@prismatic-io/lux 0.0.1
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/LICENSE +21 -0
- package/lib/answerers/claude-code/index.d.ts +17 -0
- package/lib/answerers/claude-code/index.d.ts.map +1 -0
- package/lib/answerers/claude-code/index.js +280 -0
- package/lib/answerers/claude-code/index.js.map +1 -0
- package/lib/answerers/persona/index.d.ts +40 -0
- package/lib/answerers/persona/index.d.ts.map +1 -0
- package/lib/answerers/persona/index.js +265 -0
- package/lib/answerers/persona/index.js.map +1 -0
- package/lib/answerers/scripted/index.d.ts +8 -0
- package/lib/answerers/scripted/index.d.ts.map +1 -0
- package/lib/answerers/scripted/index.js +163 -0
- package/lib/answerers/scripted/index.js.map +1 -0
- package/lib/answerers/shared.d.ts +13 -0
- package/lib/answerers/shared.d.ts.map +1 -0
- package/lib/answerers/shared.js +17 -0
- package/lib/answerers/shared.js.map +1 -0
- package/lib/answerers/terminal/index.d.ts +9 -0
- package/lib/answerers/terminal/index.d.ts.map +1 -0
- package/lib/answerers/terminal/index.js +97 -0
- package/lib/answerers/terminal/index.js.map +1 -0
- package/lib/assertions/core/command-exits-zero.d.ts +21 -0
- package/lib/assertions/core/command-exits-zero.d.ts.map +1 -0
- package/lib/assertions/core/command-exits-zero.js +87 -0
- package/lib/assertions/core/command-exits-zero.js.map +1 -0
- package/lib/assertions/core/event-checks.d.ts +57 -0
- package/lib/assertions/core/event-checks.d.ts.map +1 -0
- package/lib/assertions/core/event-checks.js +123 -0
- package/lib/assertions/core/event-checks.js.map +1 -0
- package/lib/assertions/core/file-checks.d.ts +103 -0
- package/lib/assertions/core/file-checks.d.ts.map +1 -0
- package/lib/assertions/core/file-checks.js +288 -0
- package/lib/assertions/core/file-checks.js.map +1 -0
- package/lib/assertions/core/helpers.d.ts +66 -0
- package/lib/assertions/core/helpers.d.ts.map +1 -0
- package/lib/assertions/core/helpers.js +170 -0
- package/lib/assertions/core/helpers.js.map +1 -0
- package/lib/assertions/core/index.d.ts +15 -0
- package/lib/assertions/core/index.d.ts.map +1 -0
- package/lib/assertions/core/index.js +56 -0
- package/lib/assertions/core/index.js.map +1 -0
- package/lib/assertions/core/json-pointer-equals.d.ts +19 -0
- package/lib/assertions/core/json-pointer-equals.d.ts.map +1 -0
- package/lib/assertions/core/json-pointer-equals.js +71 -0
- package/lib/assertions/core/json-pointer-equals.js.map +1 -0
- package/lib/assertions/core/output-checks.d.ts +46 -0
- package/lib/assertions/core/output-checks.d.ts.map +1 -0
- package/lib/assertions/core/output-checks.js +121 -0
- package/lib/assertions/core/output-checks.js.map +1 -0
- package/lib/assertions/core/predicate.d.ts +14 -0
- package/lib/assertions/core/predicate.d.ts.map +1 -0
- package/lib/assertions/core/predicate.js +55 -0
- package/lib/assertions/core/predicate.js.map +1 -0
- package/lib/assertions/core/resource-checks.d.ts +71 -0
- package/lib/assertions/core/resource-checks.d.ts.map +1 -0
- package/lib/assertions/core/resource-checks.js +159 -0
- package/lib/assertions/core/resource-checks.js.map +1 -0
- package/lib/assertions/core/safe-regex.d.ts +23 -0
- package/lib/assertions/core/safe-regex.d.ts.map +1 -0
- package/lib/assertions/core/safe-regex.js +102 -0
- package/lib/assertions/core/safe-regex.js.map +1 -0
- package/lib/assertions/core/tool-called-with.d.ts +21 -0
- package/lib/assertions/core/tool-called-with.d.ts.map +1 -0
- package/lib/assertions/core/tool-called-with.js +69 -0
- package/lib/assertions/core/tool-called-with.js.map +1 -0
- package/lib/assertions/core/tool-calls.d.ts +22 -0
- package/lib/assertions/core/tool-calls.d.ts.map +1 -0
- package/lib/assertions/core/tool-calls.js +17 -0
- package/lib/assertions/core/tool-calls.js.map +1 -0
- package/lib/assertions/core/tool-invocation.d.ts +36 -0
- package/lib/assertions/core/tool-invocation.d.ts.map +1 -0
- package/lib/assertions/core/tool-invocation.js +170 -0
- package/lib/assertions/core/tool-invocation.js.map +1 -0
- package/lib/assertions/core/tool-result-checks.d.ts +59 -0
- package/lib/assertions/core/tool-result-checks.d.ts.map +1 -0
