crosscheck-mcp 0.1.0 → 0.1.2

package/dist/node-stdio.js

@@ -16,6 +16,7 @@ var __dirname = /* @__PURE__ */ getDirname();
 // src/entrypoints/node-stdio.ts
 import { existsSync as existsSync8, mkdirSync as mkdirSync6 } from "fs";
 import path9 from "path";
+import { fileURLToPath as fileURLToPath3 } from "url";
 import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
 
 // src/adapters/storage/better-sqlite3.ts
@@ -1233,6 +1234,7 @@ function defaultTransient(kind) {
 // src/providers/anthropic.ts
 var ANTHROPIC_API_URL = "https://api.anthropic.com/v1/messages";
 var ANTHROPIC_VERSION_HEADER = "2023-06-01";
+var ANTHROPIC_STRUCTURED_TOOL_NAME = "structured_output";
 function buildAnthropicRequest(opts) {
   let system;
   const convo = [];
@@ -1257,6 +1259,17 @@ function buildAnthropicRequest(opts) {
   if (system !== void 0) {
     body.system = system;
   }
+  if (opts.jsonSchema) {
+    body.tools = [{
+      name: ANTHROPIC_STRUCTURED_TOOL_NAME,
+      description: "Emit a single JSON object matching the requested schema.",
+      input_schema: opts.jsonSchema
+    }];
+    body.tool_choice = {
+      type: "tool",
+      name: ANTHROPIC_STRUCTURED_TOOL_NAME
+    };
+  }
   const headers = {
     "content-type": "application/json",
     "x-api-key": opts.apiKey,
@@ -1267,6 +1280,7 @@ function buildAnthropicRequest(opts) {
 function parseAnthropicResponse(opts) {
   const r = opts.resp ?? {};
   let text = "";
+  let toolUseInput = void 0;
   const content = r["content"];
   if (!Array.isArray(content)) {
     throw new ProviderError(
@@ -1276,10 +1290,19 @@ function parseAnthropicResponse(opts) {
   }
   for (const block of content) {
     if (block && typeof block === "object") {
-      const t = block["text"];
-      if (typeof t === "string") text += t;
+      const b = block;
+      const type = b["type"];
+      if (type === "tool_use" && b["name"] === ANTHROPIC_STRUCTURED_TOOL_NAME) {
+        toolUseInput = b["input"];
+      } else {
+        const t = b["text"];
+        if (typeof t === "string") text += t;
+      }
     }
   }
+  if (toolUseInput !== void 0) {
+    text = JSON.stringify(toolUseInput);
+  }
   const u = r["usage"] ?? {};
   const prompt = Math.trunc(Number(u["input_tokens"] ?? 0)) || 0;
   const cached = Math.trunc(Number(u["cache_read_input_tokens"] ?? 0)) || 0;
@@ -1324,7 +1347,8 @@ async function sendAnthropic(args) {
     apiKey: args.apiKey,
     messages: args.messages,
     maxTokens: args.maxTokens,
-    temperature: args.temperature
+    temperature: args.temperature,
+    ...args.jsonSchema ? { jsonSchema: args.jsonSchema } : {}
   });
   const doFetch = args.fetchImpl ?? globalThis.fetch;
   const init = {
@@ -1369,6 +1393,45 @@ async function sendAnthropic(args) {
 
 // src/providers/gemini.ts
 var GEMINI_API_URL_BASE = "https://generativelanguage.googleapis.com/v1beta/models";
+function jsonSchemaToGeminiSchema(schema) {
+  const out = {};
+  const type = schema["type"];
+  if (typeof type === "string") {
+    out["type"] = type.toUpperCase();
+  } else if (Array.isArray(type)) {
+    const nonNull = type.find((t) => typeof t === "string" && t !== "null");
+    const hasNull = type.includes("null");
+    if (typeof nonNull === "string") out["type"] = nonNull.toUpperCase();
+    if (hasNull) out["nullable"] = true;
+  }
+  const passthrough = [
+    "description",
+    "enum",
