oh-my-knowledge 0.28.0 → 0.30.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +49 -3
- package/README.zh.md +49 -3
- package/dist/src/analysis/failure-clusterer.js +1 -1
- package/dist/src/analysis/failure-clusterer.js.map +1 -1
- package/dist/src/analysis/gap-analyzer.d.ts +8 -1
- package/dist/src/analysis/gap-analyzer.d.ts.map +1 -1
- package/dist/src/analysis/gap-analyzer.js +65 -33
- package/dist/src/analysis/gap-analyzer.js.map +1 -1
- package/dist/src/analysis/report-diagnostics.js +2 -0
- package/dist/src/analysis/report-diagnostics.js.map +1 -1
- package/dist/src/analysis/sample-diagnostics.js +14 -14
- package/dist/src/analysis/sample-diagnostics.js.map +1 -1
- package/dist/src/authoring/evolver.d.ts +7 -1
- package/dist/src/authoring/evolver.d.ts.map +1 -1
- package/dist/src/authoring/evolver.js +8 -5
- package/dist/src/authoring/evolver.js.map +1 -1
- package/dist/src/authoring/generator.d.ts +21 -23
- package/dist/src/authoring/generator.d.ts.map +1 -1
- package/dist/src/authoring/generator.js +621 -40
- package/dist/src/authoring/generator.js.map +1 -1
- package/dist/src/authoring/sample-fixer.d.ts +45 -0
- package/dist/src/authoring/sample-fixer.d.ts.map +1 -0
- package/dist/src/authoring/sample-fixer.js +186 -0
- package/dist/src/authoring/sample-fixer.js.map +1 -0
- package/dist/src/cli/commands/evolve.d.ts.map +1 -1
- package/dist/src/cli/commands/evolve.js +17 -0
- package/dist/src/cli/commands/evolve.js.map +1 -1
- package/dist/src/cli/commands/observe.d.ts.map +1 -1
- package/dist/src/cli/commands/observe.js +170 -0
- package/dist/src/cli/commands/observe.js.map +1 -1
- package/dist/src/cli/commands/registry.d.ts.map +1 -1
- package/dist/src/cli/commands/registry.js +9 -1
- package/dist/src/cli/commands/registry.js.map +1 -1
- package/dist/src/cli/commands/sample.d.ts +13 -0
- package/dist/src/cli/commands/sample.d.ts.map +1 -1
- package/dist/src/cli/commands/sample.js +307 -13
- package/dist/src/cli/commands/sample.js.map +1 -1
- package/dist/src/cli/commands/studio.d.ts.map +1 -1
- package/dist/src/cli/commands/studio.js +8 -0
- package/dist/src/cli/commands/studio.js.map +1 -1
- package/dist/src/cli/i18n-dict.d.ts +1 -1
- package/dist/src/cli/i18n-dict.d.ts.map +1 -1
- package/dist/src/cli/i18n-dict.js +189 -17
- package/dist/src/cli/i18n-dict.js.map +1 -1
- package/dist/src/cli/parse-run-config.d.ts +14 -2
- package/dist/src/cli/parse-run-config.d.ts.map +1 -1
- package/dist/src/cli/parse-run-config.js +62 -10
- package/dist/src/cli/parse-run-config.js.map +1 -1
- package/dist/src/doctor/health/builtin-dimensions.d.ts.map +1 -1
- package/dist/src/doctor/health/builtin-dimensions.js +20 -23
- package/dist/src/doctor/health/builtin-dimensions.js.map +1 -1
- package/dist/src/doctor/health/composer.d.ts.map +1 -1
- package/dist/src/doctor/health/composer.js +7 -3
- package/dist/src/doctor/health/composer.js.map +1 -1
- package/dist/src/doctor/health/dimension-spec.d.ts +1 -0
- package/dist/src/doctor/health/dimension-spec.d.ts.map +1 -1
- package/dist/src/doctor/health/parser.d.ts.map +1 -1
- package/dist/src/doctor/health/parser.js +1 -0
- package/dist/src/doctor/health/parser.js.map +1 -1
- package/dist/src/doctor/health/prompt-builder.js +25 -17
- package/dist/src/doctor/health/prompt-builder.js.map +1 -1
- package/dist/src/doctor/html-renderer.d.ts.map +1 -1
- package/dist/src/doctor/html-renderer.js +76 -66
- package/dist/src/doctor/html-renderer.js.map +1 -1
- package/dist/src/doctor/rules.d.ts.map +1 -1
- package/dist/src/doctor/rules.js +32 -0
- package/dist/src/doctor/rules.js.map +1 -1
- package/dist/src/eval-core/cache.d.ts +16 -2
- package/dist/src/eval-core/cache.d.ts.map +1 -1
- package/dist/src/eval-core/cache.js +86 -15
- package/dist/src/eval-core/cache.js.map +1 -1
- package/dist/src/eval-core/evaluation-execution.d.ts +11 -1
- package/dist/src/eval-core/evaluation-execution.d.ts.map +1 -1
- package/dist/src/eval-core/evaluation-execution.js +78 -6
- package/dist/src/eval-core/evaluation-execution.js.map +1 -1
- package/dist/src/eval-core/evaluation-job.d.ts +2 -1
- package/dist/src/eval-core/evaluation-job.d.ts.map +1 -1
- package/dist/src/eval-core/evaluation-job.js +2 -1
- package/dist/src/eval-core/evaluation-job.js.map +1 -1
- package/dist/src/eval-core/evaluation-reporting.d.ts +7 -0
- package/dist/src/eval-core/evaluation-reporting.d.ts.map +1 -1
- package/dist/src/eval-core/evaluation-reporting.js +19 -2
- package/dist/src/eval-core/evaluation-reporting.js.map +1 -1
- package/dist/src/eval-core/execution-strategy.d.ts +1 -1
- package/dist/src/eval-core/execution-strategy.d.ts.map +1 -1
- package/dist/src/eval-core/execution-strategy.js +12 -1
- package/dist/src/eval-core/execution-strategy.js.map +1 -1
- package/dist/src/eval-core/mock-hook.cjs +203 -0
- package/dist/src/eval-core/mocks-runtime.d.ts +96 -0
- package/dist/src/eval-core/mocks-runtime.d.ts.map +1 -0
- package/dist/src/eval-core/mocks-runtime.js +323 -0
- package/dist/src/eval-core/mocks-runtime.js.map +1 -0
- package/dist/src/eval-core/schema.d.ts.map +1 -1
- package/dist/src/eval-core/schema.js +18 -5
- package/dist/src/eval-core/schema.js.map +1 -1
- package/dist/src/eval-core/task-planner.d.ts +9 -1
- package/dist/src/eval-core/task-planner.d.ts.map +1 -1
- package/dist/src/eval-core/task-planner.js +40 -1
- package/dist/src/eval-core/task-planner.js.map +1 -1
- package/dist/src/eval-workflows/batch-evaluation-workflow.d.ts +5 -1
- package/dist/src/eval-workflows/batch-evaluation-workflow.d.ts.map +1 -1
- package/dist/src/eval-workflows/batch-evaluation-workflow.js +2 -1
- package/dist/src/eval-workflows/batch-evaluation-workflow.js.map +1 -1
- package/dist/src/eval-workflows/evaluation-pipeline.d.ts +22 -1
- package/dist/src/eval-workflows/evaluation-pipeline.d.ts.map +1 -1
- package/dist/src/eval-workflows/evaluation-pipeline.js +39 -12
- package/dist/src/eval-workflows/evaluation-pipeline.js.map +1 -1
- package/dist/src/eval-workflows/evaluation-preparation.d.ts +5 -0
- package/dist/src/eval-workflows/evaluation-preparation.d.ts.map +1 -1
- package/dist/src/eval-workflows/evaluation-preparation.js +3 -1
- package/dist/src/eval-workflows/evaluation-preparation.js.map +1 -1
- package/dist/src/eval-workflows/run-evaluation.d.ts +12 -2
- package/dist/src/eval-workflows/run-evaluation.d.ts.map +1 -1
- package/dist/src/eval-workflows/run-evaluation.js +17 -5
- package/dist/src/eval-workflows/run-evaluation.js.map +1 -1
- package/dist/src/executors/claude-cli.d.ts +1 -1
- package/dist/src/executors/claude-cli.d.ts.map +1 -1
- package/dist/src/executors/claude-cli.js +57 -6
- package/dist/src/executors/claude-cli.js.map +1 -1
- package/dist/src/executors/claude-sdk-trace.d.ts.map +1 -1
- package/dist/src/executors/claude-sdk-trace.js +10 -4
- package/dist/src/executors/claude-sdk-trace.js.map +1 -1
- package/dist/src/executors/claude-sdk.d.ts +1 -1
- package/dist/src/executors/claude-sdk.d.ts.map +1 -1
- package/dist/src/executors/claude-sdk.js +25 -2
- package/dist/src/executors/claude-sdk.js.map +1 -1
- package/dist/src/executors/shared.d.ts +1 -1
- package/dist/src/executors/shared.d.ts.map +1 -1
- package/dist/src/executors/shared.js +6 -1
- package/dist/src/executors/shared.js.map +1 -1
- package/dist/src/grading/assertions.d.ts +5 -0
- package/dist/src/grading/assertions.d.ts.map +1 -1
- package/dist/src/grading/assertions.js +27 -1
- package/dist/src/grading/assertions.js.map +1 -1
- package/dist/src/grading/debias-validate.js +4 -4
- package/dist/src/grading/debias-validate.js.map +1 -1
- package/dist/src/grading/diagnostic.d.ts +40 -0
- package/dist/src/grading/diagnostic.d.ts.map +1 -0
- package/dist/src/grading/diagnostic.js +450 -0
- package/dist/src/grading/diagnostic.js.map +1 -0
- package/dist/src/grading/gold-cli.js +2 -2
- package/dist/src/grading/gold-cli.js.map +1 -1
- package/dist/src/grading/index.d.ts +5 -0
- package/dist/src/grading/index.d.ts.map +1 -1
- package/dist/src/grading/judge.d.ts.map +1 -1
- package/dist/src/grading/judge.js +54 -2
- package/dist/src/grading/judge.js.map +1 -1
- package/dist/src/grading/layered-scores.d.ts.map +1 -1
- package/dist/src/grading/layered-scores.js +1 -0
- package/dist/src/grading/layered-scores.js.map +1 -1
- package/dist/src/inputs/eval-config.js +19 -0
- package/dist/src/inputs/eval-config.js.map +1 -1
- package/dist/src/inputs/load-samples.d.ts +22 -2
- package/dist/src/inputs/load-samples.d.ts.map +1 -1
- package/dist/src/inputs/load-samples.js +193 -22
- package/dist/src/inputs/load-samples.js.map +1 -1
- package/dist/src/inputs/skill-loader.d.ts.map +1 -1
- package/dist/src/inputs/skill-loader.js +11 -3
- package/dist/src/inputs/skill-loader.js.map +1 -1
- package/dist/src/observability/experience.d.ts +260 -0
- package/dist/src/observability/experience.d.ts.map +1 -0
- package/dist/src/observability/experience.js +1233 -0
- package/dist/src/observability/experience.js.map +1 -0
- package/dist/src/observability/inbox-view-model.d.ts +27 -0
- package/dist/src/observability/inbox-view-model.d.ts.map +1 -0
- package/dist/src/observability/inbox-view-model.js +135 -0
- package/dist/src/observability/inbox-view-model.js.map +1 -0
- package/dist/src/observability/inbox.d.ts +125 -0
- package/dist/src/observability/inbox.d.ts.map +1 -0
- package/dist/src/observability/inbox.js +741 -0
- package/dist/src/observability/inbox.js.map +1 -0
- package/dist/src/observability/problem-patterns.d.ts +51 -0
- package/dist/src/observability/problem-patterns.d.ts.map +1 -0
- package/dist/src/observability/problem-patterns.js +205 -0
- package/dist/src/observability/problem-patterns.js.map +1 -0
- package/dist/src/observability/review-state.d.ts +54 -0
- package/dist/src/observability/review-state.d.ts.map +1 -0
- package/dist/src/observability/review-state.js +131 -0
- package/dist/src/observability/review-state.js.map +1 -0
- package/dist/src/observability/skill-chain-advisories.d.ts +25 -0