- package/lib/assertions/core/tool-result-checks.js +193 -0
- package/lib/assertions/core/tool-result-checks.js.map +1 -0
- package/lib/assertions/mcp/discovery-checks.d.ts +36 -0
- package/lib/assertions/mcp/discovery-checks.d.ts.map +1 -0
- package/lib/assertions/mcp/discovery-checks.js +122 -0
- package/lib/assertions/mcp/discovery-checks.js.map +1 -0
- package/lib/assertions/mcp/index.d.ts +10 -0
- package/lib/assertions/mcp/index.d.ts.map +1 -0
- package/lib/assertions/mcp/index.js +25 -0
- package/lib/assertions/mcp/index.js.map +1 -0
- package/lib/assertions/rubric/index.d.ts +104 -0
- package/lib/assertions/rubric/index.d.ts.map +1 -0
- package/lib/assertions/rubric/index.js +191 -0
- package/lib/assertions/rubric/index.js.map +1 -0
- package/lib/assertions/rubric/internal.d.ts +14 -0
- package/lib/assertions/rubric/internal.d.ts.map +1 -0
- package/lib/assertions/rubric/internal.js +335 -0
- package/lib/assertions/rubric/internal.js.map +1 -0
- package/lib/authoring.d.ts +3 -0
- package/lib/authoring.d.ts.map +1 -0
- package/lib/authoring.js +3 -0
- package/lib/authoring.js.map +1 -0
- package/lib/cli/bin.d.ts +3 -0
- package/lib/cli/bin.d.ts.map +1 -0
- package/lib/cli/bin.js +15 -0
- package/lib/cli/bin.js.map +1 -0
- package/lib/cli/command-runtime.d.ts +5 -0
- package/lib/cli/command-runtime.d.ts.map +1 -0
- package/lib/cli/command-runtime.js +30 -0
- package/lib/cli/command-runtime.js.map +1 -0
- package/lib/cli/init.d.ts +21 -0
- package/lib/cli/init.d.ts.map +1 -0
- package/lib/cli/init.js +340 -0
- package/lib/cli/init.js.map +1 -0
- package/lib/cli/program.d.ts +22 -0
- package/lib/cli/program.d.ts.map +1 -0
- package/lib/cli/program.js +757 -0
- package/lib/cli/program.js.map +1 -0
- package/lib/cli/render/format.d.ts +14 -0
- package/lib/cli/render/format.d.ts.map +1 -0
- package/lib/cli/render/format.js +71 -0
- package/lib/cli/render/format.js.map +1 -0
- package/lib/cli/render/reporter.d.ts +73 -0
- package/lib/cli/render/reporter.d.ts.map +1 -0
- package/lib/cli/render/reporter.js +519 -0
- package/lib/cli/render/reporter.js.map +1 -0
- package/lib/cli/render/tty.d.ts +77 -0
- package/lib/cli/render/tty.d.ts.map +1 -0
- package/lib/cli/render/tty.js +283 -0
- package/lib/cli/render/tty.js.map +1 -0
- package/lib/cli/run-options.d.ts +27 -0
- package/lib/cli/run-options.d.ts.map +1 -0
- package/lib/cli/run-options.js +58 -0
- package/lib/cli/run-options.js.map +1 -0
- package/lib/cli/skills.d.ts +38 -0
- package/lib/cli/skills.d.ts.map +1 -0
- package/lib/cli/skills.js +84 -0
- package/lib/cli/skills.js.map +1 -0
- package/lib/core/annotation.d.ts +56 -0
- package/lib/core/annotation.d.ts.map +1 -0
- package/lib/core/annotation.js +92 -0
- package/lib/core/annotation.js.map +1 -0
- package/lib/core/answerer.d.ts +68 -0
- package/lib/core/answerer.d.ts.map +1 -0
- package/lib/core/answerer.js +17 -0
- package/lib/core/answerer.js.map +1 -0
- package/lib/core/artifact-evidence.d.ts +50 -0
- package/lib/core/artifact-evidence.d.ts.map +1 -0
- package/lib/core/artifact-evidence.js +185 -0
- package/lib/core/artifact-evidence.js.map +1 -0
- package/lib/core/artifact-snapshot.d.ts +37 -0
- package/lib/core/artifact-snapshot.d.ts.map +1 -0
- package/lib/core/artifact-snapshot.js +339 -0
- package/lib/core/artifact-snapshot.js.map +1 -0
- package/lib/core/assertion.d.ts +129 -0
- package/lib/core/assertion.d.ts.map +1 -0
- package/lib/core/assertion.js +75 -0
- package/lib/core/assertion.js.map +1 -0
- package/lib/core/case.d.ts +79 -0
- package/lib/core/case.d.ts.map +1 -0
- package/lib/core/case.js +90 -0
- package/lib/core/case.js.map +1 -0
- package/lib/core/conversation-turn.d.ts +23 -0
- package/lib/core/conversation-turn.d.ts.map +1 -0
- package/lib/core/conversation-turn.js +31 -0
- package/lib/core/conversation-turn.js.map +1 -0
- package/lib/core/driver.d.ts +64 -0
- package/lib/core/driver.d.ts.map +1 -0
- package/lib/core/driver.js +18 -0
- package/lib/core/driver.js.map +1 -0
- package/lib/core/emitter.d.ts +20 -0
- package/lib/core/emitter.d.ts.map +1 -0
- package/lib/core/emitter.js +34 -0
