@delegance/claude-autopilot 5.0.2 → 5.0.3

package/dist/src/adapters/review-engine/codex.js

@@ -4,6 +4,12 @@ import { GuardrailError } from "../../core/errors.js";
 import { buildSystemPrompt, classifyError } from "./prompt-builder.js";
 const DEFAULT_MODEL = process.env.CODEX_MODEL ?? 'gpt-5.3-codex';
 const MAX_OUTPUT_TOKENS = 4096;
+// Per-million-token rates for gpt-5.3-codex (override via env for other models).
+// Computed client-side because the OpenAI Responses API returns token counts
+// but no $-cost field. Without this, every codex run logged costUSD=0 even
+// though tokens were tracked correctly.
+const COST_PER_M_INPUT = Number(process.env.CODEX_COST_INPUT_PER_M ?? 1.25);
+const COST_PER_M_OUTPUT = Number(process.env.CODEX_COST_OUTPUT_PER_M ?? 10.0);
 const SYSTEM_PROMPT_TEMPLATE = `You are a senior software architect providing feedback on designs, proposals, and ideas.
 
 The codebase context:
@@ -62,10 +68,16 @@ export const codexAdapter = {
             });
         }
         const rawOutput = response.output_text ?? '';
+        const costUSD = response.usage
+            ? (response.usage.input_tokens / 1_000_000) * COST_PER_M_INPUT +
+                (response.usage.output_tokens / 1_000_000) * COST_PER_M_OUTPUT
+            : undefined;
         return {
             findings: parseReviewOutput(rawOutput, 'codex'),
             rawOutput,
-            usage: response.usage ? { input: response.usage.input_tokens, output: response.usage.output_tokens } : undefined,
+            usage: response.usage
+                ? { input: response.usage.input_tokens, output: response.usage.output_tokens, costUSD }
+                : undefined,
         };
     },
 };

package/dist/src/cli/scan.js

@@ -6,6 +6,7 @@ import { runReviewPhase } from "../core/pipeline/review-phase.js";
 import { detectStack } from "../core/detect/stack.js";
 import { loadIgnoreRules, parseConfigIgnore, applyIgnoreRules } from "../core/ignore/index.js";
 import { saveCachedFindings } from "../core/persist/findings-cache.js";
+import { appendCostLog } from "../core/persist/cost-log.js";
 import { detectLLMKey, LLM_KEY_HINTS } from "../core/detect/llm-key.js";
 const C = {
     reset: '\x1b[0m', bold: '\x1b[1m', dim: '\x1b[2m',
@@ -184,6 +185,17 @@ export async function runScan(options = {}) {
     }
     // Persist findings so `guardrail fix` can read them
     saveCachedFindings(cwd, findings);
+    // Persist run to cost log so `claude-autopilot costs` reflects scans, not
+    // just full pipeline runs. Previously scan never wrote to the log, so the
+    // costs report stayed frozen at whatever the last `run` invocation produced.
+    appendCostLog(cwd, {
+        timestamp: new Date().toISOString(),
+        files: files.length,
+        inputTokens: result.usage?.input ?? 0,
+        outputTokens: result.usage?.output ?? 0,
+        costUSD: result.costUSD ?? 0,
+        durationMs: result.durationMs,
+    });
     if (result.costUSD !== undefined) {
         console.log(fmt('dim', `  $${result.costUSD.toFixed(4)} · ${result.durationMs}ms`));
     }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@delegance/claude-autopilot",
-  "version": "5.0.2",
+  "version": "5.0.3",
   "type": "module",
   "description": "Autonomous development pipeline for Claude Code: brainstorm → spec → plan → implement → migrate → validate → PR → review → merge. Multi-model, local-first, every phase a skill you can intervene in.",
   "keywords": [