+    "format",
+    "nullable",
+    "required",
+    "minItems",
+    "maxItems",
+    "minimum",
+    "maximum"
+  ];
+  for (const k of passthrough) {
+    if (k in schema) out[k] = schema[k];
+  }
+  if (schema["properties"] && typeof schema["properties"] === "object") {
+    const props = {};
+    for (const [name, val] of Object.entries(schema["properties"])) {
+      if (val && typeof val === "object") {
+        props[name] = jsonSchemaToGeminiSchema(val);
+      }
+    }
+    out["properties"] = props;
+  }
+  if (schema["items"] && typeof schema["items"] === "object" && !Array.isArray(schema["items"])) {
+    out["items"] = jsonSchemaToGeminiSchema(schema["items"]);
+  }
+  return out;
+}
 function buildGeminiRequest(opts) {
   const contents = [];
   let systemText = null;
@@ -1394,6 +1457,10 @@ function buildGeminiRequest(opts) {
   if (systemText !== null && systemText !== "") {
     body.systemInstruction = { parts: [{ text: systemText }] };
   }
+  if (opts.jsonSchema) {
+    body.generationConfig.responseMimeType = "application/json";
+    body.generationConfig.responseSchema = jsonSchemaToGeminiSchema(opts.jsonSchema);
+  }
   const url = `${GEMINI_API_URL_BASE}/${encodeURIComponent(opts.model)}:generateContent?key=${encodeURIComponent(opts.apiKey)}`;
   return { url, headers: {}, body };
 }
@@ -1462,7 +1529,8 @@ async function sendGemini(args) {
     apiKey: args.apiKey,
     messages: args.messages,
     maxTokens: args.maxTokens,
-    temperature: args.temperature
+    temperature: args.temperature,
+    ...args.jsonSchema ? { jsonSchema: args.jsonSchema } : {}
   });
   const doFetch = args.fetchImpl ?? globalThis.fetch;
   const init = {
@@ -1513,6 +1581,17 @@ var OPENAI_COMPAT_DEFAULT_URLS = {
   groq: "https://api.groq.com/openai/v1/chat/completions",
   deepseek: "https://api.deepseek.com/v1/chat/completions"
 };
+var OPENAI_COMPAT_NATIVE_STRUCTURED = /* @__PURE__ */ new Set([
+  "openai"
+]);
+function supportsNativeJsonSchema(provider, model) {
+  if (!OPENAI_COMPAT_NATIVE_STRUCTURED.has(provider.toLowerCase())) {
+    return false;
+  }
+  const m = model.toLowerCase();
+  if (/^o1(?:-|$)/.test(m)) return false;
+  return true;
+}
 function buildOpenAICompatibleRequest(opts) {
   const msgs = opts.messages.map((m) => ({
     role: m.role,
@@ -1528,6 +1607,16 @@ function buildOpenAICompatibleRequest(opts) {
   } else {
     body.max_completion_tokens = opts.maxTokens;
   }
+  if (opts.jsonSchema && supportsNativeJsonSchema(opts.provider, opts.model)) {
+    body.response_format = {
+      type: "json_schema",
+      json_schema: {
+        name: "structured_output",
+        schema: opts.jsonSchema,
+        strict: true
+      }
+    };
+  }
   const headers = {
     "content-type": "application/json",
     Authorization: `Bearer ${opts.apiKey}`
@@ -1599,7 +1688,8 @@ async function sendOpenAICompatible(args) {
     url: args.url,
     messages: args.messages,
     maxTokens: args.maxTokens,
-    temperature: args.temperature
+    temperature: args.temperature,
+    ...args.jsonSchema ? { jsonSchema: args.jsonSchema } : {}
   });
   const doFetch = args.fetchImpl ?? globalThis.fetch;
   const init = {
@@ -2075,11 +2165,13 @@ ${schemaText}`;
         content: "Your previous response failed validation:\n- " + lastErrs.slice(0, 5).join("\n- ") + "\nFix the issues and re-emit valid JSON only."