- package/lib/core/emitter.js.map +1 -0
- package/lib/core/events.d.ts +202 -0
- package/lib/core/events.d.ts.map +1 -0
- package/lib/core/events.js +107 -0
- package/lib/core/events.js.map +1 -0
- package/lib/core/exec.d.ts +82 -0
- package/lib/core/exec.d.ts.map +1 -0
- package/lib/core/exec.js +168 -0
- package/lib/core/exec.js.map +1 -0
- package/lib/core/experiment.d.ts +285 -0
- package/lib/core/experiment.d.ts.map +1 -0
- package/lib/core/experiment.js +389 -0
- package/lib/core/experiment.js.map +1 -0
- package/lib/core/glob.d.ts +3 -0
- package/lib/core/glob.d.ts.map +1 -0
- package/lib/core/glob.js +32 -0
- package/lib/core/glob.js.map +1 -0
- package/lib/core/hash.d.ts +8 -0
- package/lib/core/hash.d.ts.map +1 -0
- package/lib/core/hash.js +214 -0
- package/lib/core/hash.js.map +1 -0
- package/lib/core/index.d.ts +29 -0
- package/lib/core/index.d.ts.map +1 -0
- package/lib/core/index.js +29 -0
- package/lib/core/index.js.map +1 -0
- package/lib/core/lifecycle-fixtures.d.ts +20 -0
- package/lib/core/lifecycle-fixtures.d.ts.map +1 -0
- package/lib/core/lifecycle-fixtures.js +109 -0
- package/lib/core/lifecycle-fixtures.js.map +1 -0
- package/lib/core/lifecycle-hooks.d.ts +52 -0
- package/lib/core/lifecycle-hooks.d.ts.map +1 -0
- package/lib/core/lifecycle-hooks.js +42 -0
- package/lib/core/lifecycle-hooks.js.map +1 -0
- package/lib/core/mcp-events.d.ts +44 -0
- package/lib/core/mcp-events.d.ts.map +1 -0
- package/lib/core/mcp-events.js +23 -0
- package/lib/core/mcp-events.js.map +1 -0
- package/lib/core/model-cli.d.ts +57 -0
- package/lib/core/model-cli.d.ts.map +1 -0
- package/lib/core/model-cli.js +237 -0
- package/lib/core/model-cli.js.map +1 -0
- package/lib/core/observer.d.ts +31 -0
- package/lib/core/observer.d.ts.map +1 -0
- package/lib/core/observer.js +25 -0
- package/lib/core/observer.js.map +1 -0
- package/lib/core/platform-process.d.ts +33 -0
- package/lib/core/platform-process.d.ts.map +1 -0
- package/lib/core/platform-process.js +214 -0
- package/lib/core/platform-process.js.map +1 -0
- package/lib/core/registry.d.ts +22 -0
- package/lib/core/registry.d.ts.map +1 -0
- package/lib/core/registry.js +45 -0
- package/lib/core/registry.js.map +1 -0
- package/lib/core/rubric.d.ts +18 -0
- package/lib/core/rubric.d.ts.map +1 -0
- package/lib/core/rubric.js +49 -0
- package/lib/core/rubric.js.map +1 -0
- package/lib/core/run.d.ts +419 -0
- package/lib/core/run.d.ts.map +1 -0
- package/lib/core/run.js +126 -0
- package/lib/core/run.js.map +1 -0
- package/lib/core/statistics.d.ts +39 -0
- package/lib/core/statistics.d.ts.map +1 -0
- package/lib/core/statistics.js +163 -0
- package/lib/core/statistics.js.map +1 -0
- package/lib/core/subject.d.ts +4 -0
- package/lib/core/subject.d.ts.map +1 -0
- package/lib/core/subject.js +46 -0
- package/lib/core/subject.js.map +1 -0
- package/lib/core/tool-events.d.ts +69 -0
- package/lib/core/tool-events.d.ts.map +1 -0
- package/lib/core/tool-events.js +242 -0
- package/lib/core/tool-events.js.map +1 -0
- package/lib/core/usage.d.ts +78 -0
- package/lib/core/usage.d.ts.map +1 -0
- package/lib/core/usage.js +104 -0
- package/lib/core/usage.js.map +1 -0
- package/lib/core/wire.d.ts +83 -0
- package/lib/core/wire.d.ts.map +1 -0
- package/lib/core/wire.js +101 -0
- package/lib/core/wire.js.map +1 -0
- package/lib/core/zod.d.ts +10 -0
- package/lib/core/zod.d.ts.map +1 -0
- package/lib/core/zod.js +18 -0
- package/lib/core/zod.js.map +1 -0
- package/lib/drivers/claude-code/index.d.ts +55 -0
- package/lib/drivers/claude-code/index.d.ts.map +1 -0
- package/lib/drivers/claude-code/index.js +779 -0
- package/lib/drivers/claude-code/index.js.map +1 -0
- package/lib/drivers/claude-code/parse-events.d.ts +46 -0
- package/lib/drivers/claude-code/parse-events.d.ts.map +1 -0
- package/lib/drivers/claude-code/parse-events.js +387 -0
- package/lib/drivers/claude-code/parse-events.js.map +1 -0
- package/lib/drivers/codex/app-events.d.ts +16 -0
- package/lib/drivers/codex/app-events.d.ts.map +1 -0
- package/lib/drivers/codex/app-events.js +184 -0
- package/lib/drivers/codex/app-events.js.map +1 -0
