@rely-ai/caliber 1.8.0 → 1.9.0

package/dist/bin.js

@@ -627,6 +627,33 @@ init_config();
 
 // src/llm/anthropic.ts
 import Anthropic from "@anthropic-ai/sdk";
+
+// src/llm/usage.ts
+var usageByModel = /* @__PURE__ */ new Map();
+function trackUsage(model, usage) {
+  const existing = usageByModel.get(model);
+  if (existing) {
+    existing.inputTokens += usage.inputTokens;
+    existing.outputTokens += usage.outputTokens;
+    existing.cacheReadTokens += usage.cacheReadTokens ?? 0;
+    existing.cacheWriteTokens += usage.cacheWriteTokens ?? 0;
+    existing.calls += 1;
+  } else {
+    usageByModel.set(model, {
+      model,
+      inputTokens: usage.inputTokens,
+      outputTokens: usage.outputTokens,
+      cacheReadTokens: usage.cacheReadTokens ?? 0,
+      cacheWriteTokens: usage.cacheWriteTokens ?? 0,
+      calls: 1
+    });
+  }
+}
+function getUsageSummary() {
+  return Array.from(usageByModel.values());
+}
+
+// src/llm/anthropic.ts
 var AnthropicProvider = class {
   client;
   defaultModel;
@@ -641,6 +668,16 @@ var AnthropicProvider = class {
       system: [{ type: "text", text: options.system, cache_control: { type: "ephemeral" } }],
       messages: [{ role: "user", content: options.prompt }]
     });
+    const model = options.model || this.defaultModel;
+    if (response.usage) {
+      const u = response.usage;
+      trackUsage(model, {
+        inputTokens: u.input_tokens ?? 0,
+        outputTokens: u.output_tokens ?? 0,
+        cacheReadTokens: u.cache_read_input_tokens,
+        cacheWriteTokens: u.cache_creation_input_tokens
+      });
+    }
     const block = response.content[0];
     return block.type === "text" ? block.text : "";
   }
@@ -667,11 +704,24 @@ var AnthropicProvider = class {
       messages
     });
     let stopReason;
+    let usage;
+    const model = options.model || this.defaultModel;
     stream.on("message", (message) => {
-      stopReason = message.stop_reason;
+      const msg = message;
+      stopReason = msg.stop_reason;
+      const u = msg.usage;
+      if (u) {
+        usage = {
+          inputTokens: u.input_tokens ?? 0,
+          outputTokens: u.output_tokens ?? 0,
+          cacheReadTokens: u.cache_read_input_tokens,
+          cacheWriteTokens: u.cache_creation_input_tokens
+        };
+        trackUsage(model, usage);
+      }
     });
     stream.on("text", (text) => callbacks.onText(text));
-    stream.on("end", () => callbacks.onEnd({ stopReason }));
+    stream.on("end", () => callbacks.onEnd({ stopReason, usage }));
     stream.on("error", (error) => callbacks.onError(error));
   }
 };
@@ -722,6 +772,16 @@ var VertexProvider = class {
       system: [{ type: "text", text: options.system, cache_control: { type: "ephemeral" } }],
       messages: [{ role: "user", content: options.prompt }]
     });
+    const model = options.model || this.defaultModel;
+    if (response.usage) {
+      const u = response.usage;
+      trackUsage(model, {
+        inputTokens: u.input_tokens ?? 0,
+        outputTokens: u.output_tokens ?? 0,
+        cacheReadTokens: u.cache_read_input_tokens,
+        cacheWriteTokens: u.cache_creation_input_tokens
+      });
+    }
     const block = response.content[0];
     return block.type === "text" ? block.text : "";
   }
@@ -740,11 +800,24 @@ var VertexProvider = class {
       messages
     });
     let stopReason;
+    let usage;
+    const model = options.model || this.defaultModel;
     stream.on("message", (message) => {
-      stopReason = message.stop_reason;
+      const msg = message;
+      stopReason = msg.stop_reason;
+      const u = msg.usage;
+      if (u) {
+        usage = {
+          inputTokens: u.input_tokens ?? 0,
+          outputTokens: u.output_tokens ?? 0,
+          cacheReadTokens: u.cache_read_input_tokens,
+          cacheWriteTokens: u.cache_creation_input_tokens
+        };
+        trackUsage(model, usage);
+      }
     });
     stream.on("text", (text) => callbacks.onText(text));
-    stream.on("end", () => callbacks.onEnd({ stopReason }));
+    stream.on("end", () => callbacks.onEnd({ stopReason, usage }));
     stream.on("error", (error) => callbacks.onError(error));
   }
 };
@@ -770,6 +843,13 @@ var OpenAICompatProvider = class {
         { role: "user", content: options.prompt }
       ]
     });
+    const model = options.model || this.defaultModel;
+    if (response.usage) {
+      trackUsage(model, {
+        inputTokens: response.usage.prompt_tokens ?? 0,
+        outputTokens: response.usage.completion_tokens ?? 0
+      });
+    }
     return response.choices[0]?.message?.content || "";
   }
   async listModels() {
@@ -797,13 +877,23 @@ var OpenAICompatProvider = class {
     });
     try {
       let stopReason;
+      let usage;
+      const model = options.model || this.defaultModel;
       for await (const chunk of stream) {
         const delta = chunk.choices[0]?.delta?.content;
         if (delta != null) callbacks.onText(delta);
         const finishReason = chunk.choices[0]?.finish_reason;
