IAC1B,WAAW,EAAE,MAAM,CAAC;IACpB,OAAO,EAAE,QAAQ,GAAG,WAAW,GAAG,QAAQ,CAAC;IAC3C,UAAU,CAAC,EAAE,MAAM,CAAC;CACpB,CAAC;AAuBF,wBAAgB,6BAA6B,CAAC,OAAO,EAAE,YAAY,EAAE,EAAE,KAAK,SAAI,GAAG,yBAAyB,EAAE,CAQ7G;AAED,wBAAgB,uBAAuB,CACtC,QAAQ,EAAE,yBAAyB,EACnC,YAAY,CAAC,EAAE,yBAAyB,EACxC,WAAW,GAAE,CAAC,KAAK,EAAE,MAAM,KAAK,MAAyB,EACzD,eAAe,GAAE,yBAAyB,EAAO,EACjD,cAAc,CAAC,EAAE,MAAM,EACvB,YAAY,CAAC,EAAE,iBAAiB,GAC9B,MAAM,CA8BR","sourcesContent":["import type { SessionEntry } from \"../session-manager.ts\";\nimport type { ModelRouterIntent } from \"./intent-classifier.ts\";\n\nexport const MODEL_ROUTER_DECISION_CUSTOM_TYPE = \"model_router_decision\";\n\nexport type ModelRouterStatusSettings = {\n\tenabled: boolean;\n\tcheapModel?: string;\n\texpensiveModel?: string;\n\tlearningModel?: string;\n};\n\nexport type ModelRouterDecisionStatus = {\n\tintent: ModelRouterIntent;\n\troutedModel: string;\n\toutcome: \"routed\" | \"escalated\" | \"failed\";\n\tretryModel?: string;\n};\n\nfunction isModelRouterDecisionStatus(data: unknown): data is ModelRouterDecisionStatus {\n\tif (!data || typeof data !== \"object\") return false;\n\tconst record = data as Partial<ModelRouterDecisionStatus>;\n\treturn (\n\t\t(record.intent === \"research\" || record.intent === \"modify\") &&\n\t\ttypeof record.routedModel === \"string\" &&\n\t\t(record.outcome === \"routed\" || record.outcome === \"escalated\" || record.outcome === \"failed\") &&\n\t\t(record.retryModel === undefined || typeof record.retryModel === \"string\")\n\t);\n}\n\nfunction formatDecision(decision: ModelRouterDecisionStatus): string {\n\tlet text = `${decision.intent} -> ${decision.routedModel}`;\n\tif (decision.outcome === \"escalated\" && decision.retryModel) {\n\t\ttext += ` (escalated -> ${decision.retryModel})`;\n\t} else if (decision.outcome === \"failed\") {\n\t\ttext += \" (failed)\";\n\t}\n\treturn text;\n}\n\nexport function getRecentModelRouterDecisions(entries: SessionEntry[], limit = 3): ModelRouterDecisionStatus[] {\n\tconst decisions: ModelRouterDecisionStatus[] = [];\n\tfor (const entry of entries) {\n\t\tif (entry.type !== \"custom\" || entry.customType !== MODEL_ROUTER_DECISION_CUSTOM_TYPE) continue;\n\t\tif (!isModelRouterDecisionStatus(entry.data)) continue;\n\t\tdecisions.push(entry.data);\n\t}\n\treturn decisions.slice(-limit);\n}\n\nexport function formatModelRouterStatus(\n\tsettings: ModelRouterStatusSettings,\n\tlastDecision?: ModelRouterDecisionStatus,\n\tformatLabel: (label: string) => string = (label) => label,\n\trecentDecisions: ModelRouterDecisionStatus[] = [],\n\tlastSkipReason?: string,\n\tlatestIntent?: ModelRouterIntent,\n): string {\n\tconst effectiveLastDecision = lastSkipReason ? undefined : lastDecision;\n\tconst lines = [\n\t\t`${formatLabel(\"Status:\")} ${settings.enabled ? \"enabled\" : \"disabled\"}`,\n\t\t`${formatLabel(\"Cheap model:\")} ${settings.cheapModel ?? \"unset\"}`,\n\t\t`${formatLabel(\"Expensive model:\")} ${settings.expensiveModel ?? \"unset\"}`,\n\t\t`${formatLabel(\"Learning model:\")} ${settings.learningModel ?? \"active\"}`,\n\t];\n\tif (!settings.enabled) {\n\t\tlines.push(`${formatLabel(\"Routing:\")} inactive (disabled)`);\n\t} else if (lastSkipReason) {\n\t\tlines.push(`${formatLabel(\"Routing:\")} skipped (${lastSkipReason})`);\n\t} else if (effectiveLastDecision) {\n\t\tlines.push(`${formatLabel(\"Routing:\")} active`);\n\t} else {\n\t\tlines.push(`${formatLabel(\"Routing:\")} waiting for prompt`);\n\t}\n\tlines.push(`${formatLabel(\"Latest intent:\")} ${latestIntent ?? \"none\"}`);\n\tif (!effectiveLastDecision) {\n\t\tlines.push(`${formatLabel(\"Last decision:\")} none`);\n\t} else {\n\t\tlines.push(`${formatLabel(\"Last decision:\")} ${formatDecision(effectiveLastDecision)}`);\n\t}\n\tif (recentDecisions.length > 0) {\n\t\tlines.push(formatLabel(\"Recent decisions:\"));\n\t\tfor (const decision of recentDecisions) {\n\t\t\tlines.push(`- ${formatDecision(decision)}`);\n\t\t}\n\t}\n\treturn lines.join(\"\\n\");\n}\n"]}
+{"version":3,"file":"status.d.ts","sourceRoot":"","sources":["../../../src/core/model-router/status.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAa,aAAa,EAAa,MAAM,0BAA0B,CAAC;AACpF,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,uBAAuB,CAAC;AAC1D,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,wBAAwB,CAAC;AAEhE,eAAO,MAAM,iCAAiC,0BAA0B,CAAC;AAEzE,MAAM,MAAM,yBAAyB,GAAG;IACvC,OAAO,EAAE,OAAO,CAAC;IACjB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,aAAa,CAAC,EAAE,MAAM,CAAC;CACvB,CAAC;AAEF,MAAM,MAAM,yBAAyB,GAAG;IACvC,KAAK,EAAE,aAAa,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,OAAO,EAAE,QAAQ,GAAG,WAAW,GAAG,QAAQ,CAAC;IAC3C,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,MAAM,CAAC,EAAE,iBAAiB,CAAC;CAC3B,CAAC;AAiDF,wBAAgB,6BAA6B,CAAC,OAAO,EAAE,YAAY,EAAE,EAAE,KAAK,SAAI,GAAG,yBAAyB,EAAE,CAQ7G;AAED,wBAAgB,uBAAuB,CACtC,QAAQ,EAAE,yBAAyB,EACnC,YAAY,CAAC,EAAE,yBAAyB,EACxC,WAAW,GAAE,CAAC,KAAK,EAAE,MAAM,KAAK,MAAyB,EACzD,eAAe,GAAE,yBAAyB,EAAO,EACjD,cAAc,CAAC,EAAE,MAAM,EACvB,YAAY,CAAC,EAAE,iBAAiB,GAC9B,MAAM,CA+BR","sourcesContent":["import type { ModelTier, RouteDecision, RouteRisk } from \"../autonomy/contracts.ts\";\nimport type { SessionEntry } from \"../session-manager.ts\";\nimport type { ModelRouterIntent } from \"./intent-classifier.ts\";\n\nexport const MODEL_ROUTER_DECISION_CUSTOM_TYPE = \"model_router_decision\";\n\nexport type ModelRouterStatusSettings = {\n\tenabled: boolean;\n\tcheapModel?: string;\n\tmediumModel?: string;\n\texpensiveModel?: string;\n\tlearningModel?: string;\n};\n\nexport type ModelRouterDecisionStatus = {\n\troute: RouteDecision;\n\troutedModel: string;\n\toutcome: \"routed\" | \"escalated\" | \"failed\";\n\tretryModel?: string;\n\tintent?: ModelRouterIntent;\n};\n\nfunction isRouteDecision(value: unknown): value is RouteDecision {\n\tif (!value || typeof value !== \"object\") return false;\n\tconst route = value as Partial<RouteDecision>;\n\tconst validTiers = new Set<ModelTier>([\"cheap\", \"medium\", \"expensive\", \"learning\"]);\n\tconst validRisks = new Set<RouteRisk>([\"read-only\", \"scoped-write\", \"high-impact\", \"approval-required\"]);\n\n\treturn (\n\t\ttypeof route.confidence === \"number\" &&\n\t\ttypeof route.reasonCode === \"string\" &&\n\t\tArray.isArray(route.reasons) &&\n\t\troute.reasons.every((r) => typeof r === \"string\") &&\n\t\tvalidTiers.has(route.tier as ModelTier) &&\n\t\tvalidRisks.has(route.risk as RouteRisk) &&\n\t\t(route.model === undefined || typeof route.model === \"string\") &&\n\t\t(route.fallbackFrom === undefined ||\n\t\t\troute.fallbackFrom === \"cheap\" ||\n\t\t\troute.fallbackFrom === \"medium\" ||\n\t\t\troute.fallbackFrom === \"expensive\" ||\n\t\t\troute.fallbackFrom === \"learning\") &&\n\t\t(route.createdAt === undefined || typeof route.createdAt === \"string\")\n\t);\n}\n\nfunction isModelRouterDecisionStatus(data: unknown): data is ModelRouterDecisionStatus {\n\tif (!data || typeof data !== \"object\") return false;\n\tconst record = data as Partial<ModelRouterDecisionStatus>;\n\treturn (\n\t\tisRouteDecision(record.route) &&\n\t\trecord.route.tier !== \"learning\" && // Validate user prompt decisions never need learning tier in runtime\n\t\ttypeof record.routedModel === \"string\" &&\n\t\t(record.outcome === \"routed\" || record.outcome === \"escalated\" || record.outcome === \"failed\") &&\n\t\t(record.retryModel === undefined || typeof record.retryModel === \"string\") &&\n\t\t(record.intent === undefined || record.intent === \"research\" || record.intent === \"modify\")\n\t);\n}\n\nfunction formatDecision(decision: ModelRouterDecisionStatus): string {\n\tconst { tier, risk, reasonCode } = decision.route;\n\tlet outcomeText: string = decision.outcome;\n\tif (decision.outcome === \"escalated\" && decision.retryModel) {\n\t\toutcomeText = `escalated -> ${decision.retryModel}`;\n\t} else if (decision.outcome === \"failed\") {\n\t\toutcomeText = \"failed\";\n\t}\n\treturn `${tier}/${risk} -> ${decision.routedModel} (${reasonCode}, ${outcomeText})`;\n}\n\nexport function getRecentModelRouterDecisions(entries: SessionEntry[], limit = 3): ModelRouterDecisionStatus[] {\n\tconst decisions: ModelRouterDecisionStatus[] = [];\n\tfor (const entry of entries) {\n\t\tif (entry.type !== \"custom\" || entry.customType !== MODEL_ROUTER_DECISION_CUSTOM_TYPE) continue;\n\t\tif (!isModelRouterDecisionStatus(entry.data)) continue;\n\t\tdecisions.push(entry.data);\n\t}\n\treturn decisions.slice(-limit);\n}\n\nexport function formatModelRouterStatus(\n\tsettings: ModelRouterStatusSettings,\n\tlastDecision?: ModelRouterDecisionStatus,\n\tformatLabel: (label: string) => string = (label) => label,\n\trecentDecisions: ModelRouterDecisionStatus[] = [],\n\tlastSkipReason?: string,\n\tlatestIntent?: ModelRouterIntent,\n): string {\n\tconst effectiveLastDecision = lastSkipReason ? undefined : lastDecision;\n\tconst lines = [\n\t\t`${formatLabel(\"Status:\")} ${settings.enabled ? \"enabled\" : \"disabled\"}`,\n\t\t`${formatLabel(\"Cheap model:\")} ${settings.cheapModel ?? \"unset\"}`,\n\t\t`${formatLabel(\"Medium model:\")} ${settings.mediumModel ?? \"unset\"}`,\n\t\t`${formatLabel(\"Expensive model:\")} ${settings.expensiveModel ?? \"unset\"}`,\n\t\t`${formatLabel(\"Learning model:\")} ${settings.learningModel ?? \"active\"}`,\n\t];\n\tif (!settings.enabled) {\n\t\tlines.push(`${formatLabel(\"Routing:\")} inactive (disabled)`);\n\t} else if (lastSkipReason) {\n\t\tlines.push(`${formatLabel(\"Routing:\")} skipped (${lastSkipReason})`);\n\t} else if (effectiveLastDecision) {\n\t\tlines.push(`${formatLabel(\"Routing:\")} active`);\n\t} else {\n\t\tlines.push(`${formatLabel(\"Routing:\")} waiting for prompt`);\n\t}\n\tlines.push(`${formatLabel(\"Latest intent:\")} ${latestIntent ?? \"none\"}`);\n\tif (!effectiveLastDecision) {\n\t\tlines.push(`${formatLabel(\"Last decision:\")} none`);\n\t} else {\n\t\tlines.push(`${formatLabel(\"Last decision:\")} ${formatDecision(effectiveLastDecision)}`);\n\t}\n\tif (recentDecisions.length > 0) {\n\t\tlines.push(formatLabel(\"Recent decisions:\"));\n\t\tfor (const decision of recentDecisions) {\n\t\t\tlines.push(`- ${formatDecision(decision)}`);\n\t\t}\n\t}\n\treturn lines.join(\"\\n\");\n}\n"]}