- package/dist/src/observability/skill-chain-advisories.d.ts.map +1 -0
- package/dist/src/observability/skill-chain-advisories.js +69 -0
- package/dist/src/observability/skill-chain-advisories.js.map +1 -0
- package/dist/src/observability/skill-chain.d.ts +61 -0
- package/dist/src/observability/skill-chain.d.ts.map +1 -0
- package/dist/src/observability/skill-chain.js +233 -0
- package/dist/src/observability/skill-chain.js.map +1 -0
- package/dist/src/observability/text-signals.d.ts +7 -0
- package/dist/src/observability/text-signals.d.ts.map +1 -0
- package/dist/src/observability/text-signals.js +22 -0
- package/dist/src/observability/text-signals.js.map +1 -0
- package/dist/src/observability/trace-adapter.d.ts +12 -64
- package/dist/src/observability/trace-adapter.d.ts.map +1 -1
- package/dist/src/observability/trace-adapter.js +9 -355
- package/dist/src/observability/trace-adapter.js.map +1 -1
- package/dist/src/observability/trace-attribution.d.ts +56 -0
- package/dist/src/observability/trace-attribution.d.ts.map +1 -0
- package/dist/src/observability/trace-attribution.js +200 -0
- package/dist/src/observability/trace-attribution.js.map +1 -0
- package/dist/src/observability/trace-segmenter.d.ts +52 -0
- package/dist/src/observability/trace-segmenter.d.ts.map +1 -0
- package/dist/src/observability/trace-segmenter.js +320 -0
- package/dist/src/observability/trace-segmenter.js.map +1 -0
- package/dist/src/observability/trace-source.d.ts +99 -0
- package/dist/src/observability/trace-source.d.ts.map +1 -0
- package/dist/src/observability/trace-source.js +492 -0
- package/dist/src/observability/trace-source.js.map +1 -0
- package/dist/src/renderer/html-renderer.d.ts.map +1 -1
- package/dist/src/renderer/html-renderer.js +324 -218
- package/dist/src/renderer/html-renderer.js.map +1 -1
- package/dist/src/renderer/layout.d.ts.map +1 -1
- package/dist/src/renderer/layout.js +155 -71
- package/dist/src/renderer/layout.js.map +1 -1
- package/dist/src/renderer/observation-inbox-renderer.d.ts +4 -0
- package/dist/src/renderer/observation-inbox-renderer.d.ts.map +1 -0
- package/dist/src/renderer/observation-inbox-renderer.js +5376 -0
- package/dist/src/renderer/observation-inbox-renderer.js.map +1 -0
- package/dist/src/renderer/skill-detail-renderer.d.ts +15 -0
- package/dist/src/renderer/skill-detail-renderer.d.ts.map +1 -0
- package/dist/src/renderer/skill-detail-renderer.js +1234 -0
- package/dist/src/renderer/skill-detail-renderer.js.map +1 -0
- package/dist/src/renderer/skill-list-renderer.d.ts +4 -0
- package/dist/src/renderer/skill-list-renderer.d.ts.map +1 -0
- package/dist/src/renderer/skill-list-renderer.js +327 -0
- package/dist/src/renderer/skill-list-renderer.js.map +1 -0
- package/dist/src/renderer/summary.d.ts +17 -0
- package/dist/src/renderer/summary.d.ts.map +1 -1
- package/dist/src/renderer/summary.js +346 -74
- package/dist/src/renderer/summary.js.map +1 -1
- package/dist/src/renderer/table.d.ts.map +1 -1
- package/dist/src/renderer/table.js +213 -167
- package/dist/src/renderer/table.js.map +1 -1
- package/dist/src/renderer/test-view.d.ts +66 -0
- package/dist/src/renderer/test-view.d.ts.map +1 -0
- package/dist/src/renderer/test-view.js +905 -0
- package/dist/src/renderer/test-view.js.map +1 -0
- package/dist/src/renderer/trends.d.ts.map +1 -1
- package/dist/src/renderer/trends.js +3 -1
- package/dist/src/renderer/trends.js.map +1 -1
- package/dist/src/server/report-server.d.ts +6 -1
- package/dist/src/server/report-server.d.ts.map +1 -1
- package/dist/src/server/report-server.js +182 -6
- package/dist/src/server/report-server.js.map +1 -1
- package/dist/src/server/report-store.d.ts.map +1 -1
- package/dist/src/server/report-store.js +33 -1
- package/dist/src/server/report-store.js.map +1 -1
- package/dist/src/server/skill-index.d.ts +77 -0
- package/dist/src/server/skill-index.d.ts.map +1 -0
- package/dist/src/server/skill-index.js +331 -0
- package/dist/src/server/skill-index.js.map +1 -0
- package/dist/src/server/skill-insights.d.ts +92 -0
- package/dist/src/server/skill-insights.d.ts.map +1 -0
- package/dist/src/server/skill-insights.js +676 -0
- package/dist/src/server/skill-insights.js.map +1 -0
- package/dist/src/shared/hard-rules.d.ts +40 -0
- package/dist/src/shared/hard-rules.d.ts.map +1 -0
- package/dist/src/shared/hard-rules.js +200 -0
- package/dist/src/shared/hard-rules.js.map +1 -0
- package/dist/src/shared/time.d.ts +2 -0
- package/dist/src/shared/time.d.ts.map +1 -0
- package/dist/src/shared/time.js +10 -0
- package/dist/src/shared/time.js.map +1 -0
- package/dist/src/shared/tool-search.d.ts +9 -0
- package/dist/src/shared/tool-search.d.ts.map +1 -0
- package/dist/src/shared/tool-search.js +104 -0
- package/dist/src/shared/tool-search.js.map +1 -0
- package/dist/src/types/eval.d.ts +83 -0
- package/dist/src/types/eval.d.ts.map +1 -1
- package/dist/src/types/executor.d.ts +47 -0
- package/dist/src/types/executor.d.ts.map +1 -1
- package/dist/src/types/judge.d.ts +76 -0
- package/dist/src/types/judge.d.ts.map +1 -1
- package/dist/src/types/judge.js +3 -1
- package/dist/src/types/judge.js.map +1 -1
- package/dist/src/types/report.d.ts +63 -3
- package/dist/src/types/report.d.ts.map +1 -1
- package/dist/src/util/safe-slice.d.ts +23 -0
- package/dist/src/util/safe-slice.d.ts.map +1 -0
- package/dist/src/util/safe-slice.js +33 -0
- package/dist/src/util/safe-slice.js.map +1 -0
- package/package.json +5 -4
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"claude-cli.d.ts","sourceRoot":"","sources":["../../../src/executors/claude-cli.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,aAAa,EAAE,MAAM,mBAAmB,CAAC;
|
|
1
|
+
{"version":3,"file":"claude-cli.d.ts","sourceRoot":"","sources":["../../../src/executors/claude-cli.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,aAAa,EAAE,MAAM,mBAAmB,CAAC;AAiDnE,wBAAsB,iBAAiB,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,GAAG,EAAE,QAAQ,EAAE,SAA8B,EAAE,aAAa,EAAE,KAAK,EAAE,YAAY,EAAE,WAAW,EAAE,IAAI,EAAE,MAAM,EAAE,EAAE,aAAa,GAAG,OAAO,CAAC,UAAU,CAAC,CA8InN"}
|
|
@@ -1,5 +1,6 @@
|
|
|
1
1
|
import { extractAgentTrace, isClaudeSdkResultMessage } from './claude-sdk-trace.js';
|
|
2
2
|
import { buildExecEnv, DEFAULT_TIMEOUT_MS, errorMessage, interruptedExecResult, MAX_BUFFER, spawnWithSigintPropagation, timeoutExecResult, } from './shared.js';
|
|
3
|
+
import { materializeForCliConfigDir } from '../eval-core/mocks-runtime.js';
|
|
3
4
|
// claude CLI 用 `--disable-slash-commands` (文档:"Disable all skills") +
|
|
4
5
|
// `--disallowedTools Skill` 实现与 SDK 等价的完全隔离。但 CLI 没有 partial whitelist
|
|
5
6
|
// 的 flag,所以 [name1, ...] 必须 throw。
|
|
@@ -32,12 +33,46 @@ function parseStreamJson(stdout) {
|
|
|
32
33
|
}
|
|
33
34
|
return messages;
|
|
34
35
|
}
|
|
35
|
-
export async function claudeCliExecutor({ model, system, prompt, cwd, skillDir, timeoutMs = DEFAULT_TIMEOUT_MS, allowedSkills }) {
|
|
36
|
-
const args = ['-p', prompt, '--output-format', 'stream-json', '--verbose', '--model', model
|
|
36
|
+
export async function claudeCliExecutor({ model, system, prompt, cwd, skillDir, timeoutMs = DEFAULT_TIMEOUT_MS, allowedSkills, mocks, mocksBaseDir, mocksStrict, lean, effort }) {
|
|
37
|
+
const args = ['-p', prompt, '--output-format', 'stream-json', '--verbose', '--model', model,
|
|
38
|
+
// 评测必须 bypass permission,否则 Bash / Edit / Write 等工具调用会卡在交互式确认。
|
|
39
|
+
// sdk executor 用 options.permissionMode='bypassPermissions',cli 用此 flag 等价。
|
|
40
|
+
'--permission-mode', 'bypassPermissions'];
|
|
37
41
|
if (system)
|
|
38
42
|
args.push('--system-prompt', system);
|
|
43
|
+
// effort 决策:
|
|
44
|
+
// - lean=true(纯文本生成):强制 'low' — 生成结构化 JSON 不需要思考,默认 high 浪费 13K tokens / 单次。
|
|
45
|
+
// - 否则用调用方传入的 effort,或 sonnet 默认(不传 flag = claude CLI 自己定)。
|
|
46
|
+
// 实测 sonnet count=3 data-warehouse: 默认 effort 240s/$0.28; --effort low 31s/$0.07。
|
|
47
|
+
const effectiveEffort = lean ? 'low' : effort;
|
|
48
|
+
if (lean) {
|
|
49
|
+
args.push('--tools', '', '--disable-slash-commands');
|
|
50
|
+
}
|
|
51
|
+
if (effectiveEffort) {
|
|
52
|
+
args.push('--effort', effectiveEffort);
|
|
53
|
+
}
|
|
39
54
|
applySkillIsolationToCliArgs(args, allowedSkills);
|
|
40
55
|
const env = buildExecEnv(skillDir);
|
|
56
|
+
// mock 注入:mktemp 临时 settings.json,通过 `claude --settings <file>` 追加(不覆盖 ~/.claude/)
|
|
57
|
+
// 这样 OAuth 登录态 / 用户主配置全部保留,只往里加 PreToolUse hook。
|
|
58
|
+
// 跑完(成功/失败/超时/abort)cleanup 必删整个临时目录。
|
|
59
|
+
const mockHandle = mocks && mocks.length > 0
|
|
60
|
+
? materializeForCliConfigDir(mocks, mocksBaseDir, !!mocksStrict)
|
|
61
|
+
: null;
|
|
62
|
+
if (mockHandle) {
|
|
63
|
+
args.push('--settings', mockHandle.settingsFile);
|
|
64
|
+
Object.assign(env, mockHandle.env);
|
|
65
|
+
}
|
|
66
|
+
const captureMockStats = () => {
|
|
67
|
+
if (!mockHandle)
|
|
68
|
+
return undefined;
|
|
69
|
+
try {
|
|
70
|
+
return mockHandle.readStats();
|
|
71
|
+
}
|
|
72
|
+
catch {
|
|
73
|
+
return undefined;
|
|
74
|
+
}
|
|
75
|
+
};
|
|
41
76
|
const start = Date.now();
|
|
42
77
|
try {
|
|
43
78
|
const { child, done } = spawnWithSigintPropagation('claude', args, {
|
|
@@ -54,17 +89,20 @@ export async function claudeCliExecutor({ model, system, prompt, cwd, skillDir,
|
|
|
54
89
|
// 提取 result 消息
|
|
55
90
|
const resultMsgs = messages.filter(isClaudeSdkResultMessage);
|
|
56
91
|
if (resultMsgs.length === 0) {
|
|
92
|
+
const ms = captureMockStats();
|
|
57
93
|
return {
|
|
58
94
|
ok: false, error: 'no result message in stream-json output',
|
|
59
95
|
durationMs, durationApiMs: 0,
|
|
60
96
|
inputTokens: 0, outputTokens: 0, cacheReadTokens: 0, cacheCreationTokens: 0,
|
|
61
97
|
costUSD: 0, output: null, stopReason: 'error', numTurns: 0,
|
|
98
|
+
...(ms && { mockStats: ms }),
|
|
62
99
|
};
|
|
63
100
|
}
|
|
64
101
|
const last = resultMsgs[resultMsgs.length - 1];
|
|
65
102
|
const usage = last.usage || {};
|
|
66
103
|
// 提取 trace
|
|
67
104
|
const trace = extractAgentTrace(messages);
