@wix/pathgrade 0.30.0 → 0.32.0

package/dist/agents/claude/sdk-options.js

@@ -67,13 +67,18 @@ export function buildClaudeSdkOptions(inputs) {
     // then layer driver-owned hermetic overrides on top — `CLAUDE_CONFIG_DIR`
     // wins on collision so an upstream leak (or a user-supplied env value)
     // cannot weaken the per-trial isolation invariant.
+    const useLocalOAuth = inputs.runtimeEnv.PATHGRADE_CLAUDE_LOCAL_OAUTH === '1';
     const env = {};
     for (const [key, value] of Object.entries(inputs.runtimeEnv)) {
         if (value === undefined)
             continue;
+        if (key === 'PATHGRADE_CLAUDE_LOCAL_OAUTH')
+            continue;
         env[key] = value;
     }
-    env.CLAUDE_CONFIG_DIR = path.join(inputs.workspacePath, CLAUDE_CONFIG_SUBDIR);
+    if (!useLocalOAuth) {
+        env.CLAUDE_CONFIG_DIR = path.join(inputs.workspacePath, CLAUDE_CONFIG_SUBDIR);
+    }
     opts.env = env;
     return opts;
 }

package/dist/providers/credentials.js

@@ -85,11 +85,19 @@ async function resolveClaude(userEnv, ports) {
     // Keychain OAuth token unusable — it's scoped to direct Anthropic, not to
     // a proxy. Skip the Keychain branch so the host-forward fallback wins.
     const hostSignalsProxy = !!ports.hostEnv('ANTHROPIC_BASE_URL');
-    // On macOS, prefer Keychain OAuth (direct Anthropic token).
+    // On macOS, prefer Claude Code's native claude.ai OAuth. Do not expose
+    // the OAuth access token as ANTHROPIC_API_KEY: Claude Code treats that env
+    // var as an external API-key auth path, and first-party OAuth tokens are
+    // rejected there as invalid API keys.
     if (ports.platform === 'darwin' && !hostSignalsProxy) {
         const token = await ports.readKeychainToken('Claude Code-credentials');
         if (token) {
-            return { env: { ANTHROPIC_API_KEY: token }, setupCommands: [], copyFromHome: [] };
+            return {
+                env: { PATHGRADE_CLAUDE_LOCAL_OAUTH: '1' },
+                setupCommands: [],
+                copyFromHome: ['.claude.json'],
+                linkFromHome: ['Library/Keychains'],
+            };
         }
     }
     // Fallback: forward host API key and base URL.

package/dist/reporters/loader.js

@@ -11,6 +11,9 @@ async function hydrateTraces(report, traceFile, resolved) {
 }
 export async function loadReports(resultsDir, opts) {
     const resolved = path.resolve(resultsDir);
+    if (!await fs.pathExists(resolved)) {
+        return [];
+    }
     const files = (await fs.readdir(resolved))
         .filter(f => f.endsWith('.json'))
         .reverse();

package/dist/sdk/agent.js

@@ -91,6 +91,14 @@ class AgentImpl {
         const turns = maxTurns ?? 30;
         return Math.ceil((turns * 80_000 + 200_000) / 1000);
     }
+    accumulateTurnUsage(turnResult) {
+        if (turnResult.inputTokens || turnResult.outputTokens) {
+            this.llm.addTokens?.(turnResult.inputTokens ?? 0, turnResult.outputTokens ?? 0);
+        }
+        if (turnResult.costUsd !== undefined) {
+            this.llm.addCost?.(turnResult.costUsd);
+        }
+    }
     async executeLoggedTurnResult(session, message, turnNumber, kind) {
         const timestamp = () => new Date().toISOString();
         this._messages.push({ role: 'user', content: message });
@@ -103,6 +111,7 @@ class AgentImpl {
         this.verbose.turnStart({ turn: turnNumber, kind, message });
         const turnStart = Date.now();
         const turnResult = await session.executeTurn(message);
+        this.accumulateTurnUsage(turnResult);
         const response = getVisibleAssistantMessage(turnResult);
         const durationMs = Date.now() - turnStart;
         this._log.push(buildModelAgentResultLogEntry({
@@ -132,7 +141,16 @@ class AgentImpl {
             if (turnResult.timedOut) {
                 throw new Error(`Agent (limit: ${timeoutSec}s) timed out (agent killed)`);
             }
-            throw new Error(`Agent exited with code ${turnResult.exitCode}`);
+            // Surface the underlying detail (e.g. an ask-bus rejection message
+            // from the Claude SDK driver) and the typed `errorSubtype` so a
+            // failing `prompt()`/`startChat()` does not collapse to a bare
+            // "Agent exited with code 1" — matching what `runConversation`
+            // already does via `converse.ts:buildTurnExitError`.
+            const detail = turnResult.rawOutput?.trim();
+            const subtype = turnResult.errorSubtype;
+            const subtypeTag = subtype ? ` (${subtype})` : '';
+            const suffix = detail ? `: ${detail}` : '';
+            throw new Error(`Agent exited with code ${turnResult.exitCode}${subtypeTag}${suffix}`);
         }
         return turnResult;
     }
@@ -165,7 +183,11 @@ class AgentImpl {
             messages: this._messages,
             log: this._log,
             exec: (cmd) => this.exec(cmd),
-            sendTurn: (message) => ms.executeTurn(message),
+            sendTurn: async (message) => {
+                const turnResult = await ms.executeTurn(message);
+                this.accumulateTurnUsage(turnResult);
+                return turnResult;
+            },
             verbose: this.verbose,
         });
     }
@@ -180,17 +202,10 @@ class AgentImpl {
         const sendTurn = async (message) => {
             const turnResult = await ms.executeTurn(message);
             turnNumber++;
-            // Accumulate agent turn tokens (from CLI stream-json) on the shared tracker
-            if (turnResult.inputTokens || turnResult.outputTokens) {
-                this.llm.addTokens?.(turnResult.inputTokens ?? 0, turnResult.outputTokens ?? 0);
-            }
-            // Accumulate agent turn cost on the same shared tracker when
-            // the upstream provider reports it (Claude SDK populates
-            // `costUsd` from `total_cost_usd`; Codex/Cursor leave it
-            // undefined, in which case this is a no-op).
-            if (turnResult.costUsd !== undefined) {
-                this.llm.addCost?.(turnResult.costUsd);
-            }
+            // Accumulate agent turn tokens (from CLI stream-json) on the shared tracker.
+            // Cost is populated by Claude SDK (from `total_cost_usd`); Codex/Cursor leave
+            // it undefined, in which case this is a no-op.
+            this.accumulateTurnUsage(turnResult);
             for (const toolEvent of turnResult.toolEvents) {
                 this._log.push({
                     type: 'tool_event',