- package/lib/drivers/codex/config.d.ts +55 -0
- package/lib/drivers/codex/config.d.ts.map +1 -0
- package/lib/drivers/codex/config.js +125 -0
- package/lib/drivers/codex/config.js.map +1 -0
- package/lib/drivers/codex/index.d.ts +8 -0
- package/lib/drivers/codex/index.d.ts.map +1 -0
- package/lib/drivers/codex/index.js +904 -0
- package/lib/drivers/codex/index.js.map +1 -0
- package/lib/drivers/mcp/index.d.ts +52 -0
- package/lib/drivers/mcp/index.d.ts.map +1 -0
- package/lib/drivers/mcp/index.js +282 -0
- package/lib/drivers/mcp/index.js.map +1 -0
- package/lib/drivers/mcp/transport.d.ts +50 -0
- package/lib/drivers/mcp/transport.d.ts.map +1 -0
- package/lib/drivers/mcp/transport.js +190 -0
- package/lib/drivers/mcp/transport.js.map +1 -0
- package/lib/drivers/shared/artifacts.d.ts +4 -0
- package/lib/drivers/shared/artifacts.d.ts.map +1 -0
- package/lib/drivers/shared/artifacts.js +81 -0
- package/lib/drivers/shared/artifacts.js.map +1 -0
- package/lib/drivers/shared/secure-copy.d.ts +2 -0
- package/lib/drivers/shared/secure-copy.d.ts.map +1 -0
- package/lib/drivers/shared/secure-copy.js +35 -0
- package/lib/drivers/shared/secure-copy.js.map +1 -0
- package/lib/drivers/subprocess/index.d.ts +24 -0
- package/lib/drivers/subprocess/index.d.ts.map +1 -0
- package/lib/drivers/subprocess/index.js +451 -0
- package/lib/drivers/subprocess/index.js.map +1 -0
- package/lib/index.d.ts +40 -0
- package/lib/index.d.ts.map +1 -0
- package/lib/index.js +48 -0
- package/lib/index.js.map +1 -0
- package/lib/optimization/reflective.d.ts +27 -0
- package/lib/optimization/reflective.d.ts.map +1 -0
- package/lib/optimization/reflective.js +257 -0
- package/lib/optimization/reflective.js.map +1 -0
- package/lib/orchestrator/annotation-loader.d.ts +65 -0
- package/lib/orchestrator/annotation-loader.d.ts.map +1 -0
- package/lib/orchestrator/annotation-loader.js +73 -0
- package/lib/orchestrator/annotation-loader.js.map +1 -0
- package/lib/orchestrator/annotation-store.d.ts +3 -0
- package/lib/orchestrator/annotation-store.d.ts.map +1 -0
- package/lib/orchestrator/annotation-store.js +100 -0
- package/lib/orchestrator/annotation-store.js.map +1 -0
- package/lib/orchestrator/authored-dependencies.d.ts +3 -0
- package/lib/orchestrator/authored-dependencies.d.ts.map +1 -0
- package/lib/orchestrator/authored-dependencies.js +146 -0
- package/lib/orchestrator/authored-dependencies.js.map +1 -0
- package/lib/orchestrator/campaign-lifecycle.d.ts +105 -0
- package/lib/orchestrator/campaign-lifecycle.d.ts.map +1 -0
- package/lib/orchestrator/campaign-lifecycle.js +141 -0
- package/lib/orchestrator/campaign-lifecycle.js.map +1 -0
- package/lib/orchestrator/candidate-integrity.d.ts +4 -0
- package/lib/orchestrator/candidate-integrity.d.ts.map +1 -0
- package/lib/orchestrator/candidate-integrity.js +14 -0
- package/lib/orchestrator/candidate-integrity.js.map +1 -0
- package/lib/orchestrator/candidates.d.ts +13 -0
- package/lib/orchestrator/candidates.d.ts.map +1 -0
- package/lib/orchestrator/candidates.js +577 -0
- package/lib/orchestrator/candidates.js.map +1 -0
- package/lib/orchestrator/compare.d.ts +80 -0
- package/lib/orchestrator/compare.d.ts.map +1 -0
- package/lib/orchestrator/compare.js +270 -0
- package/lib/orchestrator/compare.js.map +1 -0
- package/lib/orchestrator/comparison-identity.d.ts +10 -0
- package/lib/orchestrator/comparison-identity.d.ts.map +1 -0
- package/lib/orchestrator/comparison-identity.js +39 -0
- package/lib/orchestrator/comparison-identity.js.map +1 -0
- package/lib/orchestrator/config.d.ts +121 -0
- package/lib/orchestrator/config.d.ts.map +1 -0
- package/lib/orchestrator/config.js +176 -0
- package/lib/orchestrator/config.js.map +1 -0
- package/lib/orchestrator/content-bound-json.d.ts +5 -0
- package/lib/orchestrator/content-bound-json.d.ts.map +1 -0
- package/lib/orchestrator/content-bound-json.js +59 -0
- package/lib/orchestrator/content-bound-json.js.map +1 -0
- package/lib/orchestrator/discover.d.ts +45 -0
- package/lib/orchestrator/discover.d.ts.map +1 -0
- package/lib/orchestrator/discover.js +98 -0