         if (finishReason) stopReason = finishReason === "length" ? "max_tokens" : finishReason;
+        const chunkUsage = chunk.usage;
+        if (chunkUsage) {
+          usage = {
+            inputTokens: chunkUsage.prompt_tokens ?? 0,
+            outputTokens: chunkUsage.completion_tokens ?? 0
+          };
+          trackUsage(model, usage);
+        }
       }
-      callbacks.onEnd({ stopReason });
+      callbacks.onEnd({ stopReason, usage });
     } catch (error) {
       callbacks.onError(error instanceof Error ? error : new Error(String(error)));
     }
@@ -1910,9 +2000,9 @@ async function generateSetup(fingerprint, targetAgent, prompt, callbacks, failin
             }
             if (setup) {
               if (callbacks) callbacks.onComplete(setup, explanation);
-              resolve2({ setup, explanation });
+              resolve2({ setup, explanation, stopReason: stopReason ?? void 0 });
             } else {
-              resolve2({ setup: null, explanation, raw: preJsonBuffer });
+              resolve2({ setup: null, explanation, raw: preJsonBuffer, stopReason: stopReason ?? void 0 });
             }
           },
           onError: (error) => {
@@ -1922,12 +2012,12 @@ async function generateSetup(fingerprint, targetAgent, prompt, callbacks, failin
               return;
             }
             if (callbacks) callbacks.onError(error.message);
-            resolve2({ setup: null, raw: error.message });
+            resolve2({ setup: null, raw: error.message, stopReason: "error" });
           }
         }
       ).catch((error) => {
         if (callbacks) callbacks.onError(error.message);
-        resolve2({ setup: null, raw: error.message });
+        resolve2({ setup: null, raw: error.message, stopReason: "error" });
       });
     });
   };
@@ -5558,6 +5648,7 @@ async function initCommand(options) {
   genMessages.start();
   let generatedSetup = null;
   let rawOutput;
+  let genStopReason;
   try {
     const result = await generateSetup(
       fingerprint,
@@ -5583,15 +5674,18 @@ async function initCommand(options) {
       generatedSetup = result.setup;
       rawOutput = result.raw;
     }
+    genStopReason = result.stopReason;
   } catch (err) {
     genMessages.stop();
     const msg = err instanceof Error ? err.message : "Unknown error";
     genSpinner.fail(`Generation failed: ${msg}`);
+    writeErrorLog(config, void 0, msg, "exception");
     throw new Error("__exit__");
   }
   genMessages.stop();
   if (!generatedSetup) {
     genSpinner.fail("Failed to generate setup.");
+    writeErrorLog(config, rawOutput, void 0, genStopReason);
     if (rawOutput) {
       console.log(chalk8.dim("\nRaw LLM output (JSON parse failed):"));
       console.log(chalk8.dim(rawOutput.slice(0, 500)));
@@ -5825,6 +5919,9 @@ async function initCommand(options) {
   console.log(`    ${title("caliber skills")}         Discover community skills for your stack`);
   console.log(`    ${title("caliber undo")}         Revert all changes from this run`);
   console.log("");
+  if (options.showTokens) {
+    displayTokenUsage();
+  }
   if (report) {
     report.markStep("Finished");
     const reportPath = path19.join(process.cwd(), ".caliber", "debug-report.md");
@@ -6101,6 +6198,45 @@ function ensurePermissions() {
   if (!fs24.existsSync(".claude")) fs24.mkdirSync(".claude", { recursive: true });
   fs24.writeFileSync(settingsPath, JSON.stringify(settings, null, 2) + "\n");
 }
+function displayTokenUsage() {
+  const summary = getUsageSummary();
+  if (summary.length === 0) return;
+  console.log(chalk8.bold("  Token usage:\n"));
+  let totalIn = 0;
+  let totalOut = 0;
+  for (const m of summary) {
+    totalIn += m.inputTokens;
+    totalOut += m.outputTokens;
+    const cacheInfo = m.cacheReadTokens > 0 || m.cacheWriteTokens > 0 ? chalk8.dim(` (cache: ${m.cacheReadTokens.toLocaleString()} read, ${m.cacheWriteTokens.toLocaleString()} write)`) : "";
+    console.log(`    ${chalk8.dim(m.model)}: ${m.inputTokens.toLocaleString()} in / ${m.outputTokens.toLocaleString()} out  (${m.calls} call${m.calls === 1 ? "" : "s"})${cacheInfo}`);
+  }
+  if (summary.length > 1) {
+    console.log(`    ${chalk8.dim("Total")}: ${totalIn.toLocaleString()} in / ${totalOut.toLocaleString()} out`);
+  }
+  console.log("");
+}
+function writeErrorLog(config, rawOutput, error, stopReason) {
+  try {
+    const logPath = path19.join(process.cwd(), ".caliber", "error-log.md");
+    const lines = [
+      `# Generation Error \u2014 ${(/* @__PURE__ */ new Date()).toISOString()}`,
+      "",
+      `**Provider**: ${config.provider}`,
+      `**Model**: ${config.model}`,
+      `**Stop reason**: ${stopReason || "unknown"}`,
+      ""
+    ];
+    if (error) {
+      lines.push("## Error", "```", error, "```", "");
+    }
+    lines.push("## Raw LLM Output", "```", rawOutput || "(empty)", "```");
+    fs24.mkdirSync(path19.join(process.cwd(), ".caliber"), { recursive: true });
+    fs24.writeFileSync(logPath, lines.join("\n"));
+    console.log(chalk8.dim(`
+  Error log written to .caliber/error-log.md`));
+  } catch {
+  }
+}
 
