@xdxer/dingtalk-agent 0.1.5-beta.13 → 0.1.5-beta.15

package/dist/src/agent-audit.js

@@ -6,7 +6,7 @@ import { dirname, isAbsolute, join, parse, relative, resolve } from 'node:path';
 import { bootstrap } from './bootstrap.js';
 import { digest, stableStringify } from './events.js';
 import { prepareOpenCodeEvalAgent, resolveOpenCodeEvalAgent, readOpenCodeVersion, runOpenCodeEvaluation, runOpenCodePrompt, loadProbeCorePassed, OPENCODE_LOAD_PROBE_PROMPT, } from './opencode-evals.js';
-import { bindRequiredBasicSkill, verifyRequiredBasicSkill } from './opencode-workspace.js';
+import { bindRequiredBasicSkill, DEFAULT_OPENCODE_MODEL, verifyRequiredBasicSkill, } from './opencode-workspace.js';
 import { countInstructionsForPath } from './instruction-path.js';
 import { resolvePackageRoot } from './package-root.js';
 import { bundledSkillPath, isBasicSkillName } from './skill-manager.js';
@@ -125,7 +125,7 @@ export function auditAgent(root, options = {}) {
     const staticPrerequisitesReady = checks
         .filter((item) => item.blocking)
         .every((item) => item.level === 'pass');
-    const effectiveModel = options.model || 'deepseek/deepseek-chat';
+    const effectiveModel = options.model || DEFAULT_OPENCODE_MODEL;
     const loadTargetHash = currentLoadTargetHash(rootReal, options, requiredSkills, effectiveModel);
     let loadReport = null;
     let loadReportPath = '';
@@ -705,6 +705,7 @@ function executeInheritanceProbe(sourceRoot, model, command, timeoutMs, reportEv
             realpathIfExists(referenceExecution.toolPaths[0]) === referenceReal;
         const definitionRaw = persistProbeExecution(reportEvidenceRoot, 'definition', definitionExecution);
         const referenceRaw = persistProbeExecution(reportEvidenceRoot, 'reference', referenceExecution);
+        const referenceAnswerPassed = terminalSingleProbeAnswer(referenceExecution.answer, expectedReference);
         const result = {
             passed: Boolean(definitionInstruction) && basicResolvedCount === 1 &&
                 definitionPermissionPassed && referencePermissionPassed &&
@@ -712,7 +713,7 @@ function executeInheritanceProbe(sourceRoot, model, command, timeoutMs, reportEv
                 definitionExecution.toolCalls === 0 &&
                 definitionExecution.answer.trim() === expectedDefinition &&
                 referenceExecution.exitCode === 0 && referenceExecution.directoryMatched &&
-                referenceExecution.answer.trim() === expectedReference &&
+                referenceAnswerPassed &&
                 referenceRead && referenceFullRead && referenceToolsScoped,
             definition: executionProbeSummary(definitionExecution, expectedDefinition, {
                 nativeRule: definitionInstruction,
@@ -727,7 +728,7 @@ function executeInheritanceProbe(sourceRoot, model, command, timeoutMs, reportEv
                 permissionPassed: referencePermissionPassed,
                 evalAgent: referenceEvalAgent,
                 raw: referenceRaw,
-            }, sandbox),
+            }, sandbox, 'terminal-single-probe'),
             host: {
                 model,
                 openCodeVersion: preflight.openCodeVersion,
@@ -758,13 +759,19 @@ function initializeProbeWorktree(workspace) {
         throw new Error(`继承探针无法初始化隔离 git worktree: ${child.error?.message || String(child.stderr || child.stdout || '').trim()}`);
     }
 }
-function executionProbeSummary(execution, expected, extra, sandbox) {
+function executionProbeSummary(execution, expected, extra, sandbox, answerContract = 'exact') {
+    const answer = execution.answer.trim();
+    const answerPassed = answerContract === 'terminal-single-probe'
+        ? terminalSingleProbeAnswer(answer, expected)
+        : answer === expected;
     return {
         passed: execution.exitCode === 0 && execution.directoryMatched &&
-            execution.answer.trim() === expected,
-        exact: execution.answer.trim() === expected,
+            answerPassed,
+        exact: answer === expected,
+        answerContract,
+        terminalSingleProbe: terminalSingleProbeAnswer(answer, expected),
         expected,
-        answer: execution.answer.trim(),
+        answer,
         exitCode: execution.exitCode,
         directoryMatched: execution.directoryMatched,
         sessionId: execution.sessionId,
@@ -780,6 +787,15 @@ function executionProbeSummary(execution, expected, extra, sandbox) {
         ...extra,
     };
 }
+function terminalSingleProbeAnswer(answer, expected) {
+    if (typeof answer !== 'string' || typeof expected !== 'string' || !expected)
+        return false;
+    const normalized = answer.trim();
+    const lines = normalized.split(/\r?\n/);
+    const probeTokens = normalized.match(/dta-(?:definition|reference|load)-probe=[^\s`]+/g) || [];
+    return lines.at(-1) === expected &&
+        probeTokens.length === 1 && probeTokens[0] === expected;
+}
 function persistProbeExecution(reportRoot, name, execution) {
     const evidenceRoot = join(reportRoot, 'inheritance-probes');
     mkdirSync(evidenceRoot, { recursive: true });
@@ -1048,7 +1064,8 @@ function verifyLoadReport(root, reportPath, report, basic, definitionInstruction
     const referencePassed = reference.path ===
         '.agents/skills/dta-basic-behavior/references/risk-authority-and-privacy.md' &&
         validRandomProbe(reference.expected, 'dta-reference-probe') &&
-        reference.answer === reference.expected && reference.exitCode === 0 &&
+        terminalSingleProbeAnswer(reference.answer, reference.expected) &&
+        reference.exitCode === 0 &&
         reference.directoryMatched === true && referenceRead && referenceFullRead &&
         reference.fullRead === true && referenceToolsScoped &&
         reference.permissionPassed === true &&
@@ -1104,7 +1121,8 @@ function verifyLoadReport(root, reportPath, report, basic, definitionInstruction
         }
         if (!referencePassed) {
             failureSummary.push('inheritance.reference: ' +
-                `answer=${reference.answer === reference.expected ? 'exact' : 'mismatch'} ` +
+                `answer=${terminalSingleProbeAnswer(reference.answer, reference.expected)
+                    ? 'terminal-single-probe' : 'mismatch'} ` +
                 `exitCode=${reference.exitCode ?? '?'} durationMs=${reference.durationMs ?? '?'} ` +
                 `fullRead=${reference.fullRead === true}`);
         }