- package/lib/orchestrator/discover.js.map +1 -0
- package/lib/orchestrator/doctor.d.ts +141 -0
- package/lib/orchestrator/doctor.d.ts.map +1 -0
- package/lib/orchestrator/doctor.js +525 -0
- package/lib/orchestrator/doctor.js.map +1 -0
- package/lib/orchestrator/driver-session.d.ts +177 -0
- package/lib/orchestrator/driver-session.d.ts.map +1 -0
- package/lib/orchestrator/driver-session.js +396 -0
- package/lib/orchestrator/driver-session.js.map +1 -0
- package/lib/orchestrator/evaluator-identity.d.ts +7 -0
- package/lib/orchestrator/evaluator-identity.d.ts.map +1 -0
- package/lib/orchestrator/evaluator-identity.js +42 -0
- package/lib/orchestrator/evaluator-identity.js.map +1 -0
- package/lib/orchestrator/experiment-budget.d.ts +11 -0
- package/lib/orchestrator/experiment-budget.d.ts.map +1 -0
- package/lib/orchestrator/experiment-budget.js +51 -0
- package/lib/orchestrator/experiment-budget.js.map +1 -0
- package/lib/orchestrator/experiment-context.d.ts +43 -0
- package/lib/orchestrator/experiment-context.d.ts.map +1 -0
- package/lib/orchestrator/experiment-context.js +9 -0
- package/lib/orchestrator/experiment-context.js.map +1 -0
- package/lib/orchestrator/experiment-contracts.d.ts +290 -0
- package/lib/orchestrator/experiment-contracts.d.ts.map +1 -0
- package/lib/orchestrator/experiment-contracts.js +38 -0
- package/lib/orchestrator/experiment-contracts.js.map +1 -0
- package/lib/orchestrator/experiment-corpus.d.ts +21 -0
- package/lib/orchestrator/experiment-corpus.d.ts.map +1 -0
- package/lib/orchestrator/experiment-corpus.js +276 -0
- package/lib/orchestrator/experiment-corpus.js.map +1 -0
- package/lib/orchestrator/experiment-evaluation.d.ts +44 -0
- package/lib/orchestrator/experiment-evaluation.d.ts.map +1 -0
- package/lib/orchestrator/experiment-evaluation.js +542 -0
- package/lib/orchestrator/experiment-evaluation.js.map +1 -0
- package/lib/orchestrator/experiment-finish.d.ts +6 -0
- package/lib/orchestrator/experiment-finish.d.ts.map +1 -0
- package/lib/orchestrator/experiment-finish.js +19 -0
- package/lib/orchestrator/experiment-finish.js.map +1 -0
- package/lib/orchestrator/experiment-identity.d.ts +6 -0
- package/lib/orchestrator/experiment-identity.d.ts.map +1 -0
- package/lib/orchestrator/experiment-identity.js +257 -0
- package/lib/orchestrator/experiment-identity.js.map +1 -0
- package/lib/orchestrator/experiment-loader.d.ts +175 -0
- package/lib/orchestrator/experiment-loader.d.ts.map +1 -0
- package/lib/orchestrator/experiment-loader.js +77 -0
- package/lib/orchestrator/experiment-loader.js.map +1 -0
- package/lib/orchestrator/experiment-report.d.ts +450 -0
- package/lib/orchestrator/experiment-report.d.ts.map +1 -0
- package/lib/orchestrator/experiment-report.js +570 -0
- package/lib/orchestrator/experiment-report.js.map +1 -0
- package/lib/orchestrator/experiment-runs.d.ts +379 -0
- package/lib/orchestrator/experiment-runs.d.ts.map +1 -0
- package/lib/orchestrator/experiment-runs.js +796 -0
- package/lib/orchestrator/experiment-runs.js.map +1 -0
- package/lib/orchestrator/experiment-runtime-identity.d.ts +41 -0
- package/lib/orchestrator/experiment-runtime-identity.d.ts.map +1 -0
- package/lib/orchestrator/experiment-runtime-identity.js +411 -0
- package/lib/orchestrator/experiment-runtime-identity.js.map +1 -0
- package/lib/orchestrator/experiment-selection.d.ts +25 -0
- package/lib/orchestrator/experiment-selection.d.ts.map +1 -0
- package/lib/orchestrator/experiment-selection.js +286 -0
- package/lib/orchestrator/experiment-selection.js.map +1 -0
- package/lib/orchestrator/experiment.d.ts +11 -0
- package/lib/orchestrator/experiment.d.ts.map +1 -0
- package/lib/orchestrator/experiment.js +639 -0
- package/lib/orchestrator/experiment.js.map +1 -0
- package/lib/orchestrator/fixture-identity.d.ts +7 -0
- package/lib/orchestrator/fixture-identity.d.ts.map +1 -0
- package/lib/orchestrator/fixture-identity.js +55 -0
- package/lib/orchestrator/fixture-identity.js.map +1 -0
- package/lib/orchestrator/fs-read.d.ts +19 -0
- package/lib/orchestrator/fs-read.d.ts.map +1 -0
- package/lib/orchestrator/fs-read.js +113 -0