package/dist/core/model-router/status.js

@@ -1,22 +1,45 @@
 export const MODEL_ROUTER_DECISION_CUSTOM_TYPE = "model_router_decision";
+function isRouteDecision(value) {
+    if (!value || typeof value !== "object")
+        return false;
+    const route = value;
+    const validTiers = new Set(["cheap", "medium", "expensive", "learning"]);
+    const validRisks = new Set(["read-only", "scoped-write", "high-impact", "approval-required"]);
+    return (typeof route.confidence === "number" &&
+        typeof route.reasonCode === "string" &&
+        Array.isArray(route.reasons) &&
+        route.reasons.every((r) => typeof r === "string") &&
+        validTiers.has(route.tier) &&
+        validRisks.has(route.risk) &&
+        (route.model === undefined || typeof route.model === "string") &&
+        (route.fallbackFrom === undefined ||
+            route.fallbackFrom === "cheap" ||
+            route.fallbackFrom === "medium" ||
+            route.fallbackFrom === "expensive" ||
+            route.fallbackFrom === "learning") &&
+        (route.createdAt === undefined || typeof route.createdAt === "string"));
+}
 function isModelRouterDecisionStatus(data) {
     if (!data || typeof data !== "object")
         return false;
     const record = data;
-    return ((record.intent === "research" || record.intent === "modify") &&
+    return (isRouteDecision(record.route) &&
+        record.route.tier !== "learning" && // Validate user prompt decisions never need learning tier in runtime
         typeof record.routedModel === "string" &&
         (record.outcome === "routed" || record.outcome === "escalated" || record.outcome === "failed") &&
-        (record.retryModel === undefined || typeof record.retryModel === "string"));
+        (record.retryModel === undefined || typeof record.retryModel === "string") &&
+        (record.intent === undefined || record.intent === "research" || record.intent === "modify"));
 }
 function formatDecision(decision) {
-    let text = `${decision.intent} -> ${decision.routedModel}`;
+    const { tier, risk, reasonCode } = decision.route;
+    let outcomeText = decision.outcome;
     if (decision.outcome === "escalated" && decision.retryModel) {
-        text += ` (escalated -> ${decision.retryModel})`;
+        outcomeText = `escalated -> ${decision.retryModel}`;
     }
     else if (decision.outcome === "failed") {
-        text += " (failed)";
+        outcomeText = "failed";
     }
-    return text;
+    return `${tier}/${risk} -> ${decision.routedModel} (${reasonCode}, ${outcomeText})`;
 }
 export function getRecentModelRouterDecisions(entries, limit = 3) {
     const decisions = [];
@@ -34,6 +57,7 @@ export function formatModelRouterStatus(settings, lastDecision, formatLabel = (l
     const lines = [
         `${formatLabel("Status:")} ${settings.enabled ? "enabled" : "disabled"}`,
         `${formatLabel("Cheap model:")} ${settings.cheapModel ?? "unset"}`,
+        `${formatLabel("Medium model:")} ${settings.mediumModel ?? "unset"}`,
         `${formatLabel("Expensive model:")} ${settings.expensiveModel ?? "unset"}`,
         `${formatLabel("Learning model:")} ${settings.learningModel ?? "active"}`,
     ];