|
|
105
|
+
const ms = captureMockStats();
|
|
68
106
|
return {
|
|
69
107
|
ok: !last.errors?.length && last.subtype !== 'error',
|
|
70
108
|
durationMs: last.duration_ms || durationMs,
|
|
@@ -81,15 +119,20 @@ export async function claudeCliExecutor({ model, system, prompt, cwd, skillDir,
|
|
|
81
119
|
numSubAgents: trace.numSubAgents,
|
|
82
120
|
...(trace.turns.length > 0 && { turns: trace.turns }),
|
|
83
121
|
...(trace.toolCalls.length > 0 && { toolCalls: trace.toolCalls }),
|
|
122
|
+
...(ms && { mockStats: ms }),
|
|
84
123
|
};
|
|
85
124
|
}
|
|
86
125
|
catch (err) {
|
|
87
126
|
const durationMs = Date.now() - start;
|
|
88
127
|
const details = err;
|
|
89
|
-
if (details.killedByTimeout)
|
|
90
|
-
|
|
91
|
-
|
|
92
|
-
|
|
128
|
+
if (details.killedByTimeout) {
|
|
129
|
+
const ms = captureMockStats();
|
|
130
|
+
return { ...timeoutExecResult(timeoutMs, durationMs), ...(ms && { mockStats: ms }) };
|
|
131
|
+
}
|
|
132
|
+
if (details.killedBySignal) {
|
|
133
|
+
const ms = captureMockStats();
|
|
134
|
+
return { ...interruptedExecResult(durationMs), ...(ms && { mockStats: ms }) };
|
|
135
|
+
}
|
|
93
136
|
// 尝试从 stdout 解析 stream-json(即使进程退出码非 0 也可能有 result)
|
|
94
137
|
const messages = parseStreamJson(details.stdout || '');
|
|
95
138
|
const resultMsgs = messages.filter(isClaudeSdkResultMessage);
|
|
@@ -97,6 +140,7 @@ export async function claudeCliExecutor({ model, system, prompt, cwd, skillDir,
|
|
|
97
140
|
const last = resultMsgs[resultMsgs.length - 1];
|
|
98
141
|
const usage = last.usage || {};
|
|
99
142
|
const trace = extractAgentTrace(messages);
|
|
143
|
+
const ms = captureMockStats();
|
|
100
144
|
return {
|
|
101
145
|
ok: false,
|
|
102
146
|
error: last.errors?.join('; ') || last.result || errorMessage(err),
|
|
@@ -114,15 +158,22 @@ export async function claudeCliExecutor({ model, system, prompt, cwd, skillDir,
|
|
|
114
158
|
numSubAgents: trace.numSubAgents,
|
|
115
159
|
...(trace.turns.length > 0 && { turns: trace.turns }),
|
|
116
160
|
...(trace.toolCalls.length > 0 && { toolCalls: trace.toolCalls }),
|
|
161
|
+
...(ms && { mockStats: ms }),
|
|
117
162
|
};
|
|
118
163
|
}
|
|
164
|
+
const ms = captureMockStats();
|
|
119
165
|
return {
|
|
120
166
|
ok: false,
|
|
121
167
|
error: errorMessage(err),
|
|
122
168
|
durationMs, durationApiMs: 0,
|
|
123
169
|
inputTokens: 0, outputTokens: 0, cacheReadTokens: 0, cacheCreationTokens: 0,
|
|
124
170
|
costUSD: 0, output: null, stopReason: 'error', numTurns: 0,
|
|
171
|
+
...(ms && { mockStats: ms }),
|
|
125
172
|
};
|
|
126
173
|
}
|
|
174
|
+
finally {
|
|
175
|
+
// mock 注入路径:无论 try 走哪条出口,cleanup 必删临时 CLAUDE_CONFIG_DIR
|
|
176
|
+
mockHandle?.cleanup();
|
|
177
|
+
}
|
|
127
178
|
}
|
|
128
179
|
//# sourceMappingURL=claude-cli.js.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"claude-cli.js","sourceRoot":"","sources":["../../../src/executors/claude-cli.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,iBAAiB,EAAE,wBAAwB,EAAE,MAAM,uBAAuB,CAAC;AAEpF,OAAO,EACL,YAAY,EACZ,kBAAkB,EAClB,YAAY,EACZ,qBAAqB,EACrB,UAAU,EACV,0BAA0B,EAC1B,iBAAiB,GAElB,MAAM,aAAa,CAAC;
|
|
1
|
+
{"version":3,"file":"claude-cli.js","sourceRoot":"","sources":["../../../src/executors/claude-cli.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,iBAAiB,EAAE,wBAAwB,EAAE,MAAM,uBAAuB,CAAC;AAEpF,OAAO,EACL,YAAY,EACZ,kBAAkB,EAClB,YAAY,EACZ,qBAAqB,EACrB,UAAU,EACV,0BAA0B,EAC1B,iBAAiB,GAElB,MAAM,aAAa,CAAC;AACrB,OAAO,EAAE,0BAA0B,EAAE,MAAM,+BAA+B,CAAC;AAE3E,sEAAsE;AACtE,uEAAuE;AACvE,mCAAmC;AACnC,EAAE;AACF,6CAA6C;AAC7C,6EAA6E;AAC7E,iFAAiF;AACjF,+DAA+D;AAE/D,SAAS,4BAA4B,CAAC,IAAc,EAAE,aAAmC;IACvF,IAAI,aAAa,KAAK,SAAS;QAAE,OAAO;IACxC,IAAI,aAAa,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC7B,MAAM,IAAI,KAAK,CACb,2DAA2D,IAAI,CAAC,SAAS,CAAC,aAAa,CAAC,MAAM;cAC5F,oFAAoF;cACpF,mEAAmE,CACtE,CAAC;IACJ,CAAC;IACD,oDAAoD;IACpD,IAAI,CAAC,IAAI,CAAC,0BAA0B,EAAE,mBAAmB,EAAE,OAAO,CAAC,CAAC;AACtE,CAAC;AAED,SAAS,eAAe,CAAC,MAAc;IACrC,MAAM,QAAQ,GAA2B,EAAE,CAAC;IAC5C,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC;QACtC,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;QAC5B,IAAI,CAAC,OAAO;YAAE,SAAS;QACvB,IAAI,CAAC;YACH,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAyB,CAAC,CAAC;QAC7D,CAAC;QAAC,MAAM,CAAC,CAAC,yBAAyB,CAAC,CAAC;IACvC,CAAC;IACD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,iBAAiB,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,GAAG,EAAE,QAAQ,EAAE,SAAS,GAAG,kBAAkB,EAAE,aAAa,EAAE,KAAK,EAAE,YAAY,EAAE,WAAW,EAAE,IAAI,EAAE,MAAM,EAAiB;IAC5L,MAAM,IAAI,GAAG,CAAC,IAAI,EAAE,MAAM,EAAE,iBAAiB,EAAE,aAAa,EAAE,WAAW,EAAE,SAAS,EAAE,KAAK;QACzF,+DAA+D;QAC/D,4EAA4E;QAC5E,mBAAmB,EAAE,mBAAmB,CAAC,CAAC;IAC5C,IAAI,MAAM;QAAE,IAAI,CAAC,IAAI,CAAC,iBAAiB,EAAE,MAAM,CAAC,CAAC;IACjD,aAAa;IACb,+EAA+E;IAC/E,8DAA8D;IAC9D,kFAAkF;IAClF,MAAM,eAAe,GAAG,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC;IAC9C,IAAI,IAAI,EAAE,CAAC;QACT,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,EAAE,EAAE,0BAA0B,CAAC,CAAC;IACvD,CAAC;IACD,IAAI,eAAe,EAAE,CAAC;QACpB,IAAI,CAAC,IAAI,CAAC,UAAU,EAAE,eAAe,CAAC,CAAC;IACzC,CAAC;IACD,4BAA4B,CAAC,IAAI,EAAE,aAAa,CAAC,CAAC;IAElD,MAAM,GAAG,GAAG,YAAY,CAAC,QAAQ,CAAC,CAAC;IAEnC,mFAAmF;IACnF,iDAAiD;IACjD,sCAAsC;IACtC,MAAM,UAAU,GAAG,KAAK,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC;QAC1C,CAAC,CAAC,0BAA0B,CAAC,KAAK,EAAE,YAAY,EAAE,CAAC,CAAC,WAAW,CAAC;QAChE,CAAC,CAAC,IAAI,CAAC;IACT,IAAI,UAAU,EAAE,CAAC;QACf,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,UAAU,CAAC,YAAY,CAAC,CAAC;QACjD,MAAM,CAAC,MAAM,CAAC,GAAG,EAAE,UAAU,CAAC,GAAG,CAAC,CAAC;IACrC,CAAC;IACD,MAAM,gBAAgB,GAAG,GAAwC,EAAE;QACjE,IAAI,CAAC,UAAU;YAAE,OAAO,SAAS,CAAC;QAClC,IAAI,CAAC;YAAC,OAAO,UAAU,CAAC,SAAS,EAAE,CAAC;QAAC,CAAC;QAAC,MAAM,CAAC;YAAC,OAAO,SAAS,CAAC;QAAC,CAAC;IACpE,CAAC,CAAC;IAEF,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IACzB,IAAI,CAAC;QACH,MAAM,EAAE,KAAK,EAAE,IAAI,EAAE,GAAG,0BAA0B,CAAC,QAAQ,EAAE,IAAI,EAAE;YACjE,SAAS,EAAE,UAAU;YACrB,SAAS;YACT,GAAG;YACH,GAAG,CAAC,GAAG,IAAI,EAAE,GAAG,EAAE,CAAC;SACpB,CAAC,CAAC;QACH,mEAAmE;QACnE,KAAK,CAAC,KAAK,EAAE,GAAG,EAAE,CAAC;QACnB,MAAM,EAAE,MAAM,EAAE,GAAG,MAAM,IAAI,CAAC;QAC9B,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC;QACtC,MAAM,QAAQ,GAAG,eAAe,CAAC,MAAM,CAAC,CAAC;QAEzC,eAAe;QACf,MAAM,UAAU,GAAG,QAAQ,CAAC,MAAM,CAAC,wBAAwB,CAA6B,CAAC;QACzF,IAAI,UAAU,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC5B,MAAM,EAAE,GAAG,gBAAgB,EAAE,CAAC;YAC9B,OAAO;gBACL,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,yCAAyC;gBAC3D,UAAU,EAAE,aAAa,EAAE,CAAC;gBAC5B,WAAW,EAAE,CAAC,EAAE,YAAY,EAAE,CAAC,EAAE,eAAe,EAAE,CAAC,EAAE,mBAAmB,EAAE,CAAC;gBAC3E,OAAO,EAAE,CAAC,EAAE,MAAM,EAAE,IAAI,EAAE,UAAU,EAAE,OAAO,EAAE,QAAQ,EAAE,CAAC;gBAC1D,GAAG,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,EAAE,EAAE,CAAC;aAC7B,CAAC;QACJ,CAAC;QAED,MAAM,IAAI,GAAG,UAAU,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QAC/C,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,EAAE,CAAC;QAE/B,WAAW;QACX,MAAM,KAAK,GAAG,iBAAiB,CAAC,QAAQ,CAAC,CAAC;QAE1C,MAAM,EAAE,GAAG,gBAAgB,EAAE,CAAC;QAC9B,OAAO;YACL,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,EAAE,MAAM,IAAI,IAAI,CAAC,OAAO,KAAK,OAAO;YACpD,UAAU,EAAE,IAAI,CAAC,WAAW,IAAI,UAAU;YAC1C,aAAa,EAAE,IAAI,CAAC,eAAe,IAAI,CAAC;YACxC,WAAW,EAAE,KAAK,CAAC,YAAY,IAAI,CAAC;YACpC,YAAY,EAAE,KAAK,CAAC,aAAa,IAAI,CAAC;YACtC,eAAe,EAAE,KAAK,CAAC,uBAAuB,IAAI,CAAC;YACnD,mBAAmB,EAAE,KAAK,CAAC,2BAA2B,IAAI,CAAC;YAC3D,OAAO,EAAE,IAAI,CAAC,cAAc,IAAI,CAAC;YACjC,MAAM,EAAE,IAAI,CAAC,MAAM,IAAI,EAAE;YACzB,UAAU,EAAE,IAAI,CAAC,OAAO,IAAI,SAAS;YACrC,QAAQ,EAAE,IAAI,CAAC,SAAS,IAAI,CAAC;YAC7B,YAAY,EAAE,KAAK,CAAC,YAAY;YAChC,YAAY,EAAE,KAAK,CAAC,YAAY;YAChC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,CAAC;YACrD,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,SAAS,EAAE,KAAK,CAAC,SAAS,EAAE,CAAC;YACjE,GAAG,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,EAAE,EAAE,CAAC;SAC7B,CAAC;IACJ,CAAC;IAAC,OAAO,GAAY,EAAE,CAAC;QACtB,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC;QACtC,MAAM,OAAO,GAAG,GAAuB,CAAC;QACxC,IAAI,OAAO,CAAC,eAAe,EAAE,CAAC;YAC5B,MAAM,EAAE,GAAG,gBAAgB,EAAE,CAAC;YAC9B,OAAO,EAAE,GAAG,iBAAiB,CAAC,SAAS,EAAE,UAAU,CAAC,EAAE,GAAG,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,EAAE,EAAE,CAAC,EAAE,CAAC;QACvF,CAAC;QACD,IAAI,OAAO,CAAC,cAAc,EAAE,CAAC;YAC3B,MAAM,EAAE,GAAG,gBAAgB,EAAE,CAAC;YAC9B,OAAO,EAAE,GAAG,qBAAqB,CAAC,UAAU,CAAC,EAAE,GAAG,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,EAAE,EAAE,CAAC,EAAE,CAAC;QAChF,CAAC;QAED,oDAAoD;QACpD,MAAM,QAAQ,GAAG,eAAe,CAAC,OAAO,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC;QACvD,MAAM,UAAU,GAAG,QAAQ,CAAC,MAAM,CAAC,wBAAwB,CAA6B,CAAC;QACzF,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC1B,MAAM,IAAI,GAAG,UAAU,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;YAC/C,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,EAAE,CAAC;YAC/B,MAAM,KAAK,GAAG,iBAAiB,CAAC,QAAQ,CAAC,CAAC;YAC1C,MAAM,EAAE,GAAG,gBAAgB,EAAE,CAAC;YAC9B,OAAO;gBACL,EAAE,EAAE,KAAK;gBACT,KAAK,EAAE,IAAI,CAAC,MAAM,EAAE,IAAI,CAAC,IAAI,CAAC,IAAI,IAAI,CAAC,MAAM,IAAI,YAAY,CAAC,GAAG,CAAC;gBAClE,UAAU,EAAE,IAAI,CAAC,WAAW,IAAI,UAAU;gBAC1C,aAAa,EAAE,IAAI,CAAC,eAAe,IAAI,CAAC;gBACxC,WAAW,EAAE,KAAK,CAAC,YAAY,IAAI,CAAC;gBACpC,YAAY,EAAE,KAAK,CAAC,aAAa,IAAI,CAAC;gBACtC,eAAe,EAAE,KAAK,CAAC,uBAAuB,IAAI,CAAC;gBACnD,mBAAmB,EAAE,KAAK,CAAC,2BAA2B,IAAI,CAAC;gBAC3D,OAAO,EAAE,IAAI,CAAC,cAAc,IAAI,CAAC;gBACjC,MAAM,EAAE,IAAI,CAAC,MAAM,IAAI,IAAI;gBAC3B,UAAU,EAAE,OAAO;gBACnB,QAAQ,EAAE,IAAI,CAAC,SAAS,IAAI,CAAC;gBAC7B,YAAY,EAAE,KAAK,CAAC,YAAY;gBAChC,YAAY,EAAE,KAAK,CAAC,YAAY;gBAChC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,CAAC;gBACrD,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,SAAS,EAAE,KAAK,CAAC,SAAS,EAAE,CAAC;gBACjE,GAAG,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,EAAE,EAAE,CAAC;aAC7B,CAAC;QACJ,CAAC;QAED,MAAM,EAAE,GAAG,gBAAgB,EAAE,CAAC;QAC9B,OAAO;YACL,EAAE,EAAE,KAAK;YACT,KAAK,EAAE,YAAY,CAAC,GAAG,CAAC;YACxB,UAAU,EAAE,aAAa,EAAE,CAAC;YAC5B,WAAW,EAAE,CAAC,EAAE,YAAY,EAAE,CAAC,EAAE,eAAe,EAAE,CAAC,EAAE,mBAAmB,EAAE,CAAC;YAC3E,OAAO,EAAE,CAAC,EAAE,MAAM,EAAE,IAAI,EAAE,UAAU,EAAE,OAAO,EAAE,QAAQ,EAAE,CAAC;YAC1D,GAAG,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,EAAE,EAAE,CAAC;SAC7B,CAAC;IACJ,CAAC;YAAS,CAAC;QACT,wDAAwD;QACxD,UAAU,EAAE,OAAO,EAAE,CAAC;IACxB,CAAC;AACH,CAAC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"claude-sdk-trace.d.ts","sourceRoot":"","sources":["../../../src/executors/claude-sdk-trace.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,QAAQ,EAAE,MAAM,mBAAmB,CAAC;AAChE,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;