       });
     }
+    const useNative = opts.useNativeStructured !== false;
     const ans = await askOne(provider, msgs, {
       maxTokens: opts.maxTokens,
       temperature: temp,
       purpose,
-      ...opts.signal ? { signal: opts.signal } : {}
+      ...opts.signal ? { signal: opts.signal } : {},
+      ...useNative ? { jsonSchema: schema } : {}
     });
     lastAnswer = ans;
     if (ans.error !== void 0) {
@@ -2114,7 +2206,8 @@ async function askOne(provider, messages, opts) {
       maxTokens: opts.maxTokens,
       temperature: opts.temperature,
       purpose: opts.purpose,
-      ...opts.signal ? { signal: opts.signal } : {}
+      ...opts.signal ? { signal: opts.signal } : {},
+      ...opts.jsonSchema ? { jsonSchema: opts.jsonSchema } : {}
     });
     const wallMs = Math.trunc(performance.now() - startedWall);
     const cpu = process.cpuUsage(startedCpu);
@@ -2357,7 +2450,7 @@ async function runAudit(args, opts) {
   const rubricOverride = args["rubric"];
   const producing = toStringArray(args["producing_panelists"]).map((s) => s.toLowerCase());
   const explicit = typeof args["auditor"] === "string" ? args["auditor"] : null;
-  const cheapMode = boolArg(args["cheap_mode"], true);
+  const cheapMode = boolArg(args["cheap_mode"], opts.ctx?.cheap_mode ?? true);
   const allowSelf = boolArg(args["allow_self_audit"], false);
   let coalesce = boolArg(args["coalesce"], false);
   const strictMode = boolArg(args["strict_mode"], false);
@@ -3887,7 +3980,7 @@ async function runConfer(args, opts) {
       };
     }
   }
-  const { selected, unknown: unknownNames, blocked } = resolveProviders(
+  let { selected, unknown: unknownNames, blocked } = resolveProviders(
     resolvedProviders,
     opts.providers,
     opts.allowlist ?? null
@@ -3904,6 +3997,49 @@ async function runConfer(args, opts) {
     }
     return { tool: "confer", error: "no active providers have API keys in .env" };
   }
+  let cheapModePanelMeta = null;
+  if (opts.ctx?.cheap_mode === true && !callerSuppliedProviders && selected.length > 1) {
+    if (!opts.pricing) {
+      cheapModePanelMeta = {
+        before: selected.length,
+        after: selected.length,
+        picked: null,
+        reason: "ctx.cheap_mode=true but no pricing doc wired; kept full panel"
+      };
+    } else {
+      const availableSet = new Set(selected.map((p) => p.name.toLowerCase()));
+      const cheapWeights = opts.storage ? await loadProviderWeights(opts.storage, selected.map((p) => p.name)) : {};
+      const allowOnly = opts.allowlist && opts.allowlist.length > 0 ? opts.allowlist.map((s) => s.toLowerCase()) : void 0;
+      const tierPick = selectForDifficulty({
+        pricing: opts.pricing,
+        tier: "low",
+        availableProviders: availableSet,
+        providerWeights: cheapWeights,
+        ...allowOnly !== void 0 ? { allowOnly } : {}
+      });
+      if (tierPick.pick) {
+        const baseProvider = opts.providers[tierPick.pick.provider.toLowerCase()];
+        if (baseProvider) {
+          const beforeCount = selected.length;
+          const retargeted = retargetProvider(baseProvider, tierPick.pick.model);
+          selected = [retargeted];
+          cheapModePanelMeta = {
+            before: beforeCount,
+            after: 1,
+            picked: { provider: tierPick.pick.provider, model: tierPick.pick.model },
+            reason: "ctx.cheap_mode=true; narrowed to cheapest low-tier candidate"
+          };
+        }
+      } else {
+        cheapModePanelMeta = {
+          before: selected.length,
+          after: selected.length,
+          picked: null,
+          reason: `ctx.cheap_mode=true; ${tierPick.reason ?? "no candidate"}; kept full panel`
+        };
+      }
+    }
+  }
   const untrusted = Boolean(args["untrusted_input"]);
   const canary = untrusted ? mintCanary() : null;
   const baseSys = "You are part of a panel of LLMs consulted by an engineer working inside Claude Code. Answer directly, cite assumptions, and keep it crisp.";
@@ -4020,6 +4156,7 @@ ${ctxBody}` });
   if (claimsBlock !== null) result["claims"] = claimsBlock;
   if (leaks.length > 0) result["canary_leaks"] = leaks;