- package/lib/orchestrator/fs-read.js.map +1 -0
- package/lib/orchestrator/grader-experiment.d.ts +190 -0
- package/lib/orchestrator/grader-experiment.d.ts.map +1 -0
- package/lib/orchestrator/grader-experiment.js +579 -0
- package/lib/orchestrator/grader-experiment.js.map +1 -0
- package/lib/orchestrator/grading.d.ts +31 -0
- package/lib/orchestrator/grading.d.ts.map +1 -0
- package/lib/orchestrator/grading.js +186 -0
- package/lib/orchestrator/grading.js.map +1 -0
- package/lib/orchestrator/index.d.ts +22 -0
- package/lib/orchestrator/index.d.ts.map +1 -0
- package/lib/orchestrator/index.js +30 -0
- package/lib/orchestrator/index.js.map +1 -0
- package/lib/orchestrator/lifecycle-hook-execution.d.ts +4 -0
- package/lib/orchestrator/lifecycle-hook-execution.d.ts.map +1 -0
- package/lib/orchestrator/lifecycle-hook-execution.js +67 -0
- package/lib/orchestrator/lifecycle-hook-execution.js.map +1 -0
- package/lib/orchestrator/list-runs.d.ts +33 -0
- package/lib/orchestrator/list-runs.d.ts.map +1 -0
- package/lib/orchestrator/list-runs.js +308 -0
- package/lib/orchestrator/list-runs.js.map +1 -0
- package/lib/orchestrator/loader.d.ts +3 -0
- package/lib/orchestrator/loader.d.ts.map +1 -0
- package/lib/orchestrator/loader.js +17 -0
- package/lib/orchestrator/loader.js.map +1 -0
- package/lib/orchestrator/optimization-lifecycle.d.ts +12 -0
- package/lib/orchestrator/optimization-lifecycle.d.ts.map +1 -0
- package/lib/orchestrator/optimization-lifecycle.js +92 -0
- package/lib/orchestrator/optimization-lifecycle.js.map +1 -0
- package/lib/orchestrator/orchestrator.d.ts +144 -0
- package/lib/orchestrator/orchestrator.d.ts.map +1 -0
- package/lib/orchestrator/orchestrator.js +632 -0
- package/lib/orchestrator/orchestrator.js.map +1 -0
- package/lib/orchestrator/owned-lock.d.ts +17 -0
- package/lib/orchestrator/owned-lock.d.ts.map +1 -0
- package/lib/orchestrator/owned-lock.js +149 -0
- package/lib/orchestrator/owned-lock.js.map +1 -0
- package/lib/orchestrator/package-version.d.ts +2 -0
- package/lib/orchestrator/package-version.d.ts.map +1 -0
- package/lib/orchestrator/package-version.js +4 -0
- package/lib/orchestrator/package-version.js.map +1 -0
- package/lib/orchestrator/provenance.d.ts +5 -0
- package/lib/orchestrator/provenance.d.ts.map +1 -0
- package/lib/orchestrator/provenance.js +137 -0
- package/lib/orchestrator/provenance.js.map +1 -0
- package/lib/orchestrator/regrade.d.ts +80 -0
- package/lib/orchestrator/regrade.d.ts.map +1 -0
- package/lib/orchestrator/regrade.js +433 -0
- package/lib/orchestrator/regrade.js.map +1 -0
- package/lib/orchestrator/rubric-loader.d.ts +7 -0
- package/lib/orchestrator/rubric-loader.d.ts.map +1 -0
- package/lib/orchestrator/rubric-loader.js +65 -0
- package/lib/orchestrator/rubric-loader.js.map +1 -0
- package/lib/orchestrator/run-dir.d.ts +40 -0
- package/lib/orchestrator/run-dir.d.ts.map +1 -0
- package/lib/orchestrator/run-dir.js +161 -0
- package/lib/orchestrator/run-dir.js.map +1 -0
- package/lib/orchestrator/run-execution.d.ts +30 -0
- package/lib/orchestrator/run-execution.d.ts.map +1 -0
- package/lib/orchestrator/run-execution.js +531 -0
- package/lib/orchestrator/run-execution.js.map +1 -0
- package/lib/orchestrator/run-lifecycle.d.ts +115 -0
- package/lib/orchestrator/run-lifecycle.d.ts.map +1 -0
- package/lib/orchestrator/run-lifecycle.js +136 -0
- package/lib/orchestrator/run-lifecycle.js.map +1 -0
- package/lib/orchestrator/suite-compare.d.ts +76 -0
- package/lib/orchestrator/suite-compare.d.ts.map +1 -0
- package/lib/orchestrator/suite-compare.js +159 -0
- package/lib/orchestrator/suite-compare.js.map +1 -0
- package/lib/orchestrator/suite-lifecycle.d.ts +157 -0
- package/lib/orchestrator/suite-lifecycle.d.ts.map +1 -0
- package/lib/orchestrator/suite-lifecycle.js +253 -0
- package/lib/orchestrator/suite-lifecycle.js.map +1 -0
- package/lib/orchestrator/suite-runs.d.ts +91 -0
- package/lib/orchestrator/suite-runs.d.ts.map +1 -0
- package/lib/orchestrator/suite-runs.js +157 -0
- package/lib/orchestrator/suite-runs.js.map +1 -0
- package/lib/orchestrator/suite-summary.d.ts +109 -0