package/dist/core/model-router/status.js.map

@@ -1 +1 @@
-{"version":3,"file":"status.js","sourceRoot":"","sources":["../../../src/core/model-router/status.ts"],"names":[],"mappings":"AAGA,MAAM,CAAC,MAAM,iCAAiC,GAAG,uBAAuB,CAAC;AAgBzE,SAAS,2BAA2B,CAAC,IAAa,EAAqC;IACtF,IAAI,CAAC,IAAI,IAAI,OAAO,IAAI,KAAK,QAAQ;QAAE,OAAO,KAAK,CAAC;IACpD,MAAM,MAAM,GAAG,IAA0C,CAAC;IAC1D,OAAO,CACN,CAAC,MAAM,CAAC,MAAM,KAAK,UAAU,IAAI,MAAM,CAAC,MAAM,KAAK,QAAQ,CAAC;QAC5D,OAAO,MAAM,CAAC,WAAW,KAAK,QAAQ;QACtC,CAAC,MAAM,CAAC,OAAO,KAAK,QAAQ,IAAI,MAAM,CAAC,OAAO,KAAK,WAAW,IAAI,MAAM,CAAC,OAAO,KAAK,QAAQ,CAAC;QAC9F,CAAC,MAAM,CAAC,UAAU,KAAK,SAAS,IAAI,OAAO,MAAM,CAAC,UAAU,KAAK,QAAQ,CAAC,CAC1E,CAAC;AAAA,CACF;AAED,SAAS,cAAc,CAAC,QAAmC,EAAU;IACpE,IAAI,IAAI,GAAG,GAAG,QAAQ,CAAC,MAAM,OAAO,QAAQ,CAAC,WAAW,EAAE,CAAC;IAC3D,IAAI,QAAQ,CAAC,OAAO,KAAK,WAAW,IAAI,QAAQ,CAAC,UAAU,EAAE,CAAC;QAC7D,IAAI,IAAI,kBAAkB,QAAQ,CAAC,UAAU,GAAG,CAAC;IAClD,CAAC;SAAM,IAAI,QAAQ,CAAC,OAAO,KAAK,QAAQ,EAAE,CAAC;QAC1C,IAAI,IAAI,WAAW,CAAC;IACrB,CAAC;IACD,OAAO,IAAI,CAAC;AAAA,CACZ;AAED,MAAM,UAAU,6BAA6B,CAAC,OAAuB,EAAE,KAAK,GAAG,CAAC,EAA+B;IAC9G,MAAM,SAAS,GAAgC,EAAE,CAAC;IAClD,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC7B,IAAI,KAAK,CAAC,IAAI,KAAK,QAAQ,IAAI,KAAK,CAAC,UAAU,KAAK,iCAAiC;YAAE,SAAS;QAChG,IAAI,CAAC,2BAA2B,CAAC,KAAK,CAAC,IAAI,CAAC;YAAE,SAAS;QACvD,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAC5B,CAAC;IACD,OAAO,SAAS,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC;AAAA,CAC/B;AAED,MAAM,UAAU,uBAAuB,CACtC,QAAmC,EACnC,YAAwC,EACxC,WAAW,GAA8B,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,EACzD,eAAe,GAAgC,EAAE,EACjD,cAAuB,EACvB,YAAgC,EACvB;IACT,MAAM,qBAAqB,GAAG,cAAc,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,YAAY,CAAC;IACxE,MAAM,KAAK,GAAG;QACb,GAAG,WAAW,CAAC,SAAS,CAAC,IAAI,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,UAAU,EAAE;QACxE,GAAG,WAAW,CAAC,cAAc,CAAC,IAAI,QAAQ,CAAC,UAAU,IAAI,OAAO,EAAE;QAClE,GAAG,WAAW,CAAC,kBAAkB,CAAC,IAAI,QAAQ,CAAC,cAAc,IAAI,OAAO,EAAE;QAC1E,GAAG,WAAW,CAAC,iBAAiB,CAAC,IAAI,QAAQ,CAAC,aAAa,IAAI,QAAQ,EAAE;KACzE,CAAC;IACF,IAAI,CAAC,QAAQ,CAAC,OAAO,EAAE,CAAC;QACvB,KAAK,CAAC,IAAI,CAAC,GAAG,WAAW,CAAC,UAAU,CAAC,sBAAsB,CAAC,CAAC;IAC9D,CAAC;SAAM,IAAI,cAAc,EAAE,CAAC;QAC3B,KAAK,CAAC,IAAI,CAAC,GAAG,WAAW,CAAC,UAAU,CAAC,aAAa,cAAc,GAAG,CAAC,CAAC;IACtE,CAAC;SAAM,IAAI,qBAAqB,EAAE,CAAC;QAClC,KAAK,CAAC,IAAI,CAAC,GAAG,WAAW,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC;IACjD,CAAC;SAAM,CAAC;QACP,KAAK,CAAC,IAAI,CAAC,GAAG,WAAW,CAAC,UAAU,CAAC,qBAAqB,CAAC,CAAC;IAC7D,CAAC;IACD,KAAK,CAAC,IAAI,CAAC,GAAG,WAAW,CAAC,gBAAgB,CAAC,IAAI,YAAY,IAAI,MAAM,EAAE,CAAC,CAAC;IACzE,IAAI,CAAC,qBAAqB,EAAE,CAAC;QAC5B,KAAK,CAAC,IAAI,CAAC,GAAG,WAAW,CAAC,gBAAgB,CAAC,OAAO,CAAC,CAAC;IACrD,CAAC;SAAM,CAAC;QACP,KAAK,CAAC,IAAI,CAAC,GAAG,WAAW,CAAC,gBAAgB,CAAC,IAAI,cAAc,CAAC,qBAAqB,CAAC,EAAE,CAAC,CAAC;IACzF,CAAC;IACD,IAAI,eAAe,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAChC,KAAK,CAAC,IAAI,CAAC,WAAW,CAAC,mBAAmB,CAAC,CAAC,CAAC;QAC7C,KAAK,MAAM,QAAQ,IAAI,eAAe,EAAE,CAAC;YACxC,KAAK,CAAC,IAAI,CAAC,KAAK,cAAc,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC;QAC7C,CAAC;IACF,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAAA,CACxB","sourcesContent":["import type { SessionEntry } from \"../session-manager.ts\";\nimport type { ModelRouterIntent } from \"./intent-classifier.ts\";\n\nexport const MODEL_ROUTER_DECISION_CUSTOM_TYPE = \"model_router_decision\";\n\nexport type ModelRouterStatusSettings = {\n\tenabled: boolean;\n\tcheapModel?: string;\n\texpensiveModel?: string;\n\tlearningModel?: string;\n};\n\nexport type ModelRouterDecisionStatus = {\n\tintent: ModelRouterIntent;\n\troutedModel: string;\n\toutcome: \"routed\" | \"escalated\" | \"failed\";\n\tretryModel?: string;\n};\n\nfunction isModelRouterDecisionStatus(data: unknown): data is ModelRouterDecisionStatus {\n\tif (!data || typeof data !== \"object\") return false;\n\tconst record = data as Partial<ModelRouterDecisionStatus>;\n\treturn (\n\t\t(record.intent === \"research\" || record.intent === \"modify\") &&\n\t\ttypeof record.routedModel === \"string\" &&\n\t\t(record.outcome === \"routed\" || record.outcome === \"escalated\" || record.outcome === \"failed\") &&\n\t\t(record.retryModel === undefined || typeof record.retryModel === \"string\")\n\t);\n}\n\nfunction formatDecision(decision: ModelRouterDecisionStatus): string {\n\tlet text = `${decision.intent} -> ${decision.routedModel}`;\n\tif (decision.outcome === \"escalated\" && decision.retryModel) {\n\t\ttext += ` (escalated -> ${decision.retryModel})`;\n\t} else if (decision.outcome === \"failed\") {\n\t\ttext += \" (failed)\";\n\t}\n\treturn text;\n}\n\nexport function getRecentModelRouterDecisions(entries: SessionEntry[], limit = 3): ModelRouterDecisionStatus[] {\n\tconst decisions: ModelRouterDecisionStatus[] = [];\n\tfor (const entry of entries) {\n\t\tif (entry.type !== \"custom\" || entry.customType !== MODEL_ROUTER_DECISION_CUSTOM_TYPE) continue;\n\t\tif (!isModelRouterDecisionStatus(entry.data)) continue;\n\t\tdecisions.push(entry.data);\n\t}\n\treturn decisions.slice(-limit);\n}\n\nexport function formatModelRouterStatus(\n\tsettings: ModelRouterStatusSettings,\n\tlastDecision?: ModelRouterDecisionStatus,\n\tformatLabel: (label: string) => string = (label) => label,\n\trecentDecisions: ModelRouterDecisionStatus[] = [],\n\tlastSkipReason?: string,\n\tlatestIntent?: ModelRouterIntent,\n): string {\n\tconst effectiveLastDecision = lastSkipReason ? undefined : lastDecision;\n\tconst lines = [\n\t\t`${formatLabel(\"Status:\")} ${settings.enabled ? \"enabled\" : \"disabled\"}`,\n\t\t`${formatLabel(\"Cheap model:\")} ${settings.cheapModel ?? \"unset\"}`,\n\t\t`${formatLabel(\"Expensive model:\")} ${settings.expensiveModel ?? \"unset\"}`,\n\t\t`${formatLabel(\"Learning model:\")} ${settings.learningModel ?? \"active\"}`,\n\t];\n\tif (!settings.enabled) {\n\t\tlines.push(`${formatLabel(\"Routing:\")} inactive (disabled)`);\n\t} else if (lastSkipReason) {\n\t\tlines.push(`${formatLabel(\"Routing:\")} skipped (${lastSkipReason})`);\n\t} else if (effectiveLastDecision) {\n\t\tlines.push(`${formatLabel(\"Routing:\")} active`);\n\t} else {\n\t\tlines.push(`${formatLabel(\"Routing:\")} waiting for prompt`);\n\t}\n\tlines.push(`${formatLabel(\"Latest intent:\")} ${latestIntent ?? \"none\"}`);\n\tif (!effectiveLastDecision) {\n\t\tlines.push(`${formatLabel(\"Last decision:\")} none`);\n\t} else {\n\t\tlines.push(`${formatLabel(\"Last decision:\")} ${formatDecision(effectiveLastDecision)}`);\n\t}\n\tif (recentDecisions.length > 0) {\n\t\tlines.push(formatLabel(\"Recent decisions:\"));\n\t\tfor (const decision of recentDecisions) {\n\t\t\tlines.push(`- ${formatDecision(decision)}`);\n\t\t}\n\t}\n\treturn lines.join(\"\\n\");\n}\n"]}
+{"version":3,"file":"status.js","sourceRoot":"","sources":["../../../src/core/model-router/status.ts"],"names":[],"mappings":"AAIA,MAAM,CAAC,MAAM,iCAAiC,GAAG,uBAAuB,CAAC;AAkBzE,SAAS,eAAe,CAAC,KAAc,EAA0B;IAChE,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ;QAAE,OAAO,KAAK,CAAC;IACtD,MAAM,KAAK,GAAG,KAA+B,CAAC;IAC9C,MAAM,UAAU,GAAG,IAAI,GAAG,CAAY,CAAC,OAAO,EAAE,QAAQ,EAAE,WAAW,EAAE,UAAU,CAAC,CAAC,CAAC;IACpF,MAAM,UAAU,GAAG,IAAI,GAAG,CAAY,CAAC,WAAW,EAAE,cAAc,EAAE,aAAa,EAAE,mBAAmB,CAAC,CAAC,CAAC;IAEzG,OAAO,CACN,OAAO,KAAK,CAAC,UAAU,KAAK,QAAQ;QACpC,OAAO,KAAK,CAAC,UAAU,KAAK,QAAQ;QACpC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC;QAC5B,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,CAAC;QACjD,UAAU,CAAC,GAAG,CAAC,KAAK,CAAC,IAAiB,CAAC;QACvC,UAAU,CAAC,GAAG,CAAC,KAAK,CAAC,IAAiB,CAAC;QACvC,CAAC,KAAK,CAAC,KAAK,KAAK,SAAS,IAAI,OAAO,KAAK,CAAC,KAAK,KAAK,QAAQ,CAAC;QAC9D,CAAC,KAAK,CAAC,YAAY,KAAK,SAAS;YAChC,KAAK,CAAC,YAAY,KAAK,OAAO;YAC9B,KAAK,CAAC,YAAY,KAAK,QAAQ;YAC/B,KAAK,CAAC,YAAY,KAAK,WAAW;YAClC,KAAK,CAAC,YAAY,KAAK,UAAU,CAAC;QACnC,CAAC,KAAK,CAAC,SAAS,KAAK,SAAS,IAAI,OAAO,KAAK,CAAC,SAAS,KAAK,QAAQ,CAAC,CACtE,CAAC;AAAA,CACF;AAED,SAAS,2BAA2B,CAAC,IAAa,EAAqC;IACtF,IAAI,CAAC,IAAI,IAAI,OAAO,IAAI,KAAK,QAAQ;QAAE,OAAO,KAAK,CAAC;IACpD,MAAM,MAAM,GAAG,IAA0C,CAAC;IAC1D,OAAO,CACN,eAAe,CAAC,MAAM,CAAC,KAAK,CAAC;QAC7B,MAAM,CAAC,KAAK,CAAC,IAAI,KAAK,UAAU,IAAI,qEAAqE;QACzG,OAAO,MAAM,CAAC,WAAW