 // src/commands/undo.ts
 import chalk9 from "chalk";
@@ -7265,7 +7401,7 @@ function parseAgentOption(value) {
   }
   return agents;
 }
-program.command("init").description("Initialize your project for AI-assisted development").option("--agent <type>", "Target agents (comma-separated): claude, cursor, codex", parseAgentOption).option("--dry-run", "Preview changes without writing files").option("--force", "Overwrite existing setup without prompting").option("--debug-report", void 0, false).action(tracked("init", initCommand));
+program.command("init").description("Initialize your project for AI-assisted development").option("--agent <type>", "Target agents (comma-separated): claude, cursor, codex", parseAgentOption).option("--dry-run", "Preview changes without writing files").option("--force", "Overwrite existing setup without prompting").option("--debug-report", void 0, false).option("--show-tokens", "Show token usage summary at the end").action(tracked("init", initCommand));
 program.command("undo").description("Revert all config changes made by Caliber").action(tracked("undo", undoCommand));
 program.command("status").description("Show current Caliber setup status").option("--json", "Output as JSON").action(tracked("status", statusCommand));
 program.command("regenerate").alias("regen").alias("re").description("Re-analyze project and regenerate setup").option("--dry-run", "Preview changes without writing files").action(tracked("regenerate", regenerateCommand));

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@rely-ai/caliber",
-  "version": "1.8.0",
+  "version": "1.9.0",
   "description": "Analyze your codebase and generate optimized AI agent configs (CLAUDE.md, .cursorrules, skills) — no API key needed",
   "type": "module",
   "bin": {