   if (autoPanelMeta !== null) result["auto_panel"] = autoPanelMeta;
+  if (cheapModePanelMeta !== null) result["cheap_mode_panel"] = cheapModePanelMeta;
   if (requestedWorkerTools.length > 0) {
     result["worker_tools"] = {
       accepted: acceptedWorkerTools,
@@ -5198,6 +5335,40 @@ import path6 from "path";
 
 // src/tools/orchestrate.ts
 import { performance as performance3 } from "perf_hooks";
+
+// src/core/dead-models.ts
+var DEAD_MODELS_TTL_MS = 5 * 60 * 1e3;
+var deadModels = /* @__PURE__ */ new Map();
+function makeKey(provider, model) {
+  return `${provider.toLowerCase()}:${model.toLowerCase()}`;
+}
+function recordDeadModel(provider, model, reason, nowMs = Date.now()) {
+  const key = makeKey(provider, model);
+  deadModels.set(key, {
+    provider: provider.toLowerCase(),
+    model: model.toLowerCase(),
+    reason: reason.slice(0, 200),
+    expires: nowMs + DEAD_MODELS_TTL_MS
+  });
+}
+function isModelDead(provider, model, nowMs = Date.now()) {
+  const key = makeKey(provider, model);
+  const entry = deadModels.get(key);
+  if (!entry) return false;
+  if (nowMs >= entry.expires) {
+    deadModels.delete(key);
+    return false;
+  }
+  return true;
+}
+function isDeadModelError(err) {
+  if (typeof err.error !== "string") return false;
+  if (err.error_kind !== "client") return false;
+  return /HTTP\s*404|not\s*found|does not exist|deprecated|decommission/i.test(err.error);
+}
+
+// src/tools/orchestrate.ts
+var F1B_MAX_TIER_RETRIES = 2;
 var DIFFICULTY_TIERS2 = ["low", "med", "high"];
 var EST_TOKENS = {
   low: [800, 400],
@@ -5248,7 +5419,7 @@ async function runOrchestrate(args, opts) {
   const moderator = opts.providers[moderatorName.toLowerCase()] ?? selected[0];
   const failFast = boolArg2(args["fail_fast"], false);
   const planOnly = boolArg2(args["plan_only"], false);
-  const cheapMode = boolArg2(args["cheap_mode"], false);
+  const cheapMode = boolArg2(args["cheap_mode"], opts.ctx?.cheap_mode ?? false);
   const maxTokens = opts.maxTokens ?? 4096;
   const cheapAvailable = new Set(
     selected.map((p) => p.name.toLowerCase())
@@ -5334,12 +5505,51 @@ async function runOrchestrate(args, opts) {
       failedIds.add(nid);
       continue;
     }
+    const depsRaw = Array.isArray(node["depends_on"]) ? node["depends_on"] : [];
+    const upstreamBlocks = [];
+    for (const d of depsRaw) {
+      if (typeof d !== "string") continue;
+      const ur = nodeResults[d];
+      if (ur && ur.status === "ok") {
+        upstreamBlocks.push(`[node ${d} output]
+${ur.output ?? ""}`);
+      } else if (ur && ur.status === "failed") {
+        upstreamBlocks.push(`[node ${d}] [MISSING: failed \u2014 ${ur.error ?? ""}]`);
+      }
+    }
+    const ctxBlock = upstreamBlocks.length > 0 ? upstreamBlocks.join("\n\n") : "(no upstream nodes)";
+    const sysMsg = "You are a worker LLM in an orchestrated DAG. Complete the assigned task using outputs from upstream nodes when relevant. Be concise.";
+    const role = typeof node["role"] === "string" ? node["role"] : "worker";
+    const task = String(node["task"] ?? "");
+    const difficulty = String(node["difficulty"] ?? "med");
+    const msgs = [
+      { role: "system", content: sysMsg },
+      {
+        role: "user",
+        content: `ROLE: ${role}
+DIFFICULTY: ${difficulty}
+
+UPSTREAM:
+${ctxBlock}
+
+TASK:
+${task}`
+      }
+    ];
     let chosen;
+    let ans;
     let cheapReason = null;
+    const retryAttempts = [];
     const pinnedName = typeof node["provider"] === "string" ? node["provider"].toLowerCase() : null;
     const pinnedModel = typeof node["model"] === "string" ? node["model"] : null;
+    const started = performance3.now();
     if (pinnedName && opts.providers[pinnedName]) {
       chosen = opts.providers[pinnedName];
+      ans = await askOne(chosen, msgs, {
+        maxTokens,
+        temperature: 0.4,
+        purpose: "worker"
+      });
     } else if (cheapMode && !pinnedName && !pinnedModel && opts.pricing) {
       const tier = String(node["difficulty"] ?? "med");
       const pick = selectForDifficulty({
@@ -5349,21 +5559,78 @@ async function runOrchestrate(args, opts) {
         providerWeights: cheapWeights,