,KAAK,QAAQ;QACtC,CAAC,MAAM,CAAC,OAAO,KAAK,QAAQ,IAAI,MAAM,CAAC,OAAO,KAAK,WAAW,IAAI,MAAM,CAAC,OAAO,KAAK,QAAQ,CAAC;QAC9F,CAAC,MAAM,CAAC,UAAU,KAAK,SAAS,IAAI,OAAO,MAAM,CAAC,UAAU,KAAK,QAAQ,CAAC;QAC1E,CAAC,MAAM,CAAC,MAAM,KAAK,SAAS,IAAI,MAAM,CAAC,MAAM,KAAK,UAAU,IAAI,MAAM,CAAC,MAAM,KAAK,QAAQ,CAAC,CAC3F,CAAC;AAAA,CACF;AAED,SAAS,cAAc,CAAC,QAAmC,EAAU;IACpE,MAAM,EAAE,IAAI,EAAE,IAAI,EAAE,UAAU,EAAE,GAAG,QAAQ,CAAC,KAAK,CAAC;IAClD,IAAI,WAAW,GAAW,QAAQ,CAAC,OAAO,CAAC;IAC3C,IAAI,QAAQ,CAAC,OAAO,KAAK,WAAW,IAAI,QAAQ,CAAC,UAAU,EAAE,CAAC;QAC7D,WAAW,GAAG,gBAAgB,QAAQ,CAAC,UAAU,EAAE,CAAC;IACrD,CAAC;SAAM,IAAI,QAAQ,CAAC,OAAO,KAAK,QAAQ,EAAE,CAAC;QAC1C,WAAW,GAAG,QAAQ,CAAC;IACxB,CAAC;IACD,OAAO,GAAG,IAAI,IAAI,IAAI,OAAO,QAAQ,CAAC,WAAW,KAAK,UAAU,KAAK,WAAW,GAAG,CAAC;AAAA,CACpF;AAED,MAAM,UAAU,6BAA6B,CAAC,OAAuB,EAAE,KAAK,GAAG,CAAC,EAA+B;IAC9G,MAAM,SAAS,GAAgC,EAAE,CAAC;IAClD,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC7B,IAAI,KAAK,CAAC,IAAI,KAAK,QAAQ,IAAI,KAAK,CAAC,UAAU,KAAK,iCAAiC;YAAE,SAAS;QAChG,IAAI,CAAC,2BAA2B,CAAC,KAAK,CAAC,IAAI,CAAC;YAAE,SAAS;QACvD,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAC5B,CAAC;IACD,OAAO,SAAS,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC;AAAA,CAC/B;AAED,MAAM,UAAU,uBAAuB,CACtC,QAAmC,EACnC,YAAwC,EACxC,WAAW,GAA8B,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,EACzD,eAAe,GAAgC,EAAE,EACjD,cAAuB,EACvB,YAAgC,EACvB;IACT,MAAM,qBAAqB,GAAG,cAAc,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,YAAY,CAAC;IACxE,MAAM,KAAK,GAAG;QACb,GAAG,WAAW,CAAC,SAAS,CAAC,IAAI,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,UAAU,EAAE;QACxE,GAAG,WAAW,CAAC,cAAc,CAAC,IAAI,QAAQ,CAAC,UAAU,IAAI,OAAO,EAAE;QAClE,GAAG,WAAW,CAAC,eAAe,CAAC,IAAI,QAAQ,CAAC,WAAW,IAAI,OAAO,EAAE;QACpE,GAAG,WAAW,CAAC,kBAAkB,CAAC,IAAI,QAAQ,CAAC,cAAc,IAAI,OAAO,EAAE;QAC1E,GAAG,WAAW,CAAC,iBAAiB,CAAC,IAAI,QAAQ,CAAC,aAAa,IAAI,QAAQ,EAAE;KACzE,CAAC;IACF,IAAI,CAAC,QAAQ,CAAC,OAAO,EAAE,CAAC;QACvB,KAAK,CAAC,IAAI,CAAC,GAAG,WAAW,CAAC,UAAU,CAAC,sBAAsB,CAAC,CAAC;IAC9D,CAAC;SAAM,IAAI,cAAc,EAAE,CAAC;QAC3B,KAAK,CAAC,IAAI,CAAC,GAAG,WAAW,CAAC,UAAU,CAAC,aAAa,cAAc,GAAG,CAAC,CAAC;IACtE,CAAC;SAAM,IAAI,qBAAqB,EAAE,CAAC;QAClC,KAAK,CAAC,IAAI,CAAC,GAAG,WAAW,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC;IACjD,CAAC;SAAM,CAAC;QACP,KAAK,CAAC,IAAI,CAAC,GAAG,WAAW,CAAC,UAAU,CAAC,qBAAqB,CAAC,CAAC;IAC7D,CAAC;IACD,KAAK,CAAC,IAAI,CAAC,GAAG,WAAW,CAAC,gBAAgB,CAAC,IAAI,YAAY,IAAI,MAAM,EAAE,CAAC,CAAC;IACzE,IAAI,CAAC,qBAAqB,EAAE,CAAC;QAC5B,KAAK,CAAC,IAAI,CAAC,GAAG,WAAW,CAAC,gBAAgB,CAAC,OAAO,CAAC,CAAC;IACrD,CAAC;SAAM,CAAC;QACP,KAAK,CAAC,IAAI,CAAC,GAAG,WAAW,CAAC,gBAAgB,CAAC,IAAI,cAAc,CAAC,qBAAqB,CAAC,EAAE,CAAC,CAAC;IACzF,CAAC;IACD,IAAI,eAAe,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAChC,KAAK,CAAC,IAAI,CAAC,WAAW,CAAC,mBAAmB,CAAC,CAAC,CAAC;QAC7C,KAAK,MAAM,QAAQ,IAAI,eAAe,EAAE,CAAC;YACxC,KAAK,CAAC,IAAI,CAAC,KAAK,cAAc,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC;QAC7C,CAAC;IACF,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAAA,CACxB","sourcesContent":["import type { ModelTier, RouteDecision, RouteRisk } from \"../autonomy/contracts.ts\";\nimport type { SessionEntry } from \"../session-manager.ts\";\nimport type { ModelRouterIntent } from \"./intent-classifier.ts\";\n\nexport const MODEL_ROUTER_DECISION_CUSTOM_TYPE = \"model_router_decision\";\n\nexport type ModelRouterStatusSettings = {\n\tenabled: boolean;\n\tcheapModel?: string;\n\tmediumModel?: string;\n\texpensiveModel?: string;\n\tlearningModel?: string;\n};\n\nexport type ModelRouterDecisionStatus = {\n\troute: RouteDecision;\n\troutedModel: string;\n\toutcome: \"routed\" | \"escalated\" | \"failed\";\n\tretryModel?: string;\n\tintent?: ModelRouterIntent;\n};\n\nfunction isRouteDecision(value: unknown): value is RouteDecision {\n\tif (!value || typeof value !== \"object\") return false;\n\tconst route = value as Partial<RouteDecision>;\n\tconst validTiers = new Set<ModelTier>([\"cheap\", \"medium\", \"expensive\", \"learning\"]);\n\tconst validRisks = new Set<RouteRisk>([\"read-only\", \"scoped-write\", \"high-impact\", \"approval-required\"]);\n\n\treturn (\n\t\ttypeof route.confidence === \"number\" &&\n\t\ttypeof route.reasonCode === \"string\" &&\n\t\tArray.isArray(route.reasons) &&\n\t\troute.reasons.every((r) => typeof r === \"string\") &&\n\t\tvalidTiers.has(route.tier as ModelTier) &&\n\t\tvalidRisks.has(route.risk as RouteRisk) &&\n\t\t(route.model === undefined || typeof route.model === \"string\") &&\n\t\t(route.fallbackFrom === undefined ||\n\t\t\troute.fallbackFrom === \"cheap\" ||\n\t\t\troute.fallbackFrom === \"medium\" ||\n\t\t\troute.fallbackFrom === \"expensive\" ||\n\t\t\troute.fallbackFrom === \"learning\") &&\n\t\t(route.createdAt === undefined || typeof route.createdAt === \"string\")\n\t);\n}\n\nfunction isModelRouterDecisionStatus(data: unknown): data is ModelRouterDecisionStatus {\n\tif (!data || typeof data !== \"object\") return false;\n\tconst record = data as Partial<ModelRouterDecisionStatus>;\n\treturn (\n\t\tisRouteDecision(record.route) &&\n\t\trecord.route.tier !== \"learning\" && // Validate user prompt decisions never need learning tier in runtime\n\t\ttypeof record.routedModel === \"string\" &&\n\t\t(record.outcome === \"routed\" || record.outcome === \"escalated\" || record.outcome === \"failed\") &&\n\t\t(record.retryModel === undefined || typeof record.retryModel === \"string\") &&\n\t\t(record.intent === undefined || record.intent === \"research\" || record.intent === \"modify\")\n\t);\n}\n\nfunction formatDecision(decision: ModelRouterDecisionStatus): string {\n\tconst { tier, risk, reasonCode } = decision.route;\n\tlet outcomeText: string = decision.outcome;\n\tif (decision.outcome === \"escalated\" && decision.retryModel) {\n\t\toutcomeText = `escalated -> ${decision.retryModel}`;\n\t} else if (decision.outcome === \"failed\") {\n\t\toutcomeText = \"failed\";\n\t}\n\treturn `${tier}/${risk} -> ${decision.routedModel} (${reasonCode}, ${outcomeText})`;\n}\n\nexport function getRecentModelRouterDecisions(entries: SessionEntry[], limit = 3): ModelRouterDecisionStatus[] {\n\tconst decisions: ModelRouterDecisionStatus[] = [];\n\tfor (const entry of entries) {\n\t\tif (entry.type !== \"custom\" || entry.customType !== MODEL_ROUTER_DECISION_CUSTOM_TYPE) continue;\n\t\tif (!isModelRouterDecisionStatus(entry.data)) continue;\n\t\tdecisions.push(entry.data);\n\t}\n\treturn decisions.slice(-limit);\n}\n\nexport function formatModelRouterStatus(\n\tsettings: ModelRouterStatusSettings,\n\tlastDecision?: ModelRouterDecisionStatus,\n\tformatLabel: (label: string) => string = (label) => label,\n\trecentDecisions: ModelRouterDecisionStatus[] = [],\n\tlastSkipReason?: string,\n\tlatestIntent?: ModelRouterIntent,\n): string {\n\tconst effectiveLastDecision = lastSkipReason ? undefined : lastDecision;\n\tconst lines = [\n\t\t`${formatLabel(\"Status:\")} ${settings.enabled ? \"enabled\" : \"disabled\"}`,\n\t\t`${formatLabel(\"Cheap model:\")} ${settings.cheapModel ?? \"unset\"}`,\n\t\t`${formatLabel(\"Medium model:\")} ${settings.mediumModel ?? \"unset\"}`,\n\t\t`${formatLabel(\"Expensive model:\")} ${settings.expensiveModel ?? \"unset\"}`,\n\t\t`${formatLabel(\"Learning model:\")} ${settings.learningModel ?? \"active\"}`,\n\t];\n\tif (!settings.enabled) {\n\t\tlines.push(`${formatLabel(\"Routing:\")} inactive (disabled)`);\n\t} else if (lastSkipReason) {\n\t\tlines.push(`${formatLabel(\"Routing:\")} skipped (${lastSkipReason})`);\n\t} else if (effectiveLastDecision) {\n\t\tlines.push(`${formatLabel(\"Routing:\")} active`);\n\t} else {\n\t\tlines.push(`${formatLabel(\"Routing:\")} waiting for prompt`);\n\t}\n\tlines.push(`${formatLabel(\"Latest intent:\")} ${latestIntent ?? \"none\"}`);\n\tif (!effectiveLastDecision) {\n\t\tlines.push(`${formatLabel(\"Last decision:\")} none`);\n\t} else {\n\t\tlines.push(`${formatLabel(\"Last decision:\")} ${formatDecision(effectiveLastDecision)}`);\n\t}\n\tif (recentDecisions.length > 0) {\n\t\tlines.push(formatLabel(\"Recent decisions:\"));\n\t\tfor (const decision of recentDecisions) {\n\t\t\tlines.push(`- ${formatDecision(decision)}`);\n\t\t}\n\t}\n\treturn lines.join(\"\\n\");\n}\n"]}

