@wix/pathgrade 1.0.4 → 1.0.5

package/dist/agents/codex-app-server/agent.js

@@ -8,7 +8,7 @@ import { decideMcpToolCall, redactMcpSecrets, } from '../../sdk/mcp-safety.js';
 import { spawnAppServerTransport, } from './transport.js';
 import { normalizeUpstreamQuestion, toWireAnswerMap, } from './wire-translators.js';
 import { extractTurnCompletionFailure } from './turn-completion.js';
-const DEFAULT_MODEL = 'gpt-5.4';
+import { resolveCodexModel } from '../codex-model.js';
 const TURN_COMPLETED_METHOD = 'turn/completed';
 function recordFromUnknown(value) {
     if (value && typeof value === 'object' && !Array.isArray(value)) {
@@ -212,7 +212,7 @@ export class CodexAppServerAgent extends BaseAgent {
         const askBus = requireAskBusForLiveBatches(options, 'CodexAppServerAgent');
         const workspacePath = getWorkspacePath(runtime);
         const runtimeEnv = getRuntimeEnv(runtime);
-        const model = options?.model ?? DEFAULT_MODEL;
+        const model = resolveCodexModel(options?.model);
         const sandboxMode = this.deps.sandboxMode ?? 'workspace-write';
         let handle = null;
         let threadId = null;

package/dist/agents/codex-model.d.ts

@@ -0,0 +1 @@
+export declare function resolveCodexModel(explicitModel?: string): string;

package/dist/agents/codex-model.js

@@ -0,0 +1,4 @@
+const DEFAULT_CODEX_MODEL = 'gpt-5.6-luna';
+export function resolveCodexModel(explicitModel) {
+    return explicitModel ?? DEFAULT_CODEX_MODEL;
+}

package/dist/agents/codex.js

@@ -1,11 +1,13 @@
 import { TOOL_NAME_MAP, buildSummary, inferCodexExecAction, enrichSkillEvents } from '../tool-events.js';
 import { TranscriptAgent } from './transcript-agent.js';
 import { assertMcpRuntimeMountingSupportedForCodexExec } from '../providers/mcp-runtime-mounting.js';
+import { resolveCodexModel } from './codex-model.js';
 export class CodexAgent extends TranscriptAgent {
     async runTurn(instruction, runCommand, options) {
         await assertMcpRuntimeMountingSupportedForCodexExec({
             mcpConfigPath: options?.mcpConfigPath,
         });
+        await assertVerifiedCodexCli(runCommand);
         const promptPath = await this.writePromptFile(instruction, runCommand);
         const command = buildCodexExecCommand(promptPath, options?.model);
         const result = await runCommand(command);
@@ -26,17 +28,26 @@ export class CodexAgent extends TranscriptAgent {
         };
     }
 }
-const DEFAULT_CODEX_MODEL = 'gpt-5.4';
+const VERIFIED_CODEX_CLI_VERSION = '0.148.0-alpha.9';
+async function assertVerifiedCodexCli(runCommand) {
+    const result = await runCommand('codex --version');
+    const reportedVersion = (result.stdout || result.stderr).trim();
+    const expected = `codex-cli ${VERIFIED_CODEX_CLI_VERSION}`;
+    if (result.exitCode !== 0 || reportedVersion !== expected) {
+        throw new Error(`Codex CLI ${VERIFIED_CODEX_CLI_VERSION} required for PathGrade exec; received ${reportedVersion || `exit code ${result.exitCode}`}. Install @openai/codex@${VERIFIED_CODEX_CLI_VERSION}.`);
+    }
+}
 const CODEX_PROXY_PROVIDER_ID = 'pathgrade_openai_proxy';
-function buildCodexExecCommand(promptPath, model = DEFAULT_CODEX_MODEL) {
+function buildCodexExecCommand(promptPath, model) {
     const quotedPromptPath = JSON.stringify(promptPath);
-    const quotedModel = JSON.stringify(model);
-    const execArgs = `--full-auto --skip-git-repo-check -m ${quotedModel} - < ${quotedPromptPath}`;
+    const quotedModel = JSON.stringify(resolveCodexModel(model));
+    const globalArgs = '-a never -s workspace-write';
+    const execArgs = `--skip-git-repo-check -m ${quotedModel} - < ${quotedPromptPath}`;
     return [
         'if [ -n "${OPENAI_BASE_URL:-}" ]; then',
-        `codex exec ${buildCodexProxyConfigArgs()} ${execArgs};`,
+        `codex ${globalArgs} exec ${buildCodexProxyConfigArgs()} ${execArgs};`,
         'else',
-        `codex exec ${execArgs};`,
+        `codex ${globalArgs} exec ${execArgs};`,
         'fi',
     ].join(' ');
 }