         ...cheapAllowOnly ? { allowOnly: cheapAllowOnly } : {}
       });
-      if (pick.pick) {
-        const base = opts.providers[pick.pick.provider];
-        if (base) chosen = retargetProvider(base, pick.pick.model);
+      const aliveCandidates = pick.scored.filter(
+        (c) => !isModelDead(c.provider, c.model)
+      );
+      if (aliveCandidates.length === 0) {
+        cheapReason = pick.reason ?? "all tier candidates filtered by deadModels cache";
       } else {
-        cheapReason = pick.reason;
+        const tryLimit = Math.min(F1B_MAX_TIER_RETRIES, aliveCandidates.length);
+        for (let i = 0; i < tryLimit; i++) {
+          const c = aliveCandidates[i];
+          const base = opts.providers[c.provider];
+          if (!base) {
+            retryAttempts.push({
+              provider: c.provider,
+              model: c.model,
+              reason: "provider not in active registry"
+            });
+            continue;
+          }
+          const candidateProvider = retargetProvider(base, c.model);
+          const candidateAns = await askOne(candidateProvider, msgs, {
+            maxTokens,
+            temperature: 0.4,
+            purpose: "worker"
+          });
+          if (candidateAns.error === void 0) {
+            chosen = candidateProvider;
+            ans = candidateAns;
+            if (retryAttempts.length > 0) {
+              cheapReason = `recovered after ${retryAttempts.length} dead-model retry(s): ` + retryAttempts.map((a) => `${a.provider}/${a.model}`).join(", ");
+            }
+            break;
+          }
+          if (isDeadModelError(candidateAns)) {
+            recordDeadModel(c.provider, c.model, candidateAns.error ?? "");
+            retryAttempts.push({
+              provider: c.provider,
+              model: c.model,
+              reason: (candidateAns.error ?? "dead").slice(0, 100)
+            });
+            continue;
+          }
+          chosen = candidateProvider;
+          ans = candidateAns;
+          if (retryAttempts.length > 0) {
+            cheapReason = `${retryAttempts.length} dead-model retry(s) before non-dead failure: ` + retryAttempts.map((a) => `${a.provider}/${a.model}`).join(", ");
+          }
+          break;
+        }
       }
       if (!chosen) {
         const idx = djb2Hash(nid) % Math.max(1, selected.length);
         chosen = selected[idx];
+        ans = await askOne(chosen, msgs, {
+          maxTokens,
+          temperature: 0.4,
+          purpose: "worker"
+        });
+        if (retryAttempts.length > 0) {
+          cheapReason = `cheap-mode tier exhausted (` + retryAttempts.map((a) => `${a.provider}/${a.model}`).join(", ") + `); fell back to id-hash rotation`;
+        }
       }
     } else {
       const idx = djb2Hash(nid) % Math.max(1, selected.length);
       chosen = selected[idx];
+      ans = await askOne(chosen, msgs, {
+        maxTokens,
+        temperature: 0.4,
+        purpose: "worker"
+      });
     }
-    if (!chosen) {
+    const wallMs = Math.trunc(performance3.now() - started);
+    if (!chosen || !ans) {
       nodeResults[nid] = {
         id: nid,
         status: "failed",
@@ -5371,49 +5638,12 @@ async function runOrchestrate(args, opts) {
         model: null,
         error: "no provider available",
         wall_ms: 0,
-        cpu_ms: 0
+        cpu_ms: 0,
+        ...retryAttempts.length > 0 ? { retry_attempts: retryAttempts } : {}
       };
       failedIds.add(nid);
       continue;
     }
-    const depsRaw = Array.isArray(node["depends_on"]) ? node["depends_on"] : [];
-    const upstreamBlocks = [];
-    for (const d of depsRaw) {
-      if (typeof d !== "string") continue;
-      const ur = nodeResults[d];
-      if (ur && ur.status === "ok") {
-        upstreamBlocks.push(`[node ${d} output]
-${ur.output ?? ""}`);
-      } else if (ur && ur.status === "failed") {
-        upstreamBlocks.push(`[node ${d}] [MISSING: failed \u2014 ${ur.error ?? ""}]`);
-      }
-    }
-    const ctxBlock = upstreamBlocks.length > 0 ? upstreamBlocks.join("\n\n") : "(no upstream nodes)";
-    const sysMsg = "You are a worker LLM in an orchestrated DAG. Complete the assigned task using outputs from upstream nodes when relevant. Be concise.";