package/dist/core/model-router/tool-escalation.d.ts

@@ -1,9 +1,7 @@
-import type { ModelRouterIntent } from "./intent-classifier.ts";
-type ToolEscalationOptions = {
-    intent: ModelRouterIntent;
+import type { ModelTier } from "../autonomy/contracts.ts";
+export declare function shouldEscalateModelRouterTool(options: {
+    tier: ModelTier;
     toolName: string;
     args?: unknown;
-};
-export declare function shouldEscalateModelRouterTool(options: ToolEscalationOptions): boolean;
-export {};
+}): boolean;
 //# sourceMappingURL=tool-escalation.d.ts.map

package/dist/core/model-router/tool-escalation.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"tool-escalation.d.ts","sourceRoot":"","sources":["../../../src/core/model-router/tool-escalation.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,wBAAwB,CAAC;AAqDhE,KAAK,qBAAqB,GAAG;IAAE,MAAM,EAAE,iBAAiB,CAAC;IAAC,QAAQ,EAAE,MAAM,CAAC;IAAC,IAAI,CAAC,EAAE,OAAO,CAAA;CAAE,CAAC;AAwC7F,wBAAgB,6BAA6B,CAAC,OAAO,EAAE,qBAAqB,GAAG,OAAO,CAUrF","sourcesContent":["import type { ModelRouterIntent } from \"./intent-classifier.ts\";\n\nconst READ_ONLY_TOOL_NAMES = new Set([\n\t\"read\",\n\t\"grep\",\n\t\"find\",\n\t\"ls\",\n\t\"list\",\n\t\"search\",\n\t\"glob\",\n\t\"view_file\",\n\t\"list_dir\",\n\t\"grep_search\",\n\t\"search_web\",\n\t\"read_url_content\",\n\t\"read_browser_page\",\n]);\n\nconst SHELL_TOOL_NAMES = new Set([\"bash\", \"exec\", \"execute\", \"run\", \"run_command\", \"shell\"]);\n\nconst READ_ONLY_COMMANDS = new Set([\n\t\"awk\",\n\t\"cat\",\n\t\"date\",\n\t\"df\",\n\t\"du\",\n\t\"env\",\n\t\"git\",\n\t\"grep\",\n\t\"head\",\n\t\"jq\",\n\t\"ls\",\n\t\"node\",\n\t\"npm\",\n\t\"pnpm\",\n\t\"pwd\",\n\t\"rg\",\n\t\"sed\",\n\t\"tail\",\n\t\"test\",\n\t\"tsc\",\n\t\"wc\",\n\t\"which\",\n\t\"yarn\",\n]);\n\nconst READ_ONLY_GIT_SUBCOMMANDS = new Set([\"branch\", \"diff\", \"log\", \"rev-parse\", \"show\", \"status\", \"tag\"]);\nconst READ_ONLY_NPM_SUBCOMMANDS = new Set([\"info\", \"list\", \"ls\", \"outdated\", \"view\", \"whoami\"]);\nconst MUTATING_SHELL_TOKEN_RE =\n\t/(^|\\s)(>|>>|2>|&>|tee\\b|rm\\b|mv\\b|cp\\b|mkdir\\b|touch\\b|chmod\\b|chown\\b|install\\b|commit\\b|push\\b|publish\\b|deploy\\b|apply\\b|add\\b|checkout\\b|switch\\b|reset\\b|clean\\b|stash\\b|merge\\b|rebase\\b|npm\\s+(?:i|install|ci|update|publish|run)\\b|pnpm\\s+(?:i|install|update|publish|run)\\b|yarn\\s+(?:add|install|upgrade|publish|run)\\b)/i;\nconst MUTATING_TOOL_NAME_RE =\n\t/(bash|exec|execute|run|shell|write|edit|patch|replace|delete|remove|move|rename|create|mkdir|touch|install|commit|push|publish|deploy|apply)/i;\n\ntype ToolEscalationOptions = { intent: ModelRouterIntent; toolName: string; args?: unknown };\n\nfunction getShellCommand(args: unknown): string | undefined {\n\tif (!args || typeof args !== \"object\") return undefined;\n\tconst record = args as Record<string, unknown>;\n\tconst command = record.command ?? record.cmd ?? record.shellCommand;\n\treturn typeof command === \"string\" ? command.trim() : undefined;\n}\n\nfunction commandName(segment: string): string | undefined {\n\tconst first = segment.trim().match(/^[A-Za-z0-9_./-]+/)?.[0];\n\tif (!first) return undefined;\n\tconst parts = first.split(\"/\");\n\treturn parts[parts.length - 1]?.toLowerCase();\n}\n\nfunction commandArg(segment: string, index: number): string | undefined {\n\treturn segment.trim().split(/\\s+/)[index]?.toLowerCase();\n}\n\nfunction isReadOnlyShellSegment(segment: string): boolean {\n\tconst name = commandName(segment);\n\tif (!name || !READ_ONLY_COMMANDS.has(name)) return false;\n\tif (name === \"git\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_GIT_SUBCOMMANDS.has(subcommand));\n\t}\n\tif (name === \"npm\" || name === \"pnpm\" || name === \"yarn\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_NPM_SUBCOMMANDS.has(subcommand));\n\t}\n\treturn true;\n}\n\nfunction isReadOnlyShellCommand(command: string): boolean {\n\tif (!command || MUTATING_SHELL_TOKEN_RE.test(command)) return false;\n\tconst segments = command.split(/\\s*&&\\s*/).map((segment) => segment.trim());\n\treturn segments.length > 0 && segments.every(isReadOnlyShellSegment);\n}\n\nexport function shouldEscalateModelRouterTool(options: ToolEscalationOptions): boolean {\n\tif (options.intent !== \"research\") return false;\n\tconst toolName = options.toolName.trim().toLowerCase();\n\tif (!toolName) return true;\n\tif (READ_ONLY_TOOL_NAMES.has(toolName)) return false;\n\tif (SHELL_TOOL_NAMES.has(toolName)) {\n\t\tconst command = getShellCommand(options.args);\n\t\treturn command ? !isReadOnlyShellCommand(command) : true;\n\t}\n\treturn MUTATING_TOOL_NAME_RE.test(toolName) || !toolName.startsWith(\"read_\");\n}\n"]}
+{"version":3,"file":"tool-escalation.d.ts","sourceRoot":"","sources":["../../../src/core/model-router/tool-escalation.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,0BAA0B,CAAC;AA2F1D,wBAAgB,6BAA6B,CAAC,OAAO,EAAE;IAAE,IAAI,EAAE,SAAS,CAAC;IAAC,QAAQ,EAAE,MAAM,CAAC;IAAC,IAAI,CAAC,EAAE,OAAO,CAAA;CAAE,GAAG,OAAO,CAUrH","sourcesContent":["import type { ModelTier } from \"../autonomy/contracts.ts\";\n\nconst READ_ONLY_TOOL_NAMES = new Set([\n\t\"read\",\n\t\"grep\",\n\t\"find\",\n\t\"ls\",\n\t\"list\",\n\t\"search\",\n\t\"glob\",\n\t\"view_file\",\n\t\"list_dir\",\n\t\"grep_search\",\n\t\"search_web\",\n\t\"read_url_content\",\n\t\"read_browser_page\",\n]);\n\nconst SHELL_TOOL_NAMES = new Set([\"bash\", \"exec\", \"execute\", \"run\", \"run_command\", \"shell\"]);\n\nconst READ_ONLY_COMMANDS = new Set([\n\t\"awk\",\n\t\"cat\",\n\t\"date\",\n\t\"df\",\n\t\"du\",\n\t\"env\",\n\t\"git\",\n\t\"grep\",\n\t\"head\",\n\t\"jq\",\n\t\"ls\",\n\t\"node\",\n\t\"npm\",\n\t\"pnpm\",\n\t\"pwd\",\n\t\"rg\",\n\t\"sed\",\n\t\"tail\",\n\t\"test\",\n\t\"tsc\",\n\t\"wc\",\n\t\"which\",\n\t\"yarn\",\n]);\n\nconst READ_ONLY_GIT_SUBCOMMANDS = new Set([\"branch\", \"diff\", \"log\", \"rev-parse\", \"show\", \"status\", \"tag\"]);\nconst READ_ONLY_NPM_SUBCOMMANDS = new Set([\"info\", \"list\", \"ls\", \"outdated\", \"view\", \"whoami\"]);\nconst MUTATING_SHELL_TOKEN_RE =\n\t/(^|\\s)(>|>>|2>|&>|tee\\b|rm\\b|mv\\b|cp\\b|mkdir\\b|touch\\b|chmod\\b|chown\\b|install\\b|commit\\b|push\\b|publish\\b|deploy\\b|apply\\b|add\\b|checkout\\b|switch\\b|reset\\b|clean\\b|stash\\b|merge\\b|rebase\\b|npm\\s+(?:i|install|ci|update|publish|run)\\b|pnpm\\s+(?:i|install|update|publish|run)\\b|yarn\\s+(?:add|install|upgrade|publish|run)\\b)/i;\nconst MUTATING_TOOL_NAME_RE =\n\t/(bash|exec|execute|run|shell|write|edit|patch|replace|delete|remove|move|rename|create|mkdir|touch|install|commit|push|publish|deploy|apply)/i;\n\nfunction getShellCommand(args: unknown): string | undefined {\n\tif (!args || typeof args !== \"object\") return undefined;\n\tconst record = args as Record<string, unknown>;\n\tconst command = record.command ?? record.cmd ?? record.shellCommand;\n\treturn typeof command === \"string\" ? command.trim() : undefined;\n}\n\nfunction commandName(segment: string): string | undefined {\n\tconst first = segment.trim().match(/^[A-Za-z0-9_./-]+/)?.[0];\n\tif (!first) return undefined;\n\tconst parts = first.split(\"/\");\n\treturn parts[parts.length - 1]?.toLowerCase();\n}\n\nfunction commandArg(segment: string, index: number): string | undefined {\n\treturn segment.trim().split(/\\s+/)[index]?.toLowerCase();\n}\n\nfunction isReadOnlyShellSegment(segment: string): boolean {\n\tconst name = commandName(segment);\n\tif (!name || !READ_ONLY_COMMANDS.has(name)) return false;\n\tif (name === \"git\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_GIT_SUBCOMMANDS.has(subcommand));\n\t}\n\tif (name === \"npm\" || name === \"pnpm\" || name === \"yarn\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_NPM_SUBCOMMANDS.has(subcommand));\n\t}\n\treturn true;\n}\n\nfunction isReadOnlyShellCommand(command: string): boolean {\n\tif (!command || MUTATING_SHELL_TOKEN_RE.test(command)) return false;\n\tconst segments = command.split(/\\s*&&\\s*/).map((segment) => segment.trim());\n\treturn segments.length > 0 && segments.every(isReadOnlyShellSegment);\n}\n\nexport function shouldEscalateModelRouterTool(options: { tier: ModelTier; toolName: string; args?: unknown }): boolean {\n\tif (options.tier !== \"cheap\") return false;\n\tconst toolName = options.toolName.trim().toLowerCase();\n\tif (!toolName) return true;\n\tif (READ_ONLY_TOOL_NAMES.has(toolName)) return false;\n\tif (SHELL_TOOL_NAMES.has(toolName)) {\n\t\tconst command = getShellCommand(options.args);\n\t\treturn command ? !isReadOnlyShellCommand(command) : true;\n\t}\n\treturn MUTATING_TOOL_NAME_RE.test(toolName) || !toolName.startsWith(\"read_\");\n}\n"]}

package/dist/core/model-router/tool-escalation.js

@@ -81,7 +81,7 @@ function isReadOnlyShellCommand(command) {
     return segments.length > 0 && segments.every(isReadOnlyShellSegment);
 }
 export function shouldEscalateModelRouterTool(options) {
-    if (options.intent !== "research")
+    if (options.tier !== "cheap")
         return false;
     const toolName = options.toolName.trim().toLowerCase();
     if (!toolName)