|
|
1
|
+
{"version":3,"file":"claude-sdk-trace.d.ts","sourceRoot":"","sources":["../../../src/executors/claude-sdk-trace.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,QAAQ,EAAE,MAAM,mBAAmB,CAAC;AAChE,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAGxD,wBAAgB,wBAAwB,CAAC,OAAO,EAAE,oBAAoB,GAAG,OAAO,CAE/E;AAED,wBAAgB,iBAAiB,CAAC,QAAQ,EAAE,oBAAoB,EAAE,EAAE,UAAU,CAAC,EAAE,MAAM,EAAE,GAAG;IAAE,KAAK,EAAE,QAAQ,EAAE,CAAC;IAAC,SAAS,EAAE,YAAY,EAAE,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,YAAY,EAAE,MAAM,CAAA;CAAE,CAsGvL"}
|
|
@@ -1,3 +1,4 @@
|
|
|
1
|
+
import { safeSliceForJson } from '../util/safe-slice.js';
|
|
1
2
|
export function isClaudeSdkResultMessage(message) {
|
|
2
3
|
return message.type === 'result';
|
|
3
4
|
}
|
|
@@ -63,11 +64,16 @@ export function extractAgentTrace(messages, timestamps) {
|
|
|
63
64
|
: Array.isArray(resultContent)
|
|
64
65
|
? resultContent.map((c) => c.text || '').join('')
|
|
65
66
|
: '';
|
|
67
|
+
// Mock 命中:omk 用 permissionDecision='deny' 把 stub 数据塞进 result.content,
|
|
68
|
+
// is_error 因此为 true,但语义上是"mock 成功返回"。识别前缀 → 覆盖 success=true。
|
|
69
|
+
// 见 src/eval-core/mocks-runtime.ts 的 wrappedReason 前缀。
|
|
70
|
+
const isMockHit = outputText.startsWith('[mock] simulated tool output');
|
|
71
|
+
const effectiveSuccess = isMockHit ? true : !isError;
|
|
66
72
|
const tc = {
|
|
67
73
|
tool: pending?.tool || 'unknown',
|
|
68
74
|
input: pending?.input || null,
|
|
69
|
-
output: outputText
|
|
70
|
-
success:
|
|
75
|
+
output: safeSliceForJson(outputText, 500, ''),
|
|
76
|
+
success: effectiveSuccess,
|
|
71
77
|
};
|
|
72
78
|
toolCalls.push(tc);
|
|
73
79
|
if (pending) {
|
|
@@ -77,7 +83,7 @@ export function extractAgentTrace(messages, timestamps) {
|
|
|
77
83
|
const placeholder = turn.toolCalls.find((t) => t.tool === pending.tool && t.output === null);
|
|
78
84
|
if (placeholder) {
|
|
79
85
|
placeholder.output = tc.output;
|
|
80
|
-
placeholder.success =
|
|
86
|
+
placeholder.success = effectiveSuccess;
|
|
81
87
|
break;
|
|
82
88
|
}
|
|
83
89
|
}
|
|
@@ -86,7 +92,7 @@ export function extractAgentTrace(messages, timestamps) {
|
|
|
86
92
|
const toolDur = msgTs && lastTurnTs ? msgTs - lastTurnTs : undefined;
|
|
87
93
|
turns.push({
|
|
88
94
|
role: 'tool',
|
|
89
|
-
content: outputText
|
|
95
|
+
content: safeSliceForJson(outputText, 500, ''),
|
|
90
96
|
...(toolDur != null && toolDur > 0 && { durationMs: toolDur }),
|
|
91
97
|
});
|
|
92
98
|
if (msgTs)
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"claude-sdk-trace.js","sourceRoot":"","sources":["../../../src/executors/claude-sdk-trace.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"claude-sdk-trace.js","sourceRoot":"","sources":["../../../src/executors/claude-sdk-trace.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,gBAAgB,EAAE,MAAM,uBAAuB,CAAC;AAEzD,MAAM,UAAU,wBAAwB,CAAC,OAA6B;IACpE,OAAO,OAAO,CAAC,IAAI,KAAK,QAAQ,CAAC;AACnC,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,QAAgC,EAAE,UAAqB;IACvF,MAAM,KAAK,GAAe,EAAE,CAAC;IAC7B,MAAM,SAAS,GAAmB,EAAE,CAAC;IACrC,MAAM,cAAc,GAAG,IAAI,GAAG,EAA4C,CAAC;IAC3E,IAAI,UAAU,GAAG,UAAU,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;IACtC,IAAI,YAAY,GAAG,CAAC,CAAC;IACrB,IAAI,YAAY,GAAG,CAAC,CAAC;IAErB,KAAK,IAAI,MAAM,GAAG,CAAC,EAAE,MAAM,GAAG,QAAQ,CAAC,MAAM,EAAE,MAAM,EAAE,EAAE,CAAC;QACxD,MAAM,GAAG,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC;QAC7B,MAAM,KAAK,GAAG,UAAU,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;QACxC,IAAI,GAAG,CAAC,IAAI,KAAK,QAAQ,IAAI,GAAG,CAAC,IAAI,KAAK,QAAQ,IAAI,GAAG,CAAC,IAAI,KAAK,kBAAkB;YAAE,SAAS;QAEhG,MAAM,OAAO,GAAG,GAAG,CAAC,OAAO,EAAE,OAAO,CAAC;QACrC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC;YAAE,SAAS;QAEtC,IAAI,GAAG,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;YAC7B,MAAM,SAAS,GAAa,EAAE,CAAC;YAC/B,MAAM,aAAa,GAAmB,EAAE,CAAC;YACzC,IAAI,cAAc,GAAG,KAAK,CAAC;YAE3B,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;gBAC5B,IAAI,KAAK,CAAC,IAAI,KAAK,MAAM,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC;oBACxC,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;oBAC3B,cAAc,GAAG,IAAI,CAAC;gBACxB,CAAC;qBAAM,IAAI,KAAK,CAAC,IAAI,KAAK,UAAU,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC;oBACnD,cAAc,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,IAAI,EAAE,EAAE,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC;oBAC7E,aAAa,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;oBAC1F,cAAc,GAAG,IAAI,CAAC;oBACtB,IAAI,KAAK,CAAC,IAAI,KAAK,OAAO;wBAAE,YAAY,EAAE,CAAC;gBAC7C,CAAC;qBAAM,IAAI,KAAK,CAAC,IAAI,KAAK,UAAU,EAAE,CAAC;oBACrC,cAAc,GAAG,IAAI,CAAC;gBACxB,CAAC;YACH,CAAC;YACD,IAAI,cAAc;gBAAE,YAAY,EAAE,CAAC;YAEnC,IAAI,SAAS,CAAC,MAAM,GAAG,CAAC,IAAI,aAAa,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBACrD,MAAM,GAAG,GAAG,KAAK,IAAI,UAAU,CAAC,CAAC,CAAC,KAAK,GAAG,UAAU,CAAC,CAAC,CAAC,SAAS,CAAC;gBACjE,KAAK,CAAC,IAAI,CAAC;oBACT,IAAI,EAAE,WAAW;oBACjB,OAAO,EAAE,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC;oBAC7B,GAAG,CAAC,aAAa,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,SAAS,EAAE,aAAa,EAAE,CAAC;oBAC7D,GAAG,CAAC,GAAG,IAAI,IAAI,IAAI,GAAG,GAAG,CAAC,IAAI,EAAE,UAAU,EAAE,GAAG,EAAE,CAAC;iBACnD,CAAC,CAAC;gBACH,IAAI,KAAK;oBAAE,UAAU,GAAG,KAAK,CAAC;YAChC,CAAC;QACH,CAAC;QAED,IAAI,GAAG,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;YACxB,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;gBAC5B,IAAI,KAAK,CAAC,IAAI,KAAK,aAAa;oBAAE,SAAS;gBAC3C,MAAM,SAAS,GAAI,KAA6C,CAAC,WAAW,IAAI,EAAE,CAAC;gBACnF,MAAM,OAAO,GAAG,cAAc,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;gBAC9C,MAAM,OAAO,GAAI,KAA2C,CAAC,QAAQ,IAAI,KAAK,CAAC;gBAC/E,MAAM,aAAa,GAAI,KAAkF,CAAC,OAAO,CAAC;gBAClH,MAAM,UAAU,GAAG,OAAO,aAAa,KAAK,QAAQ;oBAClD,CAAC,CAAC,aAAa;oBACf,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,aAAa,CAAC;wBAC5B,CAAC,CAAC,aAAa,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC;wBACjD,CAAC,CAAC,EAAE,CAAC;gBAET,sEAAsE;gBACtE,6DAA6D;gBAC7D,uDAAuD;gBACvD,MAAM,SAAS,GAAG,UAAU,CAAC,UAAU,CAAC,8BAA8B,CAAC,CAAC;gBACxE,MAAM,gBAAgB,GAAG,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC;gBAErD,MAAM,EAAE,GAAiB;oBACvB,IAAI,EAAE,OAAO,EAAE,IAAI,IAAI,SAAS;oBAChC,KAAK,EAAE,OAAO,EAAE,KAAK,IAAI,IAAI;oBAC7B,MAAM,EAAE,gBAAgB,CAAC,UAAU,EAAE,GAAG,EAAE,EAAE,CAAC;oBAC7C,OAAO,EAAE,gBAAgB;iBAC1B,CAAC;gBACF,SAAS,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;gBAEnB,IAAI,OAAO,EAAE,CAAC;oBACZ,KAAK,IAAI,CAAC,GAAG,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;wBAC3C,MAAM,IAAI,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;wBACtB,IAAI,IAAI,CAAC,IAAI,KAAK,WAAW,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC;4BAChD,MAAM,WAAW,GAAG,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,OAAO,CAAC,IAAI,IAAI,CAAC,CAAC,MAAM,KAAK,IAAI,CAAC,CAAC;4BAC7F,IAAI,WAAW,EAAE,CAAC;gCAChB,WAAW,CAAC,MAAM,GAAG,EAAE,CAAC,MAAM,CAAC;gCAC/B,WAAW,CAAC,OAAO,GAAG,gBAAgB,CAAC;gCACvC,MAAM;4BACR,CAAC;wBACH,CAAC;oBACH,CAAC;gBACH,CAAC;gBAED,MAAM,OAAO,GAAG,KAAK,IAAI,UAAU,CAAC,CAAC,CAAC,KAAK,GAAG,UAAU,CAAC,CAAC,CAAC,SAAS,CAAC;gBACrE,KAAK,CAAC,IAAI,CAAC;oBACT,IAAI,EAAE,MAAM;oBACZ,OAAO,EAAE,gBAAgB,CAAC,UAAU,EAAE,GAAG,EAAE,EAAE,CAAC;oBAC9C,GAAG,CAAC,OAAO,IAAI,IAAI,IAAI,OAAO,GAAG,CAAC,IAAI,EAAE,UAAU,EAAE,OAAO,EAAE,CAAC;iBAC/D,CAAC,CAAC;gBACH,IAAI,KAAK;oBAAE,UAAU,GAAG,KAAK,CAAC;gBAC9B,cAAc,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC;YACnC,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,EAAE,KAAK,EAAE,SAAS,EAAE,YAAY,EAAE,YAAY,EAAE,CAAC;AAC1D,CAAC"}
|
|
@@ -12,5 +12,5 @@ export declare function buildSdkIsolationOptions(allowedSkills: string[] | undef
|
|
|
12
12
|
skills?: string[];
|
|
13
13
|
disallowedTools?: string[];
|
|
14
14
|
};
|
|
15
|
-
export declare function claudeSdkExecutor({ model, system, prompt, cwd, skillDir, timeoutMs, verbose, allowedSkills }: ExecutorInput): Promise<ExecResult>;
|
|
15
|
+
export declare function claudeSdkExecutor({ model, system, prompt, cwd, skillDir, timeoutMs, verbose, allowedSkills, mocks, mocksBaseDir, mocksStrict, lean, effort }: ExecutorInput): Promise<ExecResult>;
|
|
16
16
|
//# sourceMappingURL=claude-sdk.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"claude-sdk.d.ts","sourceRoot":"","sources":["../../../src/executors/claude-sdk.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,UAAU,EAAE,aAAa,EAAE,MAAM,mBAAmB,CAAC;
|
|
1
|
+
{"version":3,"file":"claude-sdk.d.ts","sourceRoot":"","sources":["../../../src/executors/claude-sdk.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,UAAU,EAAE,aAAa,EAAE,MAAM,mBAAmB,CAAC;AAQnE;;;;;;;;GAQG;AACH,wBAAgB,wBAAwB,CAAC,aAAa,EAAE,MAAM,EAAE,GAAG,SAAS,GAAG;IAC7E,MAAM,CAAC,EAAE,MAAM,EAAE,CAAC;IAClB,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;CAC5B,CAIA;AAUD,wBAAsB,iBAAiB,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,GAAG,EAAE,QAAQ,EAAE,SAA8B,EAAE,OAAe,EAAE,aAAa,EAAE,KAAK,EAAE,YAAY,EAAE,WAAW,EAAE,IAAI,EAAE,MAAM,EAAE,EAAE,aAAa,GAAG,OAAO,CAAC,UAAU,CAAC,CA8KpO"}
|
|
@@ -3,6 +3,7 @@ import { mkdirSync, writeFileSync } from 'node:fs';
|
|
|
3
3
|
import { tmpdir } from 'node:os';
|
|
4
4
|
import { extractAgentTrace, isClaudeSdkResultMessage } from './claude-sdk-trace.js';
|
|
5
5
|
import { asErrorLike, buildExecEnv, DEFAULT_TIMEOUT_MS, errorMessage } from './shared.js';
|
|
6
|
+
import { buildSdkHookCallback } from '../eval-core/mocks-runtime.js';
|
|
6
7
|
let sdkQuery = null;
|
|
7
8
|
/**
|
|
8
9
|
* Map ExecutorInput.allowedSkills to SDK query options for skill isolation.