package/dist/utils/llm-providers/anthropic.js

@@ -27,6 +27,35 @@ function sumInputTokens(usage) {
         + (usage?.cache_creation_input_tokens ?? 0)
         + (usage?.cache_read_input_tokens ?? 0);
 }
+const DEFAULT_ANTHROPIC_MODEL = 'claude-sonnet-5';
+const DEFAULT_ANTHROPIC_MODEL_CONFIG = {
+    transport: 'raw-messages',
+    temperature: 'default-zero',
+};
+const ANTHROPIC_MODEL_CONFIGS = {
+    'claude-haiku-4-5-20251001': { transport: 'raw-messages', temperature: 'default-zero' },
+    'claude-opus-4-7': { transport: 'raw-messages', temperature: 'unsupported' },
+    'claude-opus-5': { transport: 'raw-messages', temperature: 'unsupported' },
+    'claude-sonnet-5': { transport: 'raw-messages', temperature: 'unsupported' },
+    'claude-opus-5[1m]': { transport: 'claude-agent-sdk' },
+};
+function getAnthropicModelConfig(model) {
+    return ANTHROPIC_MODEL_CONFIGS[model] ?? DEFAULT_ANTHROPIC_MODEL_CONFIG;
+}
+function assertRawAnthropicModel(model, config) {
+    if (config.transport !== 'raw-messages') {
+        throw new Error(`Anthropic model ${model} uses Claude Agent SDK transport syntax and cannot be sent to the raw Messages API`);
+    }
+}
+function resolveTemperature(model, config, temperature) {
+    if (config.temperature === 'unsupported') {
+        if (temperature !== undefined) {
+            throw new Error(`Anthropic model ${model} does not accept temperature; omit the option`);
+        }
+        return undefined;
+    }
+    return temperature ?? 0;
+}
 async function postAnthropic(apiKey, useCache, body, env) {
     const response = await fetch(`${resolveBaseUrl(env)}/v1/messages`, {
         method: 'POST',
@@ -52,8 +81,10 @@ export const anthropicProvider = {
         if (!apiKey) {
             throw new Error('No ANTHROPIC_API_KEY available');
         }
-        const model = opts.model || 'claude-sonnet-4-20250514';
-        const temperature = opts.temperature ?? 0;
+        const model = opts.model || DEFAULT_ANTHROPIC_MODEL;
+        const modelConfig = getAnthropicModelConfig(model);
+        assertRawAnthropicModel(model, modelConfig);
+        const temperature = resolveTemperature(model, modelConfig, opts.temperature);
         const useCache = opts.cacheControl ?? false;
         const content = useCache
             ? [{ type: 'text', text: prompt, cache_control: { type: 'ephemeral' } }]
@@ -62,7 +93,7 @@ export const anthropicProvider = {
             const data = await postAnthropic(apiKey, useCache, {
                 model,
                 max_tokens: 4096,
-                temperature,
+                ...(temperature === undefined ? {} : { temperature }),
                 messages: [{ role: 'user', content }],
             }, opts.env);
             const inputTokens = sumInputTokens(data?.usage);
@@ -84,8 +115,10 @@ export const anthropicProvider = {
         if (!apiKey) {
             throw new Error('No ANTHROPIC_API_KEY available');
         }
-        const model = opts.model || 'claude-sonnet-4-20250514';
-        const temperature = opts.temperature ?? 0;
+        const model = opts.model || DEFAULT_ANTHROPIC_MODEL;
+        const modelConfig = getAnthropicModelConfig(model);
+        assertRawAnthropicModel(model, modelConfig);
+        const temperature = resolveTemperature(model, modelConfig, opts.temperature);
         const maxTokens = opts.maxTokens ?? 4096;
         const useCache = opts.cacheControl === true;
         // Tool schemas: only the last gets cache_control, serving as a cache
@@ -96,7 +129,7 @@ export const anthropicProvider = {
         const body = {
             model,
             max_tokens: maxTokens,
-            temperature,
+            ...(temperature === undefined ? {} : { temperature }),
             messages,
             tools,
         };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@wix/pathgrade",
-  "version": "1.0.4",
+  "version": "1.0.5",
   "packageManager": "yarn@4.12.0",
   "description": "Evaluate whether AI agents discover and use your skills correctly",
   "exports": {
@@ -62,6 +62,7 @@
     "test": "yarn build && vitest run && yarn test:runner-cli-smoke",
     "test:runner-cli-smoke": "node tests/runner-cli-smoke/run-smokes.mjs",
     "test:evals": "vitest run --config evals/vitest.config.mts",
+    "smoke:models": "tsx scripts/model-smoke.ts",
     "test:coverage": "vitest run --coverage",
     "dev": "tsx src/pathgrade.ts",
     "build": "tsc -p tsconfig.build.json && cp src/viewer.html dist/viewer.html && cp src/adapters/jest/reporter.cjs dist/adapters/jest/reporter.cjs"
@@ -126,5 +127,5 @@
     "typescript": "^5.9.3",
     "zod": "4.3.6"
   },
-  "falconPackageHash": "112cde138733c678a01778b225c1b3891d22417d98bbf2c479f8734c"
+  "falconPackageHash": "0d0bd8f233a4d538524961fc02c50f608b46cb42353be3b9ad1c4cde"
 }