package/dist/core/model-router/tool-escalation.js.map

@@ -1 +1 @@
-{"version":3,"file":"tool-escalation.js","sourceRoot":"","sources":["../../../src/core/model-router/tool-escalation.ts"],"names":[],"mappings":"AAEA,MAAM,oBAAoB,GAAG,IAAI,GAAG,CAAC;IACpC,MAAM;IACN,MAAM;IACN,MAAM;IACN,IAAI;IACJ,MAAM;IACN,QAAQ;IACR,MAAM;IACN,WAAW;IACX,UAAU;IACV,aAAa;IACb,YAAY;IACZ,kBAAkB;IAClB,mBAAmB;CACnB,CAAC,CAAC;AAEH,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAAC,CAAC,MAAM,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,EAAE,aAAa,EAAE,OAAO,CAAC,CAAC,CAAC;AAE7F,MAAM,kBAAkB,GAAG,IAAI,GAAG,CAAC;IAClC,KAAK;IACL,KAAK;IACL,MAAM;IACN,IAAI;IACJ,IAAI;IACJ,KAAK;IACL,KAAK;IACL,MAAM;IACN,MAAM;IACN,IAAI;IACJ,IAAI;IACJ,MAAM;IACN,KAAK;IACL,MAAM;IACN,KAAK;IACL,IAAI;IACJ,KAAK;IACL,MAAM;IACN,MAAM;IACN,KAAK;IACL,IAAI;IACJ,OAAO;IACP,MAAM;CACN,CAAC,CAAC;AAEH,MAAM,yBAAyB,GAAG,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,WAAW,EAAE,MAAM,EAAE,QAAQ,EAAE,KAAK,CAAC,CAAC,CAAC;AAC3G,MAAM,yBAAyB,GAAG,IAAI,GAAG,CAAC,CAAC,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,QAAQ,CAAC,CAAC,CAAC;AAChG,MAAM,uBAAuB,GAC5B,qUAAqU,CAAC;AACvU,MAAM,qBAAqB,GAC1B,+IAA+I,CAAC;AAIjJ,SAAS,eAAe,CAAC,IAAa,EAAsB;IAC3D,IAAI,CAAC,IAAI,IAAI,OAAO,IAAI,KAAK,QAAQ;QAAE,OAAO,SAAS,CAAC;IACxD,MAAM,MAAM,GAAG,IAA+B,CAAC;IAC/C,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC,GAAG,IAAI,MAAM,CAAC,YAAY,CAAC;IACpE,OAAO,OAAO,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;AAAA,CAChE;AAED,SAAS,WAAW,CAAC,OAAe,EAAsB;IACzD,MAAM,KAAK,GAAG,OAAO,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,mBAAmB,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;IAC7D,IAAI,CAAC,KAAK;QAAE,OAAO,SAAS,CAAC;IAC7B,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IAC/B,OAAO,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC;AAAA,CAC9C;AAED,SAAS,UAAU,CAAC,OAAe,EAAE,KAAa,EAAsB;IACvE,OAAO,OAAO,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,EAAE,WAAW,EAAE,CAAC;AAAA,CACzD;AAED,SAAS,sBAAsB,CAAC,OAAe,EAAW;IACzD,MAAM,IAAI,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC;IAClC,IAAI,CAAC,IAAI,IAAI,CAAC,kBAAkB,CAAC,GAAG,CAAC,IAAI,CAAC;QAAE,OAAO,KAAK,CAAC;IACzD,IAAI,IAAI,KAAK,KAAK,EAAE,CAAC;QACpB,MAAM,UAAU,GAAG,UAAU,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QAC1C,OAAO,OAAO,CAAC,UAAU,IAAI,yBAAyB,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC;IACzE,CAAC;IACD,IAAI,IAAI,KAAK,KAAK,IAAI,IAAI,KAAK,MAAM,IAAI,IAAI,KAAK,MAAM,EAAE,CAAC;QAC1D,MAAM,UAAU,GAAG,UAAU,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QAC1C,OAAO,OAAO,CAAC,UAAU,IAAI,yBAAyB,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC;IACzE,CAAC;IACD,OAAO,IAAI,CAAC;AAAA,CACZ;AAED,SAAS,sBAAsB,CAAC,OAAe,EAAW;IACzD,IAAI,CAAC,OAAO,IAAI,uBAAuB,CAAC,IAAI,CAAC,OAAO,CAAC;QAAE,OAAO,KAAK,CAAC;IACpE,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;IAC5E,OAAO,QAAQ,CAAC,MAAM,GAAG,CAAC,IAAI,QAAQ,CAAC,KAAK,CAAC,sBAAsB,CAAC,CAAC;AAAA,CACrE;AAED,MAAM,UAAU,6BAA6B,CAAC,OAA8B,EAAW;IACtF,IAAI,OAAO,CAAC,MAAM,KAAK,UAAU;QAAE,OAAO,KAAK,CAAC;IAChD,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IACvD,IAAI,CAAC,QAAQ;QAAE,OAAO,IAAI,CAAC;IAC3B,IAAI,oBAAoB,CAAC,GAAG,CAAC,QAAQ,CAAC;QAAE,OAAO,KAAK,CAAC;IACrD,IAAI,gBAAgB,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;QACpC,MAAM,OAAO,GAAG,eAAe,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;QAC9C,OAAO,OAAO,CAAC,CAAC,CAAC,CAAC,sBAAsB,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAC1D,CAAC;IACD,OAAO,qBAAqB,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,QAAQ,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC;AAAA,CAC7E","sourcesContent":["import type { ModelRouterIntent } from \"./intent-classifier.ts\";\n\nconst READ_ONLY_TOOL_NAMES = new Set([\n\t\"read\",\n\t\"grep\",\n\t\"find\",\n\t\"ls\",\n\t\"list\",\n\t\"search\",\n\t\"glob\",\n\t\"view_file\",\n\t\"list_dir\",\n\t\"grep_search\",\n\t\"search_web\",\n\t\"read_url_content\",\n\t\"read_browser_page\",\n]);\n\nconst SHELL_TOOL_NAMES = new Set([\"bash\", \"exec\", \"execute\", \"run\", \"run_command\", \"shell\"]);\n\nconst READ_ONLY_COMMANDS = new Set([\n\t\"awk\",\n\t\"cat\",\n\t\"date\",\n\t\"df\",\n\t\"du\",\n\t\"env\",\n\t\"git\",\n\t\"grep\",\n\t\"head\",\n\t\"jq\",\n\t\"ls\",\n\t\"node\",\n\t\"npm\",\n\t\"pnpm\",\n\t\"pwd\",\n\t\"rg\",\n\t\"sed\",\n\t\"tail\",\n\t\"test\",\n\t\"tsc\",\n\t\"wc\",\n\t\"which\",\n\t\"yarn\",\n]);\n\nconst READ_ONLY_GIT_SUBCOMMANDS = new Set([\"branch\", \"diff\", \"log\", \"rev-parse\", \"show\", \"status\", \"tag\"]);\nconst READ_ONLY_NPM_SUBCOMMANDS = new Set([\"info\", \"list\", \"ls\", \"outdated\", \"view\", \"whoami\"]);\nconst MUTATING_SHELL_TOKEN_RE =\n\t/(^|\\s)(>|>>|2>|&>|tee\\b|rm\\b|mv\\b|cp\\b|mkdir\\b|touch\\b|chmod\\b|chown\\b|install\\b|commit\\b|push\\b|publish\\b|deploy\\b|apply\\b|add\\b|checkout\\b|switch\\b|reset\\b|clean\\b|stash\\b|merge\\b|rebase\\b|npm\\s+(?:i|install|ci|update|publish|run)\\b|pnpm\\s+(?:i|install|update|publish|run)\\b|yarn\\s+(?:add|install|upgrade|publish|run)\\b)/i;\nconst MUTATING_TOOL_NAME_RE =\n\t/(bash|exec|execute|run|shell|write|edit|patch|replace|delete|remove|move|rename|create|mkdir|touch|install|commit|push|publish|deploy|apply)/i;\n\ntype ToolEscalationOptions = { intent: ModelRouterIntent; toolName: string; args?: unknown };\n\nfunction getShellCommand(args: unknown): string | undefined {\n\tif (!args || typeof args !== \"object\") return undefined;\n\tconst record = args as Record<string, unknown>;\n\tconst command = record.command ?? record.cmd ?? record.shellCommand;\n\treturn typeof command === \"string\" ? command.trim() : undefined;\n}\n\nfunction commandName(segment: string): string | undefined {\n\tconst first = segment.trim().match(/^[A-Za-z0-9_./-]+/)?.[0];\n\tif (!first) return undefined;\n\tconst parts = first.split(\"/\");\n\treturn parts[parts.length - 1]?.toLowerCase();\n}\n\nfunction commandArg(segment: string, index: number): string | undefined {\n\treturn segment.trim().split(/\\s+/)[index]?.toLowerCase();\n}\n\nfunction isReadOnlyShellSegment(segment: string): boolean {\n\tconst name = commandName(segment);\n\tif (!name || !READ_ONLY_COMMANDS.has(name)) return false;\n\tif (name === \"git\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_GIT_SUBCOMMANDS.has(subcommand));\n\t}\n\tif (name === \"npm\" || name === \"pnpm\" || name === \"yarn\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_NPM_SUBCOMMANDS.has(subcommand));\n\t}\n\treturn true;\n}\n\nfunction isReadOnlyShellCommand(command: string): boolean {\n\tif (!command || MUTATING_SHELL_TOKEN_RE.test(command)) return false;\n\tconst segments = command.split(/\\s*&&\\s*/).map((segment) => segment.trim());\n\treturn segments.length > 0 && segments.every(isReadOnlyShellSegment);\n}\n\nexport function shouldEscalateModelRouterTool(options: ToolEscalationOptions): boolean {\n\tif (options.intent !== \"research\") return false;\n\tconst toolName = options.toolName.trim().toLowerCase();\n\tif (!toolName) return true;\n\tif (READ_ONLY_TOOL_NAMES.has(toolName)) return false;\n\tif (SHELL_TOOL_NAMES.has(toolName)) {\n\t\tconst command = getShellCommand(options.args);\n\t\treturn command ? !isReadOnlyShellCommand(command) : true;\n\t}\n\treturn MUTATING_TOOL_NAME_RE.test(toolName) || !toolName.startsWith(\"read_\");\n}\n"]}
+{"version":3,"file":"tool-escalation.js","sourceRoot":"","sources":["../../../src/core/model-router/tool-escalation.ts"],"names":[],"mappings":"AAEA,MAAM,oBAAoB,GAAG,IAAI,GAAG,CAAC;IACpC,MAAM;IACN,MAAM;IACN,MAAM;IACN,IAAI;IACJ,MAAM;IACN,QAAQ;IACR,MAAM;IACN,WAAW;IACX,UAAU;IACV,aAAa;IACb,YAAY;IACZ,kBAAkB;IAClB,mBAAmB;CACnB,CAAC,CAAC;AAEH,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAAC,CAAC,MAAM,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,EAAE,aAAa,EAAE,OAAO,CAAC,CAAC,CAAC;AAE7F,MAAM,kBAAkB,GAAG,IAAI,GAAG,CAAC;IAClC,KAAK;IACL,KAAK;IACL,MAAM;IACN,IAAI;IACJ,IAAI;IACJ,KAAK;IACL,KAAK;IACL,MAAM;IACN,MAAM;IACN,IAAI;IACJ,IAAI;IACJ,MAAM;IACN,KAAK;IACL,MAAM;IACN,KAAK;IACL,IAAI;IACJ,KAAK;IACL,MAAM;IACN,MAAM;IACN,KAAK;IACL,IAAI;IACJ,OAAO;IACP,MAAM;CACN,CAAC,CAAC;AAEH,MAAM,yBAAyB,GAAG,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,WAAW,EAAE,MAAM,EAAE,QAAQ,EAAE,KAAK,CAAC,CAAC,CAAC;AAC3G,MAAM,yBAAyB,GAAG,IAAI,GAAG,CAAC,CAAC,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,QAAQ,CAAC,CAAC,CAAC;AAChG,MAAM,uBAAuB,GAC5B,qUAAqU,CAAC;AACvU,MAAM,qBAAqB,GAC1B,+IAA+I,CAAC;AAEjJ,SAAS,eAAe,CAAC,IAAa,EAAsB;IAC3D,IAAI,CAAC,IAAI,IAAI,OAAO,IAAI,KAAK,QAAQ;QAAE,OAAO,SAAS,CAAC;IACxD,MAAM,MAAM,GAAG,IAA+B,CAAC;IAC/C,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC,GAAG,IAAI,MAAM,CAAC,YAAY,CAAC;IACpE,OAAO,OAAO,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;AAAA,CAChE;AAED,SAAS,WAAW,CAAC,OAAe,EAAsB;IACzD,MAAM,KAAK,GAAG,OAAO,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,mBAAmB,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;IAC7D,IAAI,CAAC,KAAK;QAAE,OAAO,SAAS,CAAC;IAC7B,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IAC/B,OAAO,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC;AAAA,CAC9C;AAED,SAAS,UAAU,CAAC,OAAe,EAAE,KAAa,EAAsB;IACvE,OAAO,OAAO,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,EAAE,WAAW,EAAE,CAAC;AAAA,CACzD;AAED,SAAS,sBAAsB,CAAC,OAAe,EAAW;IACzD,MAAM,IAAI,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC;IAClC,IAAI,CAAC,IAAI,IAAI,CAAC,kBAAkB,CAAC,GAAG,CAAC,IAAI,CAAC;QAAE,OAAO,KAAK,CAAC;IACzD,IAAI,IAAI,KAAK,KAAK,EAAE,CAAC;QACpB,MAAM,UAAU,GAAG,UAAU,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QAC1C,OAAO,OAAO,CAAC,UAAU,IAAI,yBAAyB,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC;IACzE,CAAC;IACD,IAAI,IAAI,KAAK,KAAK,IAAI,IAAI,KAAK,MAAM,IAAI,IAAI,KAAK,MAAM,EAAE,CAAC;QAC1D,MAAM,UAAU,GAAG,UAAU,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QAC1C,OAAO,OAAO,CAAC,UAAU,IAAI,yBAAyB,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC;IACzE,CAAC;IACD,OAAO,IAAI,CAAC;AAAA,CACZ;AAED,SAAS,sBAAsB,CAAC,OAAe,EAAW;IACzD,IAAI,CAAC,OAAO,IAAI,uBAAuB,CAAC,IAAI,CAAC,OAAO,CAAC;QAAE,OAAO,KAAK,CAAC;IACpE,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;IAC5E,OAAO,QAAQ,CAAC,MAAM,GAAG,CAAC,IAAI,QAAQ,CAAC,KAAK,CAAC,sBAAsB,CAAC,CAAC;AAAA,CACrE;AAED,MAAM,UAAU,6BAA6B,CAAC,OAA8D,EAAW;IACtH,IAAI,OAAO,CAAC,IAAI,KAAK,OAAO;QAAE,OAAO,KAAK,CAAC;IAC3C,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IACvD,IAAI,CAAC,QAAQ;QAAE,OAAO,IAAI,CAAC;IAC3B,IAAI,oBAAoB,CAAC,GAAG,CAAC,QAAQ,CAAC;QAAE,OAAO,KAAK,CAAC;IACrD,IAAI,gBAAgB,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;QACpC,MAAM,OAAO,GAAG,eAAe,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;QAC9C,OAAO,OAAO,CAAC,CAAC,CAAC,CAAC,sBAAsB,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAC1D,CAAC;IACD,OAAO,qBAAqB,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,QAAQ,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC;AAAA,CAC7E","sourcesContent":["import type { ModelTier } from \"../autonomy/contracts.ts\";\n\nconst READ_ONLY_TOOL_NAMES = new Set([\n\t\"read\",\n\t\"grep\",\n\t\"find\",\n\t\"ls\",\n\t\"list\",\n\t\"search\",\n\t\"glob\",\n\t\"view_file\",\n\t\"list_dir\",\n\t\"grep_search\",\n\t\"search_web\",\n\t\"read_url_content\",\n\t\"read_browser_page\",\n]);\n\nconst SHELL_TOOL_NAMES = new Set([\"bash\", \"exec\", \"execute\", \"run\", \"run_command\", \"shell\"]);\n\nconst READ_ONLY_COMMANDS = new Set([\n\t\"awk\",\n\t\"cat\",\n\t\"date\",\n\t\"df\",\n\t\"du\",\n\t\"env\",\n\t\"git\",\n\t\"grep\",\n\t\"head\",\n\t\"jq\",\n\t\"ls\",\n\t\"node\",\n\t\"npm\",\n\t\"pnpm\",\n\t\"pwd\",\n\t\"rg\",\n\t\"sed\",\n\t\"tail\",\n\t\"test\",\n\t\"tsc\",\n\t\"wc\",\n\t\"which\",\n\t\"yarn\",\n]);\n\nconst READ_ONLY_GIT_SUBCOMMANDS = new Set([\"branch\", \"diff\", \"log\", \"rev-parse\", \"show\", \"status\", \"tag\"]);\nconst READ_ONLY_NPM_SUBCOMMANDS = new Set([\"info\", \"list\", \"ls\", \"outdated\", \"view\", \"whoami\"]);\nconst MUTATING_SHELL_TOKEN_RE =\n\t/(^|\\s)(>|>>|2>|&>|tee\\b|rm\\b|mv\\b|cp\\b|mkdir\\b|touch\\b|chmod\\b|chown\\b|install\\b|commit\\b|push\\b|publish\\b|deploy\\b|apply\\b|add\\b|checkout\\b|switch\\b|reset\\b|clean\\b|stash\\b|merge\\b|rebase\\b|npm\\s+(?:i|install|ci|update|publish|run)\\b|pnpm\\s+(?:i|install|update|publish|run)\\b|yarn\\s+(?:add|install|upgrade|publish|run)\\b)/i;\nconst MUTATING_TOOL_NAME_RE =\n\t/(bash|exec|execute|run|shell|write|edit|patch|replace|delete|remove|move|rename|create|mkdir|touch|install|commit|push|publish|deploy|apply)/i;\n\nfunction getShellCommand(args: unknown): string | undefined {\n\tif (!args || typeof args !== \"object\") return undefined;\n\tconst record = args as Record<string, unknown>;\n\tconst command = record.command ?? record.cmd ?? record.shellCommand;\n\treturn typeof command === \"string\" ? command.trim() : undefined;\n}\n\nfunction commandName(segment: string): string | undefined {\n\tconst first = segment.trim().match(/^[A-Za-z0-9_./-]+/)?.[0];\n\tif (!first) return undefined;\n\tconst parts = first.split(\"/\");\n\treturn parts[parts.length - 1]?.toLowerCase();\n}\n\nfunction commandArg(segment: string, index: number): string | undefined {\n\treturn segment.trim().split(/\\s+/)[index]?.toLowerCase();\n}\n\nfunction isReadOnlyShellSegment(segment: string): boolean {\n\tconst name = commandName(segment);\n\tif (!name || !READ_ONLY_COMMANDS.has(name)) return false;\n\tif (name === \"git\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_GIT_SUBCOMMANDS.has(subcommand));\n\t}\n\tif (name === \"npm\" || name === \"pnpm\" || name === \"yarn\") {\n\t\tconst subcommand = commandArg(segment, 1);\n\t\treturn Boolean(subcommand && READ_ONLY_NPM_SUBCOMMANDS.has(subcommand));\n\t}\n\treturn true;\n}\n\nfunction isReadOnlyShellCommand(command: string): boolean {\n\tif (!command || MUTATING_SHELL_TOKEN_RE.test(command)) return false;\n\tconst segments = command.split(/\\s*&&\\s*/).map((segment) => segment.trim());\n\treturn segments.length > 0 && segments.every(isReadOnlyShellSegment);\n}\n\nexport function shouldEscalateModelRouterTool(options: { tier: ModelTier; toolName: string; args?: unknown }): boolean {\n\tif (options.tier !== \"cheap\") return false;\n\tconst toolName = options.toolName.trim().toLowerCase();\n\tif (!toolName) return true;\n\tif (READ_ONLY_TOOL_NAMES.has(toolName)) return false;\n\tif (SHELL_TOOL_NAMES.has(toolName)) {\n\t\tconst command = getShellCommand(options.args);\n\t\treturn command ? !isReadOnlyShellCommand(command) : true;\n\t}\n\treturn MUTATING_TOOL_NAME_RE.test(toolName) || !toolName.startsWith(\"read_\");\n}\n"]}