-    const role = typeof node["role"] === "string" ? node["role"] : "worker";
-    const task = String(node["task"] ?? "");
-    const difficulty = String(node["difficulty"] ?? "med");
-    const msgs = [
-      { role: "system", content: sysMsg },
-      {
-        role: "user",
-        content: `ROLE: ${role}
-DIFFICULTY: ${difficulty}
-
-UPSTREAM:
-${ctxBlock}
-
-TASK:
-${task}`
-      }
-    ];
-    const started = performance3.now();
-    const ans = await askOne(chosen, msgs, {
-      maxTokens,
-      temperature: 0.4,
-      purpose: "worker"
-    });
-    const wallMs = Math.trunc(performance3.now() - started);
     if (ans.error !== void 0) {
       nodeResults[nid] = {
         id: nid,
@@ -5423,7 +5653,8 @@ ${task}`
         error: ans.error,
         wall_ms: wallMs,
         cpu_ms: ans.cpu_ms,
-        ...cheapReason ? { cheap_fallback_reason: cheapReason } : {}
+        ...cheapReason ? { cheap_fallback_reason: cheapReason } : {},
+        ...retryAttempts.length > 0 ? { retry_attempts: retryAttempts } : {}
       };
       failedIds.add(nid);
       continue;
@@ -5436,7 +5667,8 @@ ${task}`
       output: ans.response ?? "",
       wall_ms: wallMs,
       cpu_ms: ans.cpu_ms,
-      ...cheapReason ? { cheap_fallback_reason: cheapReason } : {}
+      ...cheapReason ? { cheap_fallback_reason: cheapReason } : {},
+      ...retryAttempts.length > 0 ? { retry_attempts: retryAttempts } : {}
     };
   }
   const missing = Object.keys(nodesById).filter(
@@ -5463,11 +5695,43 @@ Synthesize the node outputs into a single coherent deliverable. Preserve any [MI
     { role: "system", content: "You are the orchestrator. Combine node outputs into the final result." },
     { role: "user", content: recombinePrompt }
   ];
-  const synthAns = await askOne(moderator, recMsgs, {
+  let synthProvider = moderator;
+  let synthModel = null;
+  if (cheapMode && opts.pricing) {
+    const pick = selectForDifficulty({
+      pricing: opts.pricing,
+      tier: "med",
+      availableProviders: cheapAvailable,
+      providerWeights: cheapWeights,
+      ...cheapAllowOnly ? { allowOnly: cheapAllowOnly } : {}
+    });
+    for (const c of pick.scored) {
+      if (isModelDead(c.provider, c.model)) continue;
+      const base = opts.providers[c.provider];
+      if (base) {
+        synthProvider = retargetProvider(base, c.model);
+        synthModel = `${c.provider}/${c.model}`;
+        break;
+      }
+    }
+  }
+  let synthAns = await askOne(synthProvider, recMsgs, {
     maxTokens,
     temperature: 0.4,
     purpose: "synth"
   });
+  if (synthAns.error !== void 0 && synthProvider !== moderator) {
+    if (isDeadModelError(synthAns) && synthModel) {
+      const [prov, mdl] = synthModel.split("/");
+      if (prov && mdl) recordDeadModel(prov, mdl, synthAns.error ?? "");
+    }
+    synthAns = await askOne(moderator, recMsgs, {
+      maxTokens,
+      temperature: 0.4,
+      purpose: "synth"
+    });
+    synthModel = null;
+  }
   const finalText = synthAns.error !== void 0 ? "" : synthAns.response ?? "";
   const synthErr = synthAns.error;
   const publicNodes = Object.keys(nodesById).map(
@@ -5491,6 +5755,7 @@ Synthesize the node outputs into a single coherent deliverable. Preserve any [MI
     fail_fast: failFast,
     cheap_mode: cheapMode
   };
+  if (synthModel) result["synth_model"] = synthModel;
   if (synthErr) result["synth_error"] = synthErr;
   if (plannerErrors.length > 0) result["planner_errors"] = plannerErrors;
   if (unknownNames.length > 0) result["skipped_unknown_providers"] = unknownNames;
@@ -6062,6 +6327,16 @@ function errorPayload(code, message, hint, kind = "client") {
   };
 }
 
+// src/core/call-context.ts
+function buildCallContext(args) {
+  const ctx = {
+    cheap_mode: args.cheapMode,
+    session_id: args.sessionId
+  };
+  if (args.purpose) ctx.purpose = args.purpose;
+  return ctx;
+}
+
 // src/tools/create.ts
 var CREATE_DOC_MAX_BYTES = 32 * 1024;
 var CREATE_AUDIT_THRESHOLD = 0.7;
@@ -6090,6 +6365,10 @@ async function runCreate(args, opts) {
   const dryRun = boolArg3(args["dry_run"], false);
   const planOnly = boolArg3(args["plan_only"], false);