|
|
@@ -27,16 +28,32 @@ async function getSdkQuery() {
|
|
|
27
28
|
}
|
|
28
29
|
return sdkQuery;
|
|
29
30
|
}
|
|
30
|
-
export async function claudeSdkExecutor({ model, system, prompt, cwd, skillDir, timeoutMs = DEFAULT_TIMEOUT_MS, verbose = false, allowedSkills }) {
|
|
31
|
+
export async function claudeSdkExecutor({ model, system, prompt, cwd, skillDir, timeoutMs = DEFAULT_TIMEOUT_MS, verbose = false, allowedSkills, mocks, mocksBaseDir, mocksStrict, lean, effort }) {
|
|
31
32
|
const start = Date.now();
|
|
32
33
|
const abortController = new AbortController();
|
|
33
34
|
const timer = setTimeout(() => abortController.abort(), timeoutMs);
|
|
34
35
|
const env = buildExecEnv(skillDir);
|
|
35
36
|
const messages = [];
|
|
36
37
|
const messageTimestamps = [];
|
|
38
|
+
// mock 注入:in-process PreToolUse hook,无 spawn / 无文件 IO,跑完就消失。
|
|
39
|
+
const hookHandle = mocks && mocks.length > 0
|
|
40
|
+
? buildSdkHookCallback(mocks, mocksBaseDir, !!mocksStrict)
|
|
41
|
+
: null;
|
|
42
|
+
const hooksOpts = hookHandle
|
|
43
|
+
? { hooks: { PreToolUse: [{ hooks: [hookHandle.callback] }] } }
|
|
44
|
+
: {};
|
|
45
|
+
const mockStatsOf = () => hookHandle ? { ...hookHandle.stats } : undefined;
|
|
37
46
|
try {
|
|
38
47
|
const query = await getSdkQuery();
|
|
39
|
-
|
|
48
|
+
// lean 模式:纯文本生成路径,不需要工具循环 / skill 发现,
|
|
49
|
+
// 用 disallowedTools:['*'] + skills:[] 直接堵住,优先级高于 isolationOpts。
|
|
50
|
+
const isolationOpts = lean
|
|
51
|
+
? { skills: [], disallowedTools: ['*'] }
|
|
52
|
+
: buildSdkIsolationOptions(allowedSkills);
|
|
53
|
+
// effort:lean 强制 'low'(生成路径不需要思考),否则透传调用方传入。
|
|
54
|
+
// SDK 暴露 EffortLevel = 'low' | 'medium' | 'high' | 'xhigh' | 'max',直接对应。
|
|
55
|
+
const effectiveEffort = lean ? 'low' : effort;
|
|
56
|
+
const effortOpts = effectiveEffort ? { effort: effectiveEffort } : {};
|
|
40
57
|
const stream = query({
|
|
41
58
|
prompt,
|
|
42
59
|
options: {
|
|
@@ -48,6 +65,8 @@ export async function claudeSdkExecutor({ model, system, prompt, cwd, skillDir,
|
|
|
48
65
|
abortController,
|
|
49
66
|
env,
|
|
50
67
|
...isolationOpts,
|
|
68
|
+
...effortOpts,
|
|
69
|
+
...hooksOpts,
|
|
51
70
|
},
|
|
52
71
|
});
|
|
53
72
|
const resultMsgs = [];
|
|
@@ -79,6 +98,7 @@ export async function claudeSdkExecutor({ model, system, prompt, cwd, skillDir,
|
|
|
79
98
|
durationMs, durationApiMs: 0,
|
|
80
99
|
inputTokens: 0, outputTokens: 0, cacheReadTokens: 0, cacheCreationTokens: 0,
|
|
81
100
|
costUSD: 0, output: null, stopReason: 'error', numTurns: 0,
|
|
101
|
+
...(mockStatsOf() && { mockStats: mockStatsOf() }),
|
|
82
102
|
};
|
|
83
103
|
}
|
|
84
104
|
let output = '';
|
|
@@ -133,6 +153,7 @@ export async function claudeSdkExecutor({ model, system, prompt, cwd, skillDir,
|
|
|
133
153
|
numSubAgents: trace.numSubAgents,
|
|
134
154
|
...(trace.turns.length > 0 && { turns: trace.turns }),
|
|
135
155
|
...(trace.toolCalls.length > 0 && { toolCalls: trace.toolCalls }),
|
|
156
|
+
...(mockStatsOf() && { mockStats: mockStatsOf() }),
|
|
136
157
|
};
|
|
137
158
|
}
|
|
138
159
|
return {
|
|
@@ -151,6 +172,7 @@ export async function claudeSdkExecutor({ model, system, prompt, cwd, skillDir,
|
|
|
151
172
|
numSubAgents: trace.numSubAgents,
|
|
152
173
|
...(trace.turns.length > 0 && { turns: trace.turns }),
|
|
153
174
|
...(trace.toolCalls.length > 0 && { toolCalls: trace.toolCalls }),
|
|
175
|
+
...(mockStatsOf() && { mockStats: mockStatsOf() }),
|
|
154
176
|
};
|
|
155
177
|
}
|
|
156
178
|
catch (err) {
|
|
@@ -172,6 +194,7 @@ export async function claudeSdkExecutor({ model, system, prompt, cwd, skillDir,
|
|
|
172
194
|
costUSD: 0, output: null,
|
|
173
195
|
stopReason: isAbort ? 'timeout' : 'error',
|
|
174
196
|
numTurns: 0,
|
|
197
|
+
...(mockStatsOf() && { mockStats: mockStatsOf() }),
|
|
175
198
|
};
|
|
176
199
|
}
|
|
177
200
|
}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"claude-sdk.js","sourceRoot":"","sources":["../../../src/executors/claude-sdk.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,SAAS,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AACnD,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AAEjC,OAAO,EAAE,iBAAiB,EAAE,wBAAwB,EAAE,MAAM,uBAAuB,CAAC;AAEpF,OAAO,EAAE,WAAW,EAAE,YAAY,EAAE,kBAAkB,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;
|
|
1
|
+
{"version":3,"file":"claude-sdk.js","sourceRoot":"","sources":["../../../src/executors/claude-sdk.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,SAAS,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AACnD,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AAEjC,OAAO,EAAE,iBAAiB,EAAE,wBAAwB,EAAE,MAAM,uBAAuB,CAAC;AAEpF,OAAO,EAAE,WAAW,EAAE,YAAY,EAAE,kBAAkB,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAC1F,OAAO,EAAE,oBAAoB,EAAE,MAAM,+BAA+B,CAAC;AAErE,IAAI,QAAQ,GAAoC,IAAI,CAAC;AAErD;;;;;;;;GAQG;AACH,MAAM,UAAU,wBAAwB,CAAC,aAAmC;IAI1E,IAAI,aAAa,KAAK,SAAS;QAAE,OAAO,EAAE,CAAC;IAC3C,IAAI,aAAa,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,MAAM,EAAE,aAAa,EAAE,eAAe,EAAE,CAAC,OAAO,CAAC,EAAE,CAAC;IAC7F,OAAO,EAAE,MAAM,EAAE,aAAa,EAAE,CAAC;AACnC,CAAC;AAED,KAAK,UAAU,WAAW;IACxB,IAAI,CAAC,QAAQ,EAAE,CAAC;QACd,MAAM,GAAG,GAAG,MAAM,MAAM,CAAC,gCAAgC,CAAoB,CAAC;QAC9E,QAAQ,GAAG,GAAG,CAAC,KAAK,CAAC;IACvB,CAAC;IACD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,iBAAiB,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,GAAG,EAAE,QAAQ,EAAE,SAAS,GAAG,kBAAkB,EAAE,OAAO,GAAG,KAAK,EAAE,aAAa,EAAE,KAAK,EAAE,YAAY,EAAE,WAAW,EAAE,IAAI,EAAE,MAAM,EAAiB;IAC7M,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IACzB,MAAM,eAAe,GAAG,IAAI,eAAe,EAAE,CAAC;IAC9C,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,eAAe,CAAC,KAAK,EAAE,EAAE,SAAS,CAAC,CAAC;IAEnE,MAAM,GAAG,GAAG,YAAY,CAAC,QAAQ,CAAC,CAAC;IAEnC,MAAM,QAAQ,GAA2B,EAAE,CAAC;IAC5C,MAAM,iBAAiB,GAAa,EAAE,CAAC;IAEvC,6DAA6D;IAC7D,MAAM,UAAU,GAAG,KAAK,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC;QAC1C,CAAC,CAAC,oBAAoB,CAAC,KAAK,EAAE,YAAY,EAAE,CAAC,CAAC,WAAW,CAAC;QAC1D,CAAC,CAAC,IAAI,CAAC;IACT,MAAM,SAAS,GAAG,UAAU;QAC1B,CAAC,CAAC,EAAE,KAAK,EAAE,EAAE,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,UAAU,CAAC,QAAQ,CAAC,EAAE,CAAC,EAA6B,EAAE;QAC1F,CAAC,CAAC,EAAE,CAAC;IACP,MAAM,WAAW,GAAG,GAAwC,EAAE,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,GAAG,UAAU,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;IAEhH,IAAI,CAAC;QACH,MAAM,KAAK,GAAG,MAAM,WAAW,EAAE,CAAC;QAClC,sCAAsC;QACtC,gEAAgE;QAChE,MAAM,aAAa,GAAG,IAAI;YACxB,CAAC,CAAC,EAAE,MAAM,EAAE,EAAc,EAAE,eAAe,EAAE,CAAC,GAAG,CAAC,EAAE;YACpD,CAAC,CAAC,wBAAwB,CAAC,aAAa,CAAC,CAAC;QAC5C,6CAA6C;QAC7C,yEAAyE;QACzE,MAAM,eAAe,GAAG,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC;QAC9C,MAAM,UAAU,GAAG,eAAe,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,eAAe,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACtE,MAAM,MAAM,GAAG,KAAK,CAAC;YACnB,MAAM;YACN,OAAO,EAAE;gBACP,KAAK;gBACL,YAAY,EAAE,MAAM,IAAI,SAAS;gBACjC,GAAG,EAAE,GAAG,IAAI,OAAO,CAAC,GAAG,EAAE;gBACzB,cAAc,EAAE,mBAAmB;gBACnC,+BAA+B,EAAE,IAAI;gBACrC,eAAe;gBACf,GAAG;gBACH,GAAG,aAAa;gBAChB,GAAG,UAAU;gBACb,GAAG,SAAS;aACb;SACF,CAAC,CAAC;QAEH,MAAM,UAAU,GAA6B,EAAE,CAAC;QAChD,IAAI,KAAK,EAAE,MAAM,GAAG,IAAI,MAAM,EAAE,CAAC;YAC/B,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YACnB,iBAAiB,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC;YACnC,IAAI,wBAAwB,CAAC,GAAG,CAAC;gBAAE,UAAU,CAAC,IAAI,CAAC,GAA6B,CAAC,CAAC;QACpF,CAAC;QAED,YAAY,CAAC,KAAK,CAAC,CAAC;QAEpB,IAAI,OAAO,EAAE,CAAC;YACZ,IAAI,CAAC;gBACH,MAAM,QAAQ,GAAG,IAAI,CAAC,MAAM,EAAE,EAAE,WAAW,CAAC,CAAC;gBAC7C,IAAI,CAAC;oBAAC,SAAS,CAAC,QAAQ,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;gBAAC,CAAC;gBAAC,MAAM,CAAC,CAAC,CAAC;gBAC3D,MAAM,SAAS,GAAG,IAAI,CAAC,QAAQ,EAAE,cAAc,IAAI,CAAC,GAAG,EAAE,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC;gBAC5G,aAAa,CAAC,SAAS,EAAE,IAAI,CAAC,SAAS,CAAC,QAAQ,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;gBAC5D,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,wBAAwB,SAAS,KAAK,QAAQ,CAAC,MAAM,cAAc,CAAC,CAAC;YAC5F,CAAC;YAAC,MAAM,CAAC,CAAC,CAAC;QACb,CAAC;QAED,IAAI,UAAU,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC5B,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC;YACtC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,0DAA0D,QAAQ,CAAC,MAAM,cAAc,CAAC,CAAC;YAC9G,OAAO;gBACL,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,4BAA4B;gBAC9C,UAAU,EAAE,aAAa,EAAE,CAAC;gBAC5B,WAAW,EAAE,CAAC,EAAE,YAAY,EAAE,CAAC,EAAE,eAAe,EAAE,CAAC,EAAE,mBAAmB,EAAE,CAAC;gBAC3E,OAAO,EAAE,CAAC,EAAE,MAAM,EAAE,IAAI,EAAE,UAAU,EAAE,OAAO,EAAE,QAAQ,EAAE,CAAC;gBAC1D,GAAG,CAAC,WAAW,EAAE,IAAI,EAAE,SAAS,EAAE,WAAW,EAAG,EAAE,CAAC;aACpD,CAAC;QACJ,CAAC;QAED,IAAI,MAAM,GAAG,EAAE,CAAC;QAChB,IAAI,gBAAgB,GAAG,CAAC,CAAC;QACzB,IAAI,iBAAiB,GAAG,CAAC,CAAC;QAC1B,IAAI,oBAAoB,GAAG,CAAC,CAAC;QAC7B,IAAI,wBAAwB,GAAG,CAAC,CAAC;QACjC,IAAI,YAAY,GAAG,CAAC,CAAC;QACrB,IAAI,kBAAkB,GAAG,CAAC,CAAC;QAC3B,IAAI,aAAa,GAAG,CAAC,CAAC;QACtB,IAAI,WAAW,GAAG,SAAS,CAAC;QAE5B,KAAK,MAAM,CAAC,IAAI,UAAU,EAAE,CAAC;YAC3B,MAAM,IAAI,CAAC,CAAC,MAAM,IAAI,EAAE,CAAC;YACzB,8EAA8E;YAC9E,MAAM,UAAU,GAAI,CAAwC,CAAC,UAA2J,CAAC;YACzN,IAAI,UAAU,EAAE,CAAC;gBACf,KAAK,MAAM,EAAE,IAAI,MAAM,CAAC,MAAM,CAAC,UAAU,CAAC,EAAE,CAAC;oBAC3C,gBAAgB,IAAI,EAAE,CAAC,WAAW,IAAI,CAAC,CAAC;oBACxC,iBAAiB,IAAI,EAAE,CAAC,YAAY,IAAI,CAAC,CAAC;oBAC1C,oBAAoB,IAAI,EAAE,CAAC,oBAAoB,IAAI,CAAC,CAAC;oBACrD,wBAAwB,IAAI,EAAE,CAAC,wBAAwB,IAAI,CAAC,CAAC;gBAC/D,CAAC;YACH,CAAC;iBAAM,CAAC;gBACN,gBAAgB,IAAI,CAAC,CAAC,KAAK,EAAE,YAAY,IAAI,CAAC,CAAC;gBAC/C,iBAAiB,IAAI,CAAC,CAAC,KAAK,EAAE,aAAa,IAAI,CAAC,CAAC;gBACjD,oBAAoB,IAAI,CAAC,CAAC,KAAK,EAAE,uBAAuB,IAAI,CAAC,CAAC;gBAC9D,wBAAwB,IAAI,CAAC,CAAC,KAAK,EAAE,2BAA2B,IAAI,CAAC,CAAC;YACxE,CAAC;YACD,YAAY,IAAI,CAAC,CAAC,cAAc,IAAI,CAAC,CAAC;YACtC,kBAAkB,IAAI,CAAC,CAAC,eAAe,IAAI,CAAC,CAAC;YAC7C,aAAa,IAAI,CAAC,CAAC,SAAS,IAAI,CAAC,CAAC;YAClC,WAAW,GAAG,CAAC,CAAC,OAAO,IAAI,WAAW,CAAC;QACzC,CAAC;QAED,MAAM,UAAU,GAAG,UAAU,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,WAAW,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC,CAAC;QACzF,MAAM,KAAK,GAAG,iBAAiB,CAAC,QAAQ,EAAE,iBAAiB,CAAC,CAAC;QAE7D,IAAI,WAAW,KAAK,SAAS,EAAE,CAAC;YAC9B,MAAM,OAAO,GAAG,UAAU,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;YAClD,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,EAAE,IAAI,CAAC,IAAI,CAAC,IAAI,OAAO,CAAC,OAAO,CAAC;YAC7D,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,8BAA8B,OAAO,CAAC,OAAO,KAAK,MAAM,IAAI,CAAC,CAAC;YACnF,OAAO;gBACL,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM;gBACxB,UAAU;gBACV,aAAa,EAAE,kBAAkB;gBACjC,WAAW,EAAE,gBAAgB;gBAC7B,YAAY,EAAE,iBAAiB;gBAC/B,eAAe,EAAE,oBAAoB;gBACrC,mBAAmB,EAAE,wBAAwB;gBAC7C,OAAO,EAAE,YAAY;gBACrB,MAAM,EAAE,IAAI,EAAE,UAAU,EAAE,OAAO,EAAE,QAAQ,EAAE,aAAa;gBAC1D,YAAY,EAAE,KAAK,CAAC,YAAY;gBAChC,YAAY,EAAE,KAAK,CAAC,YAAY;gBAChC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,CAAC;gBACrD,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,SAAS,EAAE,KAAK,CAAC,SAAS,EAAE,CAAC;gBACjE,GAAG,CAAC,WAAW,EAAE,IAAI,EAAE,SAAS,EAAE,WAAW,EAAG,EAAE,CAAC;aACpD,CAAC;QACJ,CAAC;QAED,OAAO;YACL,EAAE,EAAE,IAAI;YACR,UAAU;YACV,aAAa,EAAE,kBAAkB;YACjC,WAAW,EAAE,gBAAgB;YAC7B,YAAY,EAAE,iBAAiB;YAC/B,eAAe,EAAE,oBAAoB;YACrC,mBAAmB,EAAE,wBAAwB;YAC7C,OAAO,EAAE,YAAY;YACrB,MAAM;YACN,UAAU,EAAE,KAAK;YACjB,QAAQ,EAAE,aAAa;YACvB,YAAY,EAAE,KAAK,CAAC,YAAY;YAChC,YAAY,EAAE,KAAK,CAAC,YAAY;YAChC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,CAAC;YACrD,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,SAAS,EAAE,KAAK,CAAC,SAAS,EAAE,CAAC;YACjE,GAAG,CAAC,WAAW,EAAE,IAAI,EAAE,SAAS,EAAE,WAAW,EAAG,EAAE,CAAC;SACpD,CAAC;IACJ,CAAC;IAAC,OAAO,GAAY,EAAE,CAAC;QACtB,YAAY,CAAC,KAAK,CAAC,CAAC;QACpB,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC;QACtC,MAAM,OAAO,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC;QACjC,MAAM,OAAO,GAAG,OAAO,CAAC,IAAI,KAAK,YAAY,CAAC;QAC9C,IAAI,OAAO,EAAE,CAAC;YACZ,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,8CAA8C,SAAS,GAAG,IAAI,KAAK,CAAC,CAAC;QAC5F,CAAC;aAAM,CAAC;YACN,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,oCAAoC,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QAClF,CAAC;QACD,OAAO;YACL,EAAE,EAAE,KAAK;YACT,KAAK,EAAE,OAAO,CAAC,CAAC,CAAC,6BAA6B,SAAS,GAAG,IAAI,GAAG,CAAC,CAAC,CAAC,YAAY,CAAC,GAAG,CAAC;YACrF,UAAU,EAAE,aAAa,EAAE,CAAC;YAC5B,WAAW,EAAE,CAAC,EAAE,YAAY,EAAE,CAAC,EAAE,eAAe,EAAE,CAAC,EAAE,mBAAmB,EAAE,CAAC;YAC3E,OAAO,EAAE,CAAC,EAAE,MAAM,EAAE,IAAI;YACxB,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO;YACzC,QAAQ,EAAE,CAAC;YACX,GAAG,CAAC,WAAW,EAAE,IAAI,EAAE,SAAS,EAAE,WAAW,EAAG,EAAE,CAAC;SACpD,CAAC;IACJ,CAAC;AACH,CAAC"}
|
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import { execFile, type ChildProcess } from 'node:child_process';
|
|
2
2
|
import type { ExecResult } from '../types/index.js';
|
|
3
3
|
export declare const execFileAsync: typeof execFile.__promisify__;
|
|
4
|
-
export declare const DEFAULT_MODEL = "
|
|
4
|
+
export declare const DEFAULT_MODEL = "opus";
|
|
5
5
|
export declare const JUDGE_MODEL = "haiku";
|
|
6
6
|
export declare const DEFAULT_TIMEOUT_MS = 120000;
|
|
7
7
|
export declare const MAX_BUFFER: number;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"shared.d.ts","sourceRoot":"","sources":["../../../src/executors/shared.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAS,KAAK,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAIxE,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAEpD,eAAO,MAAM,aAAa,+BAAsB,CAAC;
|
|
1
|
+
{"version":3,"file":"shared.d.ts","sourceRoot":"","sources":["../../../src/executors/shared.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAS,KAAK,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAIxE,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAEpD,eAAO,MAAM,aAAa,+BAAsB,CAAC;AAOjD,eAAO,MAAM,aAAa,SAAS,CAAC;AACpC,eAAO,MAAM,WAAW,UAAU,CAAC;AACnC,eAAO,MAAM,kBAAkB,SAAU,CAAC;AAC1C,eAAO,MAAM,UAAU,QAAmB,CAAC;AAE3C,MAAM,WAAW,UAAU;IACzB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,uBAAuB,CAAC,EAAE,MAAM,CAAC;IACjC,2BAA2B,CAAC,EAAE,MAAM,CAAC;CACtC;AAED,MAAM,WAAW,iBAAiB;IAChC,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,KAAK,CAAC,EAAE,UAAU,CAAC;IACnB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,WAAW;IAC1B,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,qBAAqB,CAAC,EAAE;QAAE,aAAa,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;CACpD;AAED,MAAM,WAAW,cAAc;IAC7B,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,OAAO,CAAC,EAAE,KAAK,CAAC;QAAE,OAAO,CAAC,EAAE;YAAE,OAAO,CAAC,EAAE,MAAM,CAAA;SAAE,CAAC;QAAC,aAAa,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC5E,KAAK,CAAC,EAAE;QAAE,OAAO,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;CAC9B;AAED,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,KAAK,CAAC,EAAE;QAAE,WAAW,CAAC,EAAE,MAAM,CAAC;QAAC,YAAY,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;CACzD;AAED,MAAM,WAAW,iBAAiB;IAChC,KAAK,CAAC,EAAE,UAAU,CAAC;IACnB,OAAO,CAAC,EAAE,KAAK,CAAC;QAAE,IAAI,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACnC,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,KAAK,CAAC,EAAE;QAAE,OAAO,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;CAC9B;AAED,MAAM,WAAW,qBAAqB;IACpC,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,GAAG,EAAE,MAAM,CAAC;IACZ,cAAc,EAAE,mBAAmB,CAAC;IACpC,+BAA+B,EAAE,IAAI,CAAC;IACtC,eAAe,EAAE,eAAe,CAAC;IACjC,GAAG,EAAE,MAAM,CAAC,UAAU,CAAC;CACxB;AAED,MAAM,WAAW,mBAAmB;IAClC,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,qBAAqB,CAAC;CAChC;AAED,MAAM,WAAW,oBAAoB;IACnC,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,CAAC,EAAE;QACR,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,OAAO,CAAC,EAAE,KAAK,CAAC;YACd,IAAI,EAAE,MAAM,CAAC;YACb,IAAI,CAAC,EAAE,MAAM,CAAC;YACd,EAAE,CAAC,EAAE,MAAM,CAAC;YACZ,IAAI,CAAC,EAAE,MAAM,CAAC;YACd,KAAK,CAAC,EAAE,OAAO,CAAC;SACjB,CAAC,CAAC;KACJ,CAAC;IACF,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,OAAO,CAAC,EAAE,MAAM,GAAG,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC1D,QAAQ,CAAC,EAAE,OAAO,CAAC;CACpB;AAED,MAAM,WAAW,sBAAuB,SAAQ,oBAAoB;IAClE,IAAI,EAAE,QAAQ,CAAC;IACf,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,KAAK,CAAC,EAAE,UAAU,CAAC;IACnB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,MAAM,CAAC,EAAE,MAAM,EAAE,CAAC;CACnB;AAED,MAAM,WAAW,eAAe;IAC9B,KAAK,EAAE,CAAC,IAAI,EAAE,mBAAmB,KAAK,aAAa,CAAC,oBAAoB,CAAC,CAAC;CAC3E;AAWD,MAAM,WAAW,UAAU;IACzB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE;QACN,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,mBAAmB,CAAC,EAAE,MAAM,CAAC;QAC7B,aAAa,CAAC,EAAE,MAAM,CAAC;QACvB,uBAAuB,CAAC,EAAE,MAAM,CAAC;KAClC,CAAC;IACF,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,WAAW,CAAC,EAAE,MAAM,CAAC;IAGrB,IAAI,CAAC,EAAE;QACL,EAAE,CAAC,EAAE,MAAM,CAAC;QACZ,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,iBAAiB,CAAC,EAAE,MAAM,CAAC;QAC3B,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;QAC1B,MAAM,CAAC,EAAE,MAAM,CAAC;QAChB,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,KAAK,CAAC,EAAE,MAAM,CAAC;QACf,OAAO,CAAC,EAAE,OAAO,EAAE,CAAC;QACpB,OAAO,CAAC,EAAE,KAAK,CAAC;YAAE,IAAI,CAAC,EAAE,MAAM,CAAC;YAAC,IAAI,CAAC,EAAE,MAAM,CAAA;SAAE,CAAC,CAAC;QAClD,MAAM,CAAC,EAAE,MAAM,CAAC;QAChB,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,SAAS,CAAC,EAAE,OAAO,CAAC;QACpB,MAAM,CAAC,EAAE,OAAO,CAAC;QACjB,OAAO,CAAC,EAAE,MAAM,CAAC;KAClB,CAAC;IACF,KAAK,CAAC,EAAE;QAAE,OAAO,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;IAC7B,EAAE,CAAC,EAAE,MAAM,CAAC;CACb;AAED,MAAM,WAAW,iBAAiB;IAChC,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;AAED,wBAAgB,WAAW,CAAC,GAAG,EAAE,OAAO,GAAG,iBAAiB,CAE3D;AAED,wBAAgB,YAAY,CAAC,GAAG,EAAE,OAAO,EAAE,QAAQ,GAAE,MAAwB,GAAG,MAAM,CAGrF;AAED,wBAAgB,SAAS,CAAC,CAAC,EAAE,OAAO,EAAE,MAAM,GAAG,CAAC,CAE/C;AAED,wBAAgB,YAAY,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,MAAM,CAAC,UAAU,CAcxE;AAED,wBAAgB,iBAAiB,CAAC,SAAS,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,GAAG,UAAU,CAenF;AAKD,wBAAgB,qBAAqB,CAAC,UAAU,EAAE,MAAM,GAAG,UAAU,CAepE;AA+CD,wBAAgB,4BAA4B,IAAI,IAAI,CAOnD;AAED,MAAM,WAAW,iBAAiB;IAChC,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,GAAG,IAAI,CAAC;IACpB,MAAM,EAAE,MAAM,CAAC,OAAO,GAAG,IAAI,CAAC;IAC9B,eAAe,EAAE,OAAO,CAAC;IACzB,cAAc,EAAE,MAAM,CAAC,OAAO,GAAG,IAAI,CAAC;CACvC;AAED,MAAM,WAAW,gBAAiB,SAAQ,KAAK;IAC7C,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,MAAM,CAAC,EAAE,MAAM,CAAC,OAAO,GAAG,IAAI,CAAC;IAC/B,eAAe,CAAC,EAAE,OAAO,CAAC;IAC1B,cAAc,CAAC,EAAE,MAAM,CAAC,OAAO,GAAG,IAAI,CAAC;CACxC;AAED,MAAM,WAAW,kBAAkB;IACjC,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,GAAG,CAAC,EAAE,MAAM,CAAC,UAAU,CAAC;IACxB,sEAAsE;IACtE,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,iDAAiD;IACjD,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,2DAA2D;IAC3D,WAAW,CAAC,EAAE,WAAW,CAAC;CAC3B;AAED;;;;;;GAMG;AACH,wBAAgB,0BAA0B,CACxC,OAAO,EAAE,MAAM,EACf,IAAI,EAAE,MAAM,EAAE,EACd,OAAO,GAAE,kBAAuB,GAC/B;IAAE,KAAK,EAAE,YAAY,CAAC;IAAC,IAAI,EAAE,OAAO,CAAC,iBAAiB,CAAC,CAAA;CAAE,CAmH3D"}
|
|
@@ -3,7 +3,12 @@ import { existsSync } from 'node:fs';
|
|
|
3
3
|
import { delimiter, join } from 'node:path';
|
|
4
4
|
import { promisify } from 'node:util';
|
|
5
5
|
export const execFileAsync = promisify(execFile);
|
|
6
|
-
|
|
6
|
+
// 系统默认任务执行模型。用 alias 'opus' 而非 pinned 'claude-opus-4-7',
|
|
7
|
+
// 让 Anthropic 推新一代 opus 时(opus 5 等)自动跟进,无需改代码。
|
|
8
|
+
// 副作用:跨 opus 代际的历史报告分数不可严格直接比 — 但 report.meta.executorRuntime
|
|
9
|
+
// 会记录每次跑实际解析到的模型 ID(如 'claude-opus-4-7'),renderer / verdict 据此识别版本切换。
|
|
10
|
+
// 子场景可显式覆盖:generator 走 'sonnet'(lean 路径不需要 opus 质量,且 sonnet 便宜 5x)。
|
|
11
|
+
export const DEFAULT_MODEL = 'opus';
|
|
7
12
|
export const JUDGE_MODEL = 'haiku';
|
|
8
13
|
export const DEFAULT_TIMEOUT_MS = 120_000;
|
|
9
14
|
export const MAX_BUFFER = 10 * 1024 * 1024;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"shared.js","sourceRoot":"","sources":["../../../src/executors/shared.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,KAAK,EAAqB,MAAM,oBAAoB,CAAC;AACxE,OAAO,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AACrC,OAAO,EAAE,SAAS,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAC5C,OAAO,EAAE,SAAS,EAAE,MAAM,WAAW,CAAC;AAGtC,MAAM,CAAC,MAAM,aAAa,GAAG,SAAS,CAAC,QAAQ,CAAC,CAAC;AAEjD,MAAM,CAAC,MAAM,aAAa,GAAG,
|
|
1
|
+
{"version":3,"file":"shared.js","sourceRoot":"","sources":["../../../src/executors/shared.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,KAAK,EAAqB,MAAM,oBAAoB,CAAC;AACxE,OAAO,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AACrC,OAAO,EAAE,SAAS,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAC5C,OAAO,EAAE,SAAS,EAAE,MAAM,WAAW,CAAC;AAGtC,MAAM,CAAC,MAAM,aAAa,GAAG,SAAS,CAAC,QAAQ,CAAC,CAAC;AAEjD,yDAAyD;AACzD,+CAA+C;AAC/C,8DAA8D;AAC9D,sEAAsE;AACtE,oEAAoE;AACpE,MAAM,CAAC,MAAM,aAAa,GAAG,MAAM,CAAC;AACpC,MAAM,CAAC,MAAM,WAAW,GAAG,OAAO,CAAC;AACnC,MAAM,CAAC,MAAM,kBAAkB,GAAG,OAAO,CAAC;AAC1C,MAAM,CAAC,MAAM,UAAU,GAAG,EAAE,GAAG,IAAI,GAAG,IAAI,CAAC;AAgJ3C,MAAM,UAAU,WAAW,CAAC,GAAY;IACtC,OAAO,OAAO,GAAG,KAAK,QAAQ,IAAI,GAAG,KAAK,IAAI,CAAC,CAAC,CAAC,GAAwB,CAAC,CAAC,CAAC,EAAE,CAAC;AACjF,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,GAAY,EAAE,WAAmB,eAAe;IAC3E,MAAM,OAAO,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC;IACjC,OAAO,OAAO,CAAC,OAAO,IAAI,QAAQ,CAAC;AACrC,CAAC;AAED,MAAM,UAAU,SAAS,CAAI,OAAe;IAC1C,OAAO,IAAI,CAAC,KAAK,CAAC,OAAO,CAAM,CAAC;AAClC,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,QAAwB;IACnD,MAAM,QAAQ,GAAG,OAAO,CAAC,GAAG,CAAC,aAAa,IAAI,SAAS,CAAC;IACxD,MAAM,GAAG,GAAsB,QAAQ;QACrC,CAAC,CAAC,EAAE,GAAG,OAAO,CAAC,GAAG,EAAE,kBAAkB,EAAE,QAAQ,EAAE;QAClD,CAAC,CAAC,EAAE,GAAG,OAAO,CAAC,GAAG,EAAE,CAAC;IAEvB,IAAI,QAAQ,EAAE,CAAC;QACb,MAAM,OAAO,GAAG,IAAI,CAAC,QAAQ,EAAE,cAAc,EAAE,MAAM,CAAC,CAAC;QACvD,IAAI,UAAU,CAAC,OAAO,CAAC,EAAE,CAAC;YACxB,GAAG,CAAC,IAAI,GAAG,GAAG,OAAO,GAAG,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,GAAG,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;QACjE,CAAC;IACH,CAAC;IAED,OAAO,GAAG,CAAC;AACb,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,SAAiB,EAAE,UAAkB;IACrE,OAAO;QACL,EAAE,EAAE,KAAK;QACT,KAAK,EAAE,6BAA6B,SAAS,GAAG,IAAI,GAAG;QACvD,UAAU;QACV,aAAa,EAAE,CAAC;QAChB,WAAW,EAAE,CAAC;QACd,YAAY,EAAE,CAAC;QACf,eAAe,EAAE,CAAC;QAClB,mBAAmB,EAAE,CAAC;QACtB,OAAO,EAAE,CAAC;QACV,MAAM,EAAE,IAAI;QACZ,UAAU,EAAE,SAAS;QACrB,QAAQ,EAAE,CAAC;KACZ,CAAC;AACJ,CAAC;AAED,iFAAiF;AACjF,2EAA2E;AAC3E,2DAA2D;AAC3D,MAAM,UAAU,qBAAqB,CAAC,UAAkB;IACtD,OAAO;QACL,EAAE,EAAE,KAAK;QACT,KAAK,EAAE,gCAAgC;QACvC,UAAU;QACV,aAAa,EAAE,CAAC;QAChB,WAAW,EAAE,CAAC;QACd,YAAY,EAAE,CAAC;QACf,eAAe,EAAE,CAAC;QAClB,mBAAmB,EAAE,CAAC;QACtB,OAAO,EAAE,CAAC;QACV,MAAM,EAAE,IAAI;QACZ,UAAU,EAAE,aAAa;QACzB,QAAQ,EAAE,CAAC;KACZ,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,qEAAqE;AACrE,0DAA0D;AAC1D,8EAA8E;AAC9E,EAAE;AACF,QAAQ;AACR,sEAAsE;AACtE,2DAA2D;AAC3D,2EAA2E;AAC3E,8EAA8E;AAC9E,gFAAgF;AAChF,gFAAgF;AAChF,qBAAqB;AACrB,+DAA+D;AAE/D,MAAM,cAAc,GAAG,IAAI,GAAG,EAAgB,CAAC;AAC/C,IAAI,uBAAuB,GAAG,KAAK,CAAC;AACpC,IAAI,YAAY,GAAG,KAAK,CAAC;AACzB,MAAM,gBAAgB,GAAG,GAAG,CAAC;AAE7B,SAAS,iBAAiB;IACxB,IAAI,YAAY;QAAE,OAAO;IACzB,YAAY,GAAG,IAAI,CAAC;IACpB,KAAK,MAAM,KAAK,IAAI,cAAc,EAAE,CAAC;QACnC,IAAI,CAAC;YAAC,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QAAC,CAAC;QAAC,MAAM,CAAC,CAAC,kBAAkB,CAAC,CAAC;QAC3D,UAAU,CAAC,GAAG,EAAE;YACd,IAAI,CAAC;gBAAC,IAAI,KAAK,CAAC,QAAQ,KAAK,IAAI,IAAI,KAAK,CAAC,UAAU,KAAK,IAAI;oBAAE,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;YAAC,CAAC;YAAC,MAAM,CAAC,CAAC,KAAK,CAAC,CAAC;QAC1G,CAAC,EAAE,gBAAgB,CAAC,CAAC,KAAK,EAAE,CAAC;IAC/B,CAAC;IACD,yEAAyE;IACzE,OAAO,CAAC,cAAc,CAAC,QAAQ,EAAE,aAAa,CAAC,CAAC;IAChD,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,EAAE,QAAQ,CAAC,CAAC;AACtC,CAAC;AAED,SAAS,aAAa;IACpB,iBAAiB,EAAE,CAAC;AACtB,CAAC;AAED,SAAS,oBAAoB;IAC3B,IAAI,uBAAuB;QAAE,OAAO;IACpC,uBAAuB,GAAG,IAAI,CAAC;IAC/B,OAAO,CAAC,EAAE,CAAC,QAAQ,EAAE,aAAa,CAAC,CAAC;AACtC,CAAC;AAED,oCAAoC;AACpC,MAAM,UAAU,4BAA4B;IAC1C,cAAc,CAAC,KAAK,EAAE,CAAC;IACvB,IAAI,uBAAuB,EAAE,CAAC;QAC5B,OAAO,CAAC,cAAc,CAAC,QAAQ,EAAE,aAAa,CAAC,CAAC;QAChD,uBAAuB,GAAG,KAAK,CAAC;IAClC,CAAC;IACD,YAAY,GAAG,KAAK,CAAC;AACvB,CAAC;AA+BD;;;;;;GAMG;AACH,MAAM,UAAU,0BAA0B,CACxC,OAAe,EACf,IAAc,EACd,UAA8B,EAAE;IAEhC,MAAM,EAAE,GAAG,EAAE,GAAG,EAAE,SAAS,EAAE,SAAS,GAAG,UAAU,EAAE,WAAW,EAAE,GAAG,OAAO,CAAC;IAC7E,oBAAoB,EAAE,CAAC;IAEvB,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,EAAE,IAAI,EAAE;QACjC,KAAK,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC;QAC/B,GAAG,CAAC,GAAG,IAAI,EAAE,GAAG,EAAE,CAAC;QACnB,GAAG,CAAC,GAAG,IAAI,EAAE,GAAG,EAAE,CAAC;KACpB,CAAC,CAAC;IAEH,cAAc,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;IAE1B,IAAI,MAAM,GAAG,EAAE,CAAC;IAChB,IAAI,MAAM,GAAG,EAAE,CAAC;IAChB,IAAI,cAAc,GAAG,KAAK,CAAC;IAC3B,IAAI,eAAe,GAAG,KAAK,CAAC;IAC5B,IAAI,oBAAoB,GAA0B,IAAI,CAAC;IACvD,IAAI,UAAU,GAA0B,IAAI,CAAC;IAC7C,IAAI,YAAY,GAA0B,IAAI,CAAC;IAE/C,SAAS,aAAa,CAAC,MAAsC;QAC3D,sDAAsD;QACtD,IAAI,MAAM,KAAK,SAAS,IAAI,CAAC,oBAAoB;YAAE,eAAe,GAAG,IAAI,CAAC;QAC1E,IAAI,MAAM,KAAK,OAAO,IAAI,CAAC,eAAe;YAAE,oBAAoB,GAAG,SAAS,CAAC;QAC7E,IAAI,CAAC;YAAC,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QAAC,CAAC;QAAC,MAAM,CAAC,CAAC,kBAAkB,CAAC,CAAC;QAC3D,IAAI,UAAU;YAAE,OAAO;QACvB,gDAAgD;QAChD,UAAU,GAAG,UAAU,CAAC,GAAG,EAAE;YAC3B,IAAI,CAAC;gBAAC,IAAI,KAAK,CAAC,QAAQ,KAAK,IAAI,IAAI,KAAK,CAAC,UAAU,KAAK,IAAI;oBAAE,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;YAAC,CAAC;YAAC,MAAM,CAAC,CAAC,KAAK,CAAC,CAAC;QAC1G,CAAC,EAAE,gBAAgB,CAAC,CAAC;QACrB,UAAU,CAAC,KAAK,EAAE,CAAC;IACrB,CAAC;IAED,IAAI,SAAS,IAAI,SAAS,GAAG,CAAC,EAAE,CAAC;QAC/B,YAAY,