package/dist/core/models/fitness-store.d.ts

@@ -0,0 +1,40 @@
+import type { ModelFitnessReport } from "../research/model-fitness.ts";
+/**
+ * Durable, HOST-KEYED storage for model fitness reports. Fitness is a property of a model ON a
+ * host (tok/s and latency-driven failures do not travel between machines), so reports are keyed
+ * by a hardware fingerprint: the same model can be "the heavy lifter" on one machine and
+ * "waiting for better hardware" on another, and pi remembers both without confusing them —
+ * including when settings/dotfiles are synced across machines.
+ */
+export interface HostFingerprint {
+    /** Stable, human-readable id derived from the specs below. */
+    id: string;
+    cpu: string;
+    cores: number;
+    totalMemGb: number;
+}
+export declare function currentHostFingerprint(): HostFingerprint;
+export interface StoredFitnessReport {
+    model: string;
+    report: ModelFitnessReport;
+    at: string;
+    host: HostFingerprint;
+}
+export declare class FitnessStore {
+    private readonly filePath;
+    private readonly fingerprint;
+    constructor(filePath: string, options?: {
+        fingerprint?: () => HostFingerprint;
+    });
+    static forAgentDir(agentDir: string, options?: {
+        fingerprint?: () => HostFingerprint;
+    }): FitnessStore;
+    private load;
+    /** Persist the latest report for a model on the CURRENT host. Best-effort, returns the entry. */
+    save(model: string, report: ModelFitnessReport, at?: string): StoredFitnessReport;
+    /** Reports for the current host (default) or an explicit host id. */
+    getForHost(hostId?: string): StoredFitnessReport[];
+    /** Every stored report across all hosts (for cross-machine comparisons). */
+    getAll(): StoredFitnessReport[];
+}
+//# sourceMappingURL=fitness-store.d.ts.map

package/dist/core/models/fitness-store.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"fitness-store.d.ts","sourceRoot":"","sources":["../../../src/core/models/fitness-store.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,8BAA8B,CAAC;AAEvE;;;;;;GAMG;AAEH,MAAM,WAAW,eAAe;IAC/B,8DAA8D;IAC9D,EAAE,EAAE,MAAM,CAAC;IACX,GAAG,EAAE,MAAM,CAAC;IACZ,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;CACnB;AAED,wBAAgB,sBAAsB,IAAI,eAAe,CAWxD;AAED,MAAM,WAAW,mBAAmB;IACnC,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,kBAAkB,CAAC;IAC3B,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,eAAe,CAAC;CACtB;AAQD,qBAAa,YAAY;IACxB,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAS;IAClC,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAwB;IAEpD,YAAY,QAAQ,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,WAAW,CAAC,EAAE,MAAM,eAAe,CAAA;KAAE,EAG9E;IAED,MAAM,CAAC,WAAW,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,WAAW,CAAC,EAAE,MAAM,eAAe,CAAA;KAAE,GAAG,YAAY,CAEpG;IAED,OAAO,CAAC,IAAI;IAaZ,iGAAiG;IACjG,IAAI,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,kBAAkB,EAAE,EAAE,CAAC,EAAE,MAAM,GAAG,mBAAmB,CAQhF;IAED,qEAAqE;IACrE,UAAU,CAAC,MAAM,CAAC,EAAE,MAAM,GAAG,mBAAmB,EAAE,CAGjD;IAED,4EAA4E;IAC5E,MAAM,IAAI,mBAAmB,EAAE,CAG9B;CACD","sourcesContent":["import { existsSync, mkdirSync, readFileSync, writeFileSync } from \"node:fs\";\nimport { cpus, totalmem } from \"node:os\";\nimport { dirname, join } from \"node:path\";\nimport type { ModelFitnessReport } from \"../research/model-fitness.ts\";\n\n/**\n * Durable, HOST-KEYED storage for model fitness reports. Fitness is a property of a model ON a\n * host (tok/s and latency-driven failures do not travel between machines), so reports are keyed\n * by a hardware fingerprint: the same model can be \"the heavy lifter\" on one machine and\n * \"waiting for better hardware\" on another, and pi remembers both without confusing them —\n * including when settings/dotfiles are synced across machines.\n */\n\nexport interface HostFingerprint {\n\t/** Stable, human-readable id derived from the specs below. */\n\tid: string;\n\tcpu: string;\n\tcores: number;\n\ttotalMemGb: number;\n}\n\nexport function currentHostFingerprint(): HostFingerprint {\n\tconst cpuList = cpus();\n\tconst cpu = (cpuList[0]?.model ?? \"unknown-cpu\").trim();\n\tconst cores = cpuList.length;\n\tconst totalMemGb = Math.round(totalmem() / 1024 ** 3);\n\tconst id = `${cpu\n\t\t.toLowerCase()\n\t\t.replace(/[^a-z0-9]+/g, \"-\")\n\t\t.replace(/^-+|-+$/g, \"\")\n\t\t.slice(0, 48)}-${cores}c-${totalMemGb}g`;\n\treturn { id, cpu, cores, totalMemGb };\n}\n\nexport interface StoredFitnessReport {\n\tmodel: string;\n\treport: ModelFitnessReport;\n\tat: string;\n\thost: HostFingerprint;\n}\n\ninterface FitnessStoreFile {\n\tversion: 1;\n\t/** hostId -> modelRef -> latest stored report. */\n\thosts: Record<string, Record<string, StoredFitnessReport>>;\n}\n\nexport class FitnessStore {\n\tprivate readonly filePath: string;\n\tprivate readonly fingerprint: () => HostFingerprint;\n\n\tconstructor(filePath: string, options?: { fingerprint?: () => HostFingerprint }) {\n\t\tthis.filePath = filePath;\n\t\tthis.fingerprint = options?.fingerprint ?? currentHostFingerprint;\n\t}\n\n\tstatic forAgentDir(agentDir: string, options?: { fingerprint?: () => HostFingerprint }): FitnessStore {\n\t\treturn new FitnessStore(join(agentDir, \"state\", \"model-fitness.json\"), options);\n\t}\n\n\tprivate load(): FitnessStoreFile {\n\t\ttry {\n\t\t\tif (!existsSync(this.filePath)) return { version: 1, hosts: {} };\n\t\t\tconst parsed = JSON.parse(readFileSync(this.filePath, \"utf-8\")) as FitnessStoreFile;\n\t\t\tif (parsed && parsed.version === 1 && parsed.hosts && typeof parsed.hosts === \"object\") {\n\t\t\t\treturn parsed;\n\t\t\t}\n\t\t} catch {\n\t\t\t// Unreadable/corrupt store: start fresh in memory; the next save rewrites the file.\n\t\t}\n\t\treturn { version: 1, hosts: {} };\n\t}\n\n\t/** Persist the latest report for a model on the CURRENT host. Best-effort, returns the entry. */\n\tsave(model: string, report: ModelFitnessReport, at?: string): StoredFitnessReport {\n\t\tconst host = this.fingerprint();\n\t\tconst entry: StoredFitnessReport = { model, report, at: at ?? new Date().toISOString(), host };\n\t\tconst file = this.load();\n\t\tfile.hosts[host.id] = { ...(file.hosts[host.id] ?? {}), [model]: entry };\n\t\tmkdirSync(dirname(this.filePath), { recursive: true });\n\t\twriteFileSync(this.filePath, `${JSON.stringify(file, null, \"\\t\")}\\n`, \"utf-8\");\n\t\treturn entry;\n\t}\n\n\t/** Reports for the current host (default) or an explicit host id. */\n\tgetForHost(hostId?: string): StoredFitnessReport[] {\n\t\tconst file = this.load();\n\t\treturn Object.values(file.hosts[hostId ?? this.fingerprint().id] ?? {});\n\t}\n\n\t/** Every stored report across all hosts (for cross-machine comparisons). */\n\tgetAll(): StoredFitnessReport[] {\n\t\tconst file = this.load();\n\t\treturn Object.values(file.hosts).flatMap((models) => Object.values(models));\n\t}\n}\n"]}