   const moderator = typeof args["moderator"] === "string" && args["moderator"] ? args["moderator"] : opts.moderator ?? "anthropic";
+  const ctx = opts.ctx ?? buildCallContext({
+    cheapMode,
+    sessionId
+  });
   const descriptors = await ingestDocuments(documents, sessionId, opts);
   const documentsPayload = formatDocumentsPayload(descriptors);
   const okCount = descriptors.filter((d) => d.status === "ok").length;
@@ -6107,7 +6386,8 @@ ${documentsPayload}`;
   }, {
     providers: opts.providers,
     allowlist: opts.allowlist ?? null,
-    ...opts.bridge ? { bridge: opts.bridge } : {}
+    ...opts.bridge ? { bridge: opts.bridge } : {},
+    ctx
   });
   const scopeAnswer = Array.isArray(scope["answers"]) ? scope["answers"].filter((a) => typeof a["response"] === "string" && a["response"]).map((a) => `[${a["provider"]}]
 ${a["response"]}`).join("\n\n") : "";
@@ -6129,7 +6409,8 @@ ${scopeAnswer || "(none)"}
     providers: opts.providers,
     allowlist: opts.allowlist ?? null,
     moderator,
-    ...opts.bridge ? { bridge: opts.bridge } : {}
+    ...opts.bridge ? { bridge: opts.bridge } : {},
+    ctx
   });
   let attempts = 1;
   if (planOnly) {
@@ -6163,7 +6444,8 @@ ${scopeAnswer || "(none)"}
     }, {
       providers: opts.providers,
       allowlist: opts.allowlist ?? null,
-      ...opts.bridge ? { bridge: opts.bridge } : {}
+      ...opts.bridge ? { bridge: opts.bridge } : {},
+      ctx
     });
   }
   let auditEnvelope = null;
@@ -6185,7 +6467,8 @@ ${scopeAnswer || "(none)"}
     }, {
       providers: opts.providers,
       allowlist: opts.allowlist ?? null,
-      ...opts.bridge ? { bridge: opts.bridge } : {}
+      ...opts.bridge ? { bridge: opts.bridge } : {},
+      ctx
     });
     let overallF = toNumberOrNull(auditEnvelope["overall_score"]);
     if (overallF !== null && overallF < auditThreshold && !cheapMode && !boolArg3(args["_no_retry"], false)) {
@@ -6209,7 +6492,8 @@ ${documentsPayload}`
           providers: opts.providers,
           allowlist: opts.allowlist ?? null,
           moderator,
-          ...opts.bridge ? { bridge: opts.bridge } : {}
+          ...opts.bridge ? { bridge: opts.bridge } : {},
+          ctx
         }
       );
       orchestration = retryOrchestration;
@@ -6228,7 +6512,8 @@ ${documentsPayload}`
         }, {
           providers: opts.providers,
           allowlist: opts.allowlist ?? null,
-          ...opts.bridge ? { bridge: opts.bridge } : {}
+          ...opts.bridge ? { bridge: opts.bridge } : {},
+          ctx
         });
         overallF = toNumberOrNull(auditEnvelope["overall_score"]);
       } else {
@@ -8079,7 +8364,7 @@ CRITERIA:
 ${criteriaBlock}`
     }
   ];
-  const maxTokens = opts.maxTokens ?? 2048;
+  const maxTokens = opts.maxTokens ?? 4096;
   const scoresByProvider = {};
   const answersCollected = [];
   const scoringErrors = {};
@@ -8275,9 +8560,23 @@ function toArray(v) {
 }
 
 // src/tools/plan.ts
+var PLAN_MODE_PRESETS = {
+  fast: { max_rounds: 2, early_stop: true, early_stop_threshold: 0.7 },
+  thorough: { max_rounds: 5, early_stop: false, early_stop_threshold: 0.7 }
+};
+var DEFAULT_PLAN_MODE = "fast";
 async function runPlan(args, opts) {
   const goal = typeof args["goal"] === "string" ? args["goal"] : String(args["goal"] ?? "");
   const constraints = typeof args["constraints"] === "string" ? args["constraints"] : "";
+  const modeArg = args["mode"];
+  const mode = modeArg === "fast" || modeArg === "thorough" ? modeArg : DEFAULT_PLAN_MODE;
+  const preset = PLAN_MODE_PRESETS[mode];
+  const maxRoundsRaw = args["max_rounds"];
+  const earlyStopRaw = args["early_stop"];
+  const earlyStopThresholdRaw = args["early_stop_threshold"];
+  const maxRounds = typeof maxRoundsRaw === "number" && Number.isFinite(maxRoundsRaw) && maxRoundsRaw > 0 ? Math.trunc(maxRoundsRaw) : preset.max_rounds;
+  const earlyStop = typeof earlyStopRaw === "boolean" ? earlyStopRaw : preset.early_stop;
+  const earlyStopThreshold = typeof earlyStopThresholdRaw === "number" && Number.isFinite(earlyStopThresholdRaw) ? earlyStopThresholdRaw : preset.early_stop_threshold;
   const merged = `We need a step-by-step plan to achieve this goal.