- package/lib/orchestrator/suite-summary.d.ts.map +1 -0
- package/lib/orchestrator/suite-summary.js +213 -0
- package/lib/orchestrator/suite-summary.js.map +1 -0
- package/lib/orchestrator/trace-limits.d.ts +20 -0
- package/lib/orchestrator/trace-limits.d.ts.map +1 -0
- package/lib/orchestrator/trace-limits.js +28 -0
- package/lib/orchestrator/trace-limits.js.map +1 -0
- package/lib/orchestrator/validate-plugin-config.d.ts +10 -0
- package/lib/orchestrator/validate-plugin-config.d.ts.map +1 -0
- package/lib/orchestrator/validate-plugin-config.js +26 -0
- package/lib/orchestrator/validate-plugin-config.js.map +1 -0
- package/package.json +84 -0
- package/skills/lux/SKILL.md +85 -0
- package/skills/lux/references/cli.md +102 -0
- package/skills/lux/references/eval-authoring.md +105 -0
- package/skills/lux-answerer/SKILL.md +169 -0
- package/src/answerers/claude-code/index.ts +342 -0
- package/src/answerers/persona/index.ts +327 -0
- package/src/answerers/scripted/index.ts +186 -0
- package/src/answerers/shared.ts +23 -0
- package/src/answerers/terminal/index.ts +117 -0
- package/src/assertions/core/command-exits-zero.ts +94 -0
- package/src/assertions/core/event-checks.ts +138 -0
- package/src/assertions/core/file-checks.ts +336 -0
- package/src/assertions/core/helpers.ts +203 -0
- package/src/assertions/core/index.ts +80 -0
- package/src/assertions/core/json-pointer-equals.ts +84 -0
- package/src/assertions/core/output-checks.ts +134 -0
- package/src/assertions/core/predicate.ts +67 -0
- package/src/assertions/core/resource-checks.ts +180 -0
- package/src/assertions/core/safe-regex.ts +196 -0
- package/src/assertions/core/tool-called-with.ts +68 -0
- package/src/assertions/core/tool-calls.ts +28 -0
- package/src/assertions/core/tool-invocation.ts +198 -0
- package/src/assertions/core/tool-result-checks.ts +211 -0
- package/src/assertions/mcp/discovery-checks.ts +139 -0
- package/src/assertions/mcp/index.ts +37 -0
- package/src/assertions/rubric/index.ts +258 -0
- package/src/assertions/rubric/internal.ts +367 -0
- package/src/authoring.ts +2 -0
- package/src/cli/bin.ts +14 -0
- package/src/cli/command-runtime.ts +34 -0
- package/src/cli/init.ts +410 -0
- package/src/cli/program.ts +992 -0
- package/src/cli/render/format.ts +69 -0
- package/src/cli/render/reporter.ts +643 -0
- package/src/cli/render/tty.ts +314 -0
- package/src/cli/run-options.ts +82 -0
- package/src/cli/skills.ts +129 -0
- package/src/core/annotation.ts +104 -0
- package/src/core/answerer.ts +75 -0
- package/src/core/artifact-evidence.ts +249 -0
- package/src/core/artifact-snapshot.ts +429 -0
- package/src/core/assertion.ts +156 -0
- package/src/core/case.ts +103 -0
- package/src/core/conversation-turn.ts +47 -0
- package/src/core/driver.ts +75 -0
- package/src/core/emitter.ts +38 -0
- package/src/core/events.ts +135 -0
- package/src/core/exec.ts +263 -0
- package/src/core/experiment.ts +467 -0
- package/src/core/glob.ts +31 -0
- package/src/core/hash.ts +237 -0
- package/src/core/index.ts +28 -0
- package/src/core/lifecycle-fixtures.ts +145 -0
- package/src/core/lifecycle-hooks.ts +101 -0
- package/src/core/mcp-events.ts +57 -0
- package/src/core/model-cli.ts +271 -0
- package/src/core/observer.ts +50 -0
- package/src/core/platform-process.ts +284 -0
- package/src/core/registry.ts +57 -0
- package/src/core/rubric.ts +52 -0
- package/src/core/run.ts +144 -0
- package/src/core/statistics.ts +194 -0
- package/src/core/subject.ts +50 -0
- package/src/core/tool-events.ts +292 -0
- package/src/core/usage.ts +123 -0
- package/src/core/wire.ts +135 -0
- package/src/core/zod.ts +20 -0
- package/src/drivers/claude-code/index.ts +866 -0
- package/src/drivers/claude-code/parse-events.ts +464 -0
- package/src/drivers/codex/app-events.ts +218 -0
- package/src/drivers/codex/config.ts +128 -0
- package/src/drivers/codex/index.ts +995 -0
- package/src/drivers/mcp/index.ts +351 -0
- package/src/drivers/mcp/transport.ts +222 -0
- package/src/drivers/shared/artifacts.ts +86 -0
- package/src/drivers/shared/secure-copy.ts +40 -0
- package/src/drivers/subprocess/index.ts +490 -0
- package/src/index.ts +594 -0