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,aAAa,CAAC,SAAS,CAAC,EAAE,SAAS,CAAC,CAAC;QACrE,YAAY,CAAC,KAAK,EAAE,CAAC;IACvB,CAAC;IAED,IAAI,aAAa,GAAwB,IAAI,CAAC;IAC9C,IAAI,WAAW,EAAE,CAAC;QAChB,aAAa,GAAG,GAAS,EAAE,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC;QACnD,WAAW,CAAC,gBAAgB,CAAC,OAAO,EAAE,aAAa,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;IACvE,CAAC;IAED,KAAK,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,EAAE,CAAC,KAAa,EAAE,EAAE;QACzC,MAAM,IAAI,KAAK,CAAC,QAAQ,EAAE,CAAC;QAC3B,IAAI,MAAM,CAAC,MAAM,GAAG,SAAS,IAAI,CAAC,cAAc,EAAE,CAAC;YACjD,cAAc,GAAG,IAAI,CAAC;YACtB,+DAA+D;YAC/D,aAAa,CAAC,QAAQ,CAAC,CAAC;QAC1B,CAAC;IACH,CAAC,CAAC,CAAC;IACH,KAAK,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,EAAE,CAAC,KAAa,EAAE,EAAE;QACzC,MAAM,IAAI,KAAK,CAAC,QAAQ,EAAE,CAAC;IAC7B,CAAC,CAAC,CAAC;IAEH,SAAS,OAAO;QACd,cAAc,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QAC7B,IAAI,YAAY;YAAE,YAAY,CAAC,YAAY,CAAC,CAAC;QAC7C,IAAI,UAAU;YAAE,YAAY,CAAC,UAAU,CAAC,CAAC;QACzC,IAAI,aAAa,IAAI,WAAW;YAAE,WAAW,CAAC,mBAAmB,CAAC,OAAO,EAAE,aAAa,CAAC,CAAC;IAC5F,CAAC;IAED,MAAM,IAAI,GAAG,IAAI,OAAO,CAAoB,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QAC9D,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,GAAU,EAAE,EAAE;YAC/B,OAAO,EAAE,CAAC;YACV,MAAM,CAAC,GAAG,GAAuB,CAAC;YAClC,CAAC,CAAC,MAAM,GAAG,MAAM,CAAC;YAClB,CAAC,CAAC,MAAM,GAAG,MAAM,CAAC;YAClB,MAAM,CAAC,CAAC,CAAC,CAAC;QACZ,CAAC,CAAC,CAAC;QACH,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,IAAmB,EAAE,MAA6B,EAAE,EAAE;YACvE,OAAO,EAAE,CAAC;YACV,MAAM,SAAS,GAA0B,oBAAoB;mBACxD,CAAC,MAAM,IAAI,CAAC,eAAe,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;YAClD,MAAM,MAAM,GAAsB;gBAChC,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM;gBAC5B,eAAe;gBACf,cAAc,EAAE,SAAS;aAC1B,CAAC;YACF,uDAAuD;YACvD,IAAI,cAAc,EAAE,CAAC;gBACnB,MAAM,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,qBAAqB,SAAS,YAAY,CAAC,EAAE,MAAM,CAAqB,CAAC;gBAC3G,MAAM,CAAC,CAAC,CAAC,CAAC;gBACV,OAAO;YACT,CAAC;YACD,mEAAmE;YACnE,6DAA6D;YAC7D,+DAA+D;YAC/D,IAAI,IAAI,KAAK,CAAC,EAAE,CAAC;gBACf,OAAO,CAAC,MAAM,CAAC,CAAC;gBAChB,OAAO;YACT,CAAC;YACD,IAAI,eAAe,EAAE,CAAC;gBACpB,MAAM,IAAI,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC;gBAC7D,MAAM,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,6BAA6B,IAAI,GAAG,CAAC,EAAE,MAAM,CAAqB,CAAC;gBACrG,MAAM,CAAC,CAAC,CAAC,CAAC;gBACV,OAAO;YACT,CAAC;YACD,IAAI,SAAS,EAAE,CAAC;gBACd,MAAM,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,mCAAmC,SAAS,EAAE,CAAC,EAAE,MAAM,CAAqB,CAAC;gBAC/G,MAAM,CAAC,CAAC,CAAC,CAAC;gBACV,OAAO;YACT,CAAC;YACD,IAAI,IAAI,KAAK,IAAI,EAAE,CAAC;gBAClB,MAAM,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,GAAG,OAAO,qBAAqB,IAAI,EAAE,CAAC,EAAE,MAAM,CAAqB,CAAC;gBACtG,MAAM,CAAC,CAAC,CAAC,CAAC;gBACV,OAAO;YACT,CAAC;YACD,uDAAuD;YACvD,OAAO,CAAC,MAAM,CAAC,CAAC;QAClB,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC;AACzB,CAAC"}
|
|
@@ -29,6 +29,11 @@ export declare function runAssertions(output: string, assertions: Assertion[], c
|
|
|
29
29
|
durationMs?: number;
|
|
30
30
|
numTurns?: number;
|
|
31
31
|
toolCalls?: ToolCallInfo[];
|
|
32
|
+
mockStats?: {
|
|
33
|
+
hits: number;
|
|
34
|
+
misses: number;
|
|
35
|
+
perMock: Record<string, number>;
|
|
36
|
+
};
|
|
32
37
|
}): AssertionResults;
|
|
33
38
|
export declare function runAsyncAssertions(output: string, assertions: Assertion[], { executor, judgeModel, sample, samplesDir }: AsyncAssertionContext): Promise<AssertionResults>;
|
|
34
39
|
//# sourceMappingURL=assertions.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"assertions.d.ts","sourceRoot":"","sources":["../../../src/grading/assertions.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,SAAS,EAAmB,gBAAgB,EAAE,UAAU,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,mBAAmB,CAAC;AAMxH,eAAO,MAAM,qBAAqB,aAQhC,CAAC;AAEH,MAAM,WAAW,qBAAqB;IACpC,QAAQ,EAAE,UAAU,CAAC;IACrB,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,EAAE,MAAM,CAAC;CACpB;AAwBD,wBAAgB,YAAY,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,CAElD;AAkCD;;;;;GAKG;AACH,wBAAgB,MAAM,CAAC,SAAS,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,GAAG,MAAM,CAK9E;AAED,kEAAkE;AAClE,wBAAgB,WAAW,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,GAAG,MAAM,CAmBxD;AAED;;;;;GAKG;AACH,wBAAgB,IAAI,CAAC,SAAS,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,EAAE,IAAI,SAAI,GAAG,MAAM,CAgB3E;AAED,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,OAAO,CAQ1F;
|
|
1
|
+
{"version":3,"file":"assertions.d.ts","sourceRoot":"","sources":["../../../src/grading/assertions.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,SAAS,EAAmB,gBAAgB,EAAE,UAAU,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,mBAAmB,CAAC;AAMxH,eAAO,MAAM,qBAAqB,aAQhC,CAAC;AAEH,MAAM,WAAW,qBAAqB;IACpC,QAAQ,EAAE,UAAU,CAAC;IACrB,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,EAAE,MAAM,CAAC;CACpB;AAwBD,wBAAgB,YAAY,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,CAElD;AAkCD;;;;;GAKG;AACH,wBAAgB,MAAM,CAAC,SAAS,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,GAAG,MAAM,CAK9E;AAED,kEAAkE;AAClE,wBAAgB,WAAW,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,GAAG,MAAM,CAmBxD;AAED;;;;;GAKG;AACH,wBAAgB,IAAI,CAAC,SAAS,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,EAAE,IAAI,SAAI,GAAG,MAAM,CAgB3E;AAED,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,OAAO,CAQ1F;AA8ID,wBAAgB,aAAa,CAC3B,MAAM,EAAE,MAAM,EACd,UAAU,EAAE,SAAS,EAAE,EACvB,OAAO,GAAE;IAAE,OAAO,CAAC,EAAE,MAAM,CAAC;IAAC,UAAU,CAAC,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAAC,SAAS,CAAC,EAAE,YAAY,EAAE,CAAC;IAAC,SAAS,CAAC,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;KAAE,CAAA;CAAO,GACpL,gBAAgB,CA8BlB;AAED,wBAAsB,kBAAkB,CAAC,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,SAAS,EAAE,EAAE,EAAE,QAAQ,EAAE,UAAU,EAAE,MAAM,EAAE,UAAU,EAAE,EAAE,qBAAqB,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAqGhL"}
|
|
@@ -227,6 +227,32 @@ function evalAssertion(output, assertion, ctx) {
|
|
|
227
227
|
return toolCalls.some((tc) => tc.tool.toLowerCase() === targetTool &&
|
|
228
228
|
JSON.stringify(tc.input || '').toLowerCase().includes(expected));
|
|
229
229
|
}
|
|
230
|
+
case 'tool_input_not_contains': {
|
|
231
|
+
const sep = String(assertion.value).indexOf(':');
|
|
232
|
+
// value 必须是 "Tool:needle" 格式。无冒号(generator 偶尔产 "--force" / "lastTaskPatrol"
|
|
233
|
+
// 这种裸 needle)在 _contains 语义里算 false(没法判定哪个工具),但 _not_contains 语义
|
|
234
|
+
// 应该 trivially pass(没有匹配 = 没传 needle = 满足"不应包含"约束)。否则会变成
|
|
235
|
+
// 假阳性失败,扭曲通过率。loader 已加格式校验拒绝这种 value,这里是 grader 兜底。
|
|
236
|
+
if (sep <= 0)
|
|
237
|
+
return true;
|
|
238
|
+
const targetTool = String(assertion.value).slice(0, sep).toLowerCase();
|
|
239
|
+
const expected = String(assertion.value).slice(sep + 1).toLowerCase();
|
|
240
|
+
return !toolCalls.some((tc) => tc.tool.toLowerCase() === targetTool &&
|
|
241
|
+
JSON.stringify(tc.input || '').toLowerCase().includes(expected));
|
|
242
|
+
}
|
|
243
|
+
// mock_hit:校验"驱动流程"——指定 mock 是否被命中至少 N 次
|
|
244
|
+
// value 是 mockStats.perMock 的 key,格式 "Tool:N"(N 为 sample.mocks 数组里的 1-based 序号)
|
|
245
|
+
// threshold 可选,默认 >=1
|
|
246
|
+
// 例: { type: 'mock_hit', value: 'Bash:2' } 检查第 2 条 Bash mock 被调过
|
|
247
|
+
case 'mock_hit': {
|
|
248
|
+
const stats = ctx.mockStats;
|
|
249
|
+
if (!stats)
|
|
250
|
+
return false;
|
|
251
|
+
const key = String(assertion.value);
|
|
252
|
+
const hits = stats.perMock[key] ?? 0;
|
|
253
|
+
const min = assertion.threshold ?? 1;
|
|
254
|
+
return hits >= min;
|
|
255
|
+
}
|
|
230
256
|
case 'rouge_n_min':
|
|
231
257
|
return rougeN(output, String(assertion.reference ?? assertion.value ?? ''), assertion.n ?? 1)
|
|
232
258
|
>= (assertion.threshold ?? 0.5);
|
|
@@ -244,7 +270,7 @@ export function runAssertions(output, assertions, context = {}) {
|
|
|
244
270
|
const outputLower = output.toLowerCase();
|
|
245
271
|
const toolCalls = context.toolCalls || [];
|
|
246
272
|
const toolNames = toolCalls.map((tc) => tc.tool.toLowerCase());
|
|
247
|
-
const ctx = { outputLower, toolCalls, toolNames, costUSD: context.costUSD, durationMs: context.durationMs, numTurns: context.numTurns };
|
|
273
|
+
const ctx = { outputLower, toolCalls, toolNames, costUSD: context.costUSD, durationMs: context.durationMs, numTurns: context.numTurns, mockStats: context.mockStats };
|
|
248
274
|
const details = [];
|
|
249
275
|
for (const assertion of assertions) {
|
|
250
276
|
const weight = assertion.weight ?? 1;
|