 
 GOAL: ${goal}
@@ -8288,7 +8587,10 @@ Return: (1) the plan as numbered steps, (2) risks, (3) alternatives considered.`
   const debateArgs = {
     topic: merged,
     context: typeof args["context"] === "string" ? args["context"] : "",
-    structured: Boolean(args["structured"])
+    structured: Boolean(args["structured"]),
+    max_rounds: maxRounds,
+    early_stop: earlyStop,
+    early_stop_threshold: earlyStopThreshold
   };
   if (args["providers"] !== void 0) debateArgs["providers"] = args["providers"];
   if (args["moderator"] !== void 0) debateArgs["moderator"] = args["moderator"];
@@ -9998,7 +10300,7 @@ function critiqueTool(providers, allowlist, bridge) {
 function planTool(providers, allowlist, bridge) {
   return {
     name: "plan",
-    description: "Have an LLM panel debate a step-by-step plan for the stated goal under the given constraints. Returns the debate envelope with a moderator-synthesised plan. Use `structured: true` (bridge) for schema-validated synthesis.",
+    description: 'Have an LLM panel debate a step-by-step plan for the stated goal under the given constraints. Returns the debate envelope with a moderator-synthesised plan. Defaults to `mode: "fast"` (2 rounds + early_stop) \u2014 set `mode: "thorough"` for 5 rounds without early_stop. Explicit `max_rounds` / `early_stop` / `early_stop_threshold` always override the mode preset. Use `structured: true` for schema-validated synthesis.',
     inputSchema: {
       type: "object",
       additionalProperties: true,
@@ -10009,7 +10311,11 @@ function planTool(providers, allowlist, bridge) {
         providers: { type: "array", items: { type: "string" } },
         moderator: { type: "string" },
         session_id: { type: "string" },
-        structured: { type: "boolean" }
+        structured: { type: "boolean" },
+        mode: { type: "string", enum: ["fast", "thorough"] },
+        max_rounds: { type: "integer", minimum: 1 },
+        early_stop: { type: "boolean" },
+        early_stop_threshold: { type: "number", minimum: 0, maximum: 1 }
       },
       required: ["goal"]
     },
@@ -10402,8 +10708,19 @@ async function main() {
     );
   }
   installShutdownHandlers(bridge);
-  const pricingPath = process.env["CROSSCHECK_PRICING_PATH"] ?? resolveRepoFile("config/pricing.json");
-  const pricing = pricingPath && existsSync8(pricingPath) ? loadPricing(pricingPath) : {};
+  const bundledPricing = (() => {
+    try {
+      return path9.join(path9.dirname(fileURLToPath3(import.meta.url)), "pricing.json");
+    } catch {
+      return void 0;
+    }
+  })();
+  const pricingPath = [
+    process.env["CROSSCHECK_PRICING_PATH"],
+    resolveRepoFile("config/pricing.json"),
+    bundledPricing
+  ].find((p) => p && existsSync8(p));
+  const pricing = pricingPath ? loadPricing(pricingPath) : {};
   const providers = buildProviders({ env: process.env, pricing });
   if (Object.keys(providers).length > 0) {
     process.stderr.write(