- package/src/optimization/reflective.ts +319 -0
- package/src/orchestrator/annotation-loader.ts +97 -0
- package/src/orchestrator/annotation-store.ts +112 -0
- package/src/orchestrator/authored-dependencies.ts +193 -0
- package/src/orchestrator/campaign-lifecycle.ts +181 -0
- package/src/orchestrator/candidate-integrity.ts +18 -0
- package/src/orchestrator/candidates.ts +683 -0
- package/src/orchestrator/compare.ts +358 -0
- package/src/orchestrator/comparison-identity.ts +45 -0
- package/src/orchestrator/config.ts +205 -0
- package/src/orchestrator/content-bound-json.ts +71 -0
- package/src/orchestrator/discover.ts +132 -0
- package/src/orchestrator/doctor.ts +781 -0
- package/src/orchestrator/driver-session.ts +507 -0
- package/src/orchestrator/evaluator-identity.ts +45 -0
- package/src/orchestrator/experiment-budget.ts +61 -0
- package/src/orchestrator/experiment-context.ts +41 -0
- package/src/orchestrator/experiment-contracts.ts +43 -0
- package/src/orchestrator/experiment-corpus.ts +416 -0
- package/src/orchestrator/experiment-evaluation.ts +749 -0
- package/src/orchestrator/experiment-finish.ts +33 -0
- package/src/orchestrator/experiment-identity.ts +356 -0
- package/src/orchestrator/experiment-loader.ts +105 -0
- package/src/orchestrator/experiment-report.ts +788 -0
- package/src/orchestrator/experiment-runs.ts +960 -0
- package/src/orchestrator/experiment-runtime-identity.ts +465 -0
- package/src/orchestrator/experiment-selection.ts +467 -0
- package/src/orchestrator/experiment.ts +991 -0
- package/src/orchestrator/fixture-identity.ts +67 -0
- package/src/orchestrator/fs-read.ts +126 -0
- package/src/orchestrator/grader-experiment.ts +856 -0
- package/src/orchestrator/grading.ts +260 -0
- package/src/orchestrator/index.ts +186 -0
- package/src/orchestrator/lifecycle-hook-execution.ts +90 -0
- package/src/orchestrator/list-runs.ts +384 -0
- package/src/orchestrator/loader.ts +19 -0
- package/src/orchestrator/optimization-lifecycle.ts +126 -0
- package/src/orchestrator/orchestrator.ts +919 -0
- package/src/orchestrator/owned-lock.ts +147 -0
- package/src/orchestrator/package-version.ts +5 -0
- package/src/orchestrator/provenance.ts +140 -0
- package/src/orchestrator/regrade.ts +580 -0
- package/src/orchestrator/rubric-loader.ts +74 -0
- package/src/orchestrator/run-dir.ts +209 -0
- package/src/orchestrator/run-execution.ts +694 -0
- package/src/orchestrator/run-lifecycle.ts +174 -0
- package/src/orchestrator/suite-compare.ts +195 -0
- package/src/orchestrator/suite-lifecycle.ts +335 -0
- package/src/orchestrator/suite-runs.ts +202 -0
- package/src/orchestrator/suite-summary.ts +276 -0
- package/src/orchestrator/trace-limits.ts +65 -0
- package/src/orchestrator/validate-plugin-config.ts +34 -0
|
@@ -0,0 +1,308 @@
|
|
|
1
|
+
import { existsSync } from "node:fs";
|
|
2
|
+
import { constants, lstat, open, readdir } from "node:fs/promises";
|
|
3
|
+
import { basename, join, relative, sep } from "node:path";
|
|
4
|
+
import { GradingReportSchema, harnessUsage, PersistedCaseSchema, RunEventSchema, RunMetadataSchema, ToolCallPayloadSchema, ToolResultPayloadSchema, totalTokens, } from "../core/index.js";
|
|
5
|
+
import { readBoundedHandle, safeReadJson, walkFiles } from "./fs-read.js";
|
|
6
|
+
import { MAX_RUN_TRACE_BYTES, MAX_RUN_TRACE_EVENTS } from "./trace-limits.js";
|
|
7
|
+
// The trailing `-N` is the collision suffix openRunDir adds when two runs of
|
|
8
|
+
// the same case+driver start in the same millisecond.
|
|
9
|
+
const TS_DIR_RE = /^\d{4}-\d{2}-\d{2}T[\d-]+Z(?:-\d+)?$/;
|
|
10
|
+
const summarizeRun = async (caseSlug, driverName, ts, runDir) => {
|
|
11
|
+
const [meta, grading] = await Promise.all([
|
|
12
|
+
safeReadJson(join(runDir, "run.json"), RunMetadataSchema),
|
|
13
|
+
safeReadJson(join(runDir, "grading.json"), GradingReportSchema),
|
|
14
|
+
]);
|
|
15
|
+
const startedAt = meta?.startedAt ?? null;
|
|
16
|
+
const endedAt = meta?.endedAt ?? null;
|
|
17
|
+
const started = Date.parse(startedAt ?? "");
|
|
18
|
+
const ended = Date.parse(endedAt ?? "");
|
|
19
|
+
const durationMs = Number.isNaN(