package/dist/core/models/fitness-store.js

@@ -0,0 +1,61 @@
+import { existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs";
+import { cpus, totalmem } from "node:os";
+import { dirname, join } from "node:path";
+export function currentHostFingerprint() {
+    const cpuList = cpus();
+    const cpu = (cpuList[0]?.model ?? "unknown-cpu").trim();
+    const cores = cpuList.length;
+    const totalMemGb = Math.round(totalmem() / 1024 ** 3);
+    const id = `${cpu
+        .toLowerCase()
+        .replace(/[^a-z0-9]+/g, "-")
+        .replace(/^-+|-+$/g, "")
+        .slice(0, 48)}-${cores}c-${totalMemGb}g`;
+    return { id, cpu, cores, totalMemGb };
+}
+export class FitnessStore {
+    filePath;
+    fingerprint;
+    constructor(filePath, options) {
+        this.filePath = filePath;
+        this.fingerprint = options?.fingerprint ?? currentHostFingerprint;
+    }
+    static forAgentDir(agentDir, options) {
+        return new FitnessStore(join(agentDir, "state", "model-fitness.json"), options);
+    }
+    load() {
+        try {
+            if (!existsSync(this.filePath))
+                return { version: 1, hosts: {} };
+            const parsed = JSON.parse(readFileSync(this.filePath, "utf-8"));
+            if (parsed && parsed.version === 1 && parsed.hosts && typeof parsed.hosts === "object") {
+                return parsed;
+            }
+        }
+        catch {
+            // Unreadable/corrupt store: start fresh in memory; the next save rewrites the file.
+        }
+        return { version: 1, hosts: {} };
+    }
+    /** Persist the latest report for a model on the CURRENT host. Best-effort, returns the entry. */
+    save(model, report, at) {
+        const host = this.fingerprint();
+        const entry = { model, report, at: at ?? new Date().toISOString(), host };
+        const file = this.load();
+        file.hosts[host.id] = { ...(file.hosts[host.id] ?? {}), [model]: entry };
+        mkdirSync(dirname(this.filePath), { recursive: true });
+        writeFileSync(this.filePath, `${JSON.stringify(file, null, "\t")}\n`, "utf-8");
+        return entry;
+    }
+    /** Reports for the current host (default) or an explicit host id. */
+    getForHost(hostId) {
+        const file = this.load();
+        return Object.values(file.hosts[hostId ?? this.fingerprint().id] ?? {});
+    }
+    /** Every stored report across all hosts (for cross-machine comparisons). */
+    getAll() {
+        const file = this.load();
+        return Object.values(file.hosts).flatMap((models) => Object.values(models));
+    }
+}
+//# sourceMappingURL=fitness-store.js.map

package/dist/core/models/fitness-store.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"fitness-store.js","sourceRoot":"","sources":["../../../src/core/models/fitness-store.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,SAAS,EAAE,YAAY,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AAC7E,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,SAAS,CAAC;AACzC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAmB1C,MAAM,UAAU,sBAAsB,GAAoB;IACzD,MAAM,OAAO,GAAG,IAAI,EAAE,CAAC;IACvB,MAAM,GAAG,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,KAAK,IAAI,aAAa,CAAC,CAAC,IAAI,EAAE,CAAC;IACxD,MAAM,KAAK,GAAG,OAAO,CAAC,MAAM,CAAC;IAC7B,MAAM,UAAU,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,EAAE,GAAG,IAAI,IAAI,CAAC,CAAC,CAAC;IACtD,MAAM,EAAE,GAAG,GAAG,GAAG;SACf,WAAW,EAAE;SACb,OAAO,CAAC,aAAa,EAAE,GAAG,CAAC;SAC3B,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC;SACvB,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,KAAK,KAAK,UAAU,GAAG,CAAC;IAC1C,OAAO,EAAE,EAAE,EAAE,GAAG,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC;AAAA,CACtC;AAeD,MAAM,OAAO,YAAY;IACP,QAAQ,CAAS;IACjB,WAAW,CAAwB;IAEpD,YAAY,QAAgB,EAAE,OAAiD,EAAE;QAChF,IAAI,CAAC,QAAQ,GAAG,QAAQ,CAAC;QACzB,IAAI,CAAC,WAAW,GAAG,OAAO,EAAE,WAAW,IAAI,sBAAsB,CAAC;IAAA,CAClE;IAED,MAAM,CAAC,WAAW,CAAC,QAAgB,EAAE,OAAiD,EAAgB;QACrG,OAAO,IAAI,YAAY,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,oBAAoB,CAAC,EAAE,OAAO,CAAC,CAAC;IAAA,CAChF;IAEO,IAAI,GAAqB;QAChC,IAAI,CAAC;YACJ,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,QAAQ,CAAC;gBAAE,OAAO,EAAE,OAAO,EAAE,CAAC,EAAE,KAAK,EAAE,EAAE,EAAE,CAAC;YACjE,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAqB,CAAC;YACpF,IAAI,MAAM,IAAI,MAAM,CAAC,OAAO,KAAK,CAAC,IAAI,MAAM,CAAC,KAAK,IAAI,OAAO,MAAM,CAAC,KAAK,KAAK,QAAQ,EAAE,CAAC;gBACxF,OAAO,MAAM,CAAC;YACf,CAAC;QACF,CAAC;QAAC,MAAM,CAAC;YACR,oFAAoF;QACrF,CAAC;QACD,OAAO,EAAE,OAAO,EAAE,CAAC,EAAE,KAAK,EAAE,EAAE,EAAE,CAAC;IAAA,CACjC;IAED,iGAAiG;IACjG,IAAI,CAAC,KAAa,EAAE,MAA0B,EAAE,EAAW,EAAuB;QACjF,MAAM,IAAI,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;QAChC,MAAM,KAAK,GAAwB,EAAE,KAAK,EAAE,MAAM,EAAE,EAAE,EAAE,EAAE,IAAI,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,EAAE,IAAI,EAAE,CAAC;QAC/F,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;QACzB,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,EAAE,CAAC;QACzE,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACvD,aAAa,CAAC,IAAI,CAAC,QAAQ,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;QAC/E,OAAO,KAAK,CAAC;IAAA,CACb;IAED,qEAAqE;IACrE,UAAU,CAAC,MAAe,EAAyB;QAClD,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;QACzB,OAAO,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,IAAI,IAAI,CAAC,WAAW,EAAE,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC;IAAA,CACxE;IAED,4EAA4E;IAC5E,MAAM,GAA0B;QAC/B,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;QACzB,OAAO,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC;IAAA,CAC5E;CACD","sourcesContent":["import { existsSync, mkdirSync, readFileSync, writeFileSync } from \"node:fs\";\nimport { cpus, totalmem } from \"node:os\";\nimport { dirname, join } from \"node:path\";\nimport type { ModelFitnessReport } from \"../research/model-fitness.ts\";\n\n/**\n * Durable, HOST-KEYED storage for model fitness reports. Fitness is a property of a model ON a\n * host (tok/s and latency-driven failures do not travel between machines), so reports are keyed\n * by a hardware fingerprint: the same model can be \"the heavy lifter\" on one machine and\n * \"waiting for better hardware\" on another, and pi remembers both without confusing them —\n * including when settings/dotfiles are synced across machines.\n */\n\nexport interface HostFingerprint {\n\t/** Stable, human-readable id derived from the specs below. */\n\tid: string;\n\tcpu: string;\n\tcores: number;\n\ttotalMemGb: number;\n}\n\nexport function currentHostFingerprint(): HostFingerprint {\n\tconst cpuList = cpus();\n\tconst cpu = (cpuList[0]?.model ?? \"unknown-cpu\").trim();\n\tconst cores = cpuList.length;\n\tconst totalMemGb = Math.round(totalmem() / 1024 ** 3);\n\tconst id = `${cpu\n\t\t.toLowerCase()\n\t\t.replace(/[^a-z0-9]+/g, \"-\")\n\t\t.replace(/^-+|-+$/g, \"\")\n\t\t.slice(0, 48)}-${cores}c-${totalMemGb}g`;\n\treturn { id, cpu, cores, totalMemGb };\n}\n\nexport interface StoredFitnessReport {\n\tmodel: string;\n\treport: ModelFitnessReport;\n\tat: string;\n\thost: HostFingerprint;\n}\n\ninterface FitnessStoreFile {\n\tversion: 1;\n\t/** hostId -> modelRef -> latest stored report. */\n\thosts: Record<string, Record<string, StoredFitnessReport>>;\n}\n\nexport class FitnessStore {\n\tprivate readonly filePath: string;\n\tprivate readonly fingerprint: () => HostFingerprint;\n\n\tconstructor(filePath: string, options?: { fingerprint?: () => HostFingerprint }) {\n\t\tthis.filePath = filePath;\n\t\tthis.fingerprint = options?.fingerprint ?? currentHostFingerprint;\n\t}\n\n\tstatic forAgentDir(agentDir: string, options?: { fingerprint?: () => HostFingerprint }): FitnessStore {\n\t\treturn new FitnessStore(join(agentDir, \"state\", \"model-fitness.json\"), options);\n\t}\n\n\tprivate load(): FitnessStoreFile {\n\t\ttry {\n\t\t\tif (!existsSync(this.filePath)) return { version: 1, hosts: {} };\n\t\t\tconst parsed = JSON.parse(readFileSync(this.filePath, \"utf-8\")) as FitnessStoreFile;\n\t\t\tif (parsed && parsed.version === 1 && parsed.hosts && typeof parsed.hosts === \"object\") {\n\t\t\t\treturn parsed;\n\t\t\t}\n\t\t} catch {\n\t\t\t// Unreadable/corrupt store: start fresh in memory; the next save rewrites the file.\n\t\t}\n\t\treturn { version: 1, hosts: {} };\n\t}\n\n\t/** Persist the latest report for a model on the CURRENT host. Best-effort, returns the entry. */\n\tsave(model: string, report: ModelFitnessReport, at?: string): StoredFitnessReport {\n\t\tconst host = this.fingerprint();\n\t\tconst entry: StoredFitnessReport = { model, report, at: at ?? new Date().toISOString(), host };\n\t\tconst file = this.load();\n\t\tfile.hosts[host.id] = { ...(file.hosts[host.id] ?? {}), [model]: entry };\n\t\tmkdirSync(dirname(this.filePath), { recursive: true });\n\t\twriteFileSync(this.filePath, `${JSON.stringify(file, null, \"\\t\")}\\n`, \"utf-8\");\n\t\treturn entry;\n\t}\n\n\t/** Reports for the current host (default) or an explicit host id. */\n\tgetForHost(hostId?: string): StoredFitnessReport[] {\n\t\tconst file = this.load();\n\t\treturn Object.values(file.hosts[hostId ?? this.fingerprint().id] ?? {});\n\t}\n\n\t/** Every stored report across all hosts (for cross-machine comparisons). */\n\tgetAll(): StoredFitnessReport[] {\n\t\tconst file = this.load();\n\t\treturn Object.values(file.hosts).flatMap((models) => Object.values(models));\n\t}\n}\n"]}