package/dist/sdk/chat.js

@@ -78,7 +78,15 @@ export class ChatSessionImpl {
         });
         if (turnResult.exitCode !== 0) {
             this._done = true;
-            throw new Error(`Agent exited with code ${turnResult.exitCode}`);
+            // Match `agent.ts:executeLoggedTurnResult` — append the driver's
+            // captured `rawOutput` and typed `errorSubtype` so chat-session
+            // failures expose actionable detail (e.g. ask-bus rejection from
+            // the Claude SDK driver) instead of a bare exit-code string.
+            const detail = turnResult.rawOutput?.trim();
+            const subtype = turnResult.errorSubtype;
+            const subtypeTag = subtype ? ` (${subtype})` : '';
+            const suffix = detail ? `: ${detail}` : '';
+            throw new Error(`Agent exited with code ${turnResult.exitCode}${subtypeTag}${suffix}`);
         }
         this._turn = turnNumber;
         this._lastMessage = response;

package/dist/sdk/index.d.ts

@@ -18,7 +18,7 @@ export { buildAskBatchLogEntries } from './agent-result-log.js';
 export { getAgentCapabilities } from './types.js';
 export type { AgentTransport, AgentCapabilities, AgentName } from './types.js';
 export type { AskBus, AskBatch, AskQuestion, AskOption, AskAnswer, AskResolution, AskBatchSnapshot, AskAnswerSnapshot, AskResolutionSnapshot, AskHandle, AskHandler, AskSource, AskLifecycle, AskAnswerSource, Unsubscribe as AskBusUnsubscribe, } from './ask-bus/types.js';
-export type { Agent, AgentOptions, Message, Scorer, CheckScorer, ScoreScorer, JudgeScorer, ToolUsageScorer, ScorerContext, EvalResult, ScorerResultEntry, ScorerStatus, ChatSession, ConversationResult, ConverseOptions, Reaction, TextReaction, AskUserReaction, AskUserQuestion, AskUserOption, ReactionPreviewEntry, TextReactionPreviewEntry, AskUserReactionPreviewEntry, ReactionPreviewResult, ReactionPreviewTurn, Persona, PersonaConfig, ConversationWindowConfig, PathgradePluginOptions, PathgradeMeta, TurnTiming, TokenUsage, EvaluateOptions, ReactionPreviewStatus, } from './types.js';
+export type { Agent, AgentOptions, Message, Scorer, CheckScorer, ScoreScorer, JudgeScorer, ToolUsageScorer, ScorerContext, EvalResult, ScorerResultEntry, ScorerStatus, ChatSession, ConversationResult, ConverseOptions, UntilPredicate, UntilContext, Reaction, TextReaction, AskUserReaction, AskUserQuestion, AskUserOption, ReactionPreviewEntry, TextReactionPreviewEntry, AskUserReactionPreviewEntry, ReactionPreviewResult, ReactionPreviewTurn, StepScorer, Persona, PersonaConfig, ConversationWindowConfig, TurnDetail, ReactionFiredEntry, PathgradePluginOptions, PathgradeMeta, TurnTiming, TokenUsage, EvaluateOptions, ReactionPreviewStatus, ScoreResult, JudgeInput, CodeJudgeToolName, ToolExpectation, SessionArtifactMatchOptions, SessionArtifactContent, SessionArtifacts, RecordedEvalResult, PathgradeTestMeta, } from './types.js';
 export type { ConversationWindow, ConversationWindowOptions } from './conversation-window.js';
 export type { JudgePipelineOptions } from './judge-pipeline.js';
 export type { RunScorerOptions } from './run-scorer.js';

package/package.json

@@ -1,7 +1,19 @@
 {
   "name": "@wix/pathgrade",
-  "version": "0.30.0",
+  "version": "0.32.0",
   "description": "Evaluate whether AI agents discover and use your skills correctly",
+  "main": "./dist/sdk/index.js",
+  "types": "./dist/sdk/index.d.ts",
+  "typesVersions": {
+    "*": {
+      "plugin": [
+        "./dist/plugin/index.d.ts"
+      ],
+      "mcp-mock": [
+        "./dist/core/mcp-mock.d.ts"
+      ]
+    }
+  },
   "exports": {
     ".": {
       "types": "./dist/sdk/index.d.ts",
@@ -84,5 +96,5 @@
     "typescript": "^5.9.3",
     "zod": "4.3.6"
   },
-  "falconPackageHash": "d98d954f1467c94ec73646e4a73cd9096d86991e79e3786c42690dd7"
+  "falconPackageHash": "2eac6cd3615471de5203991805252e0bd5555626a7e922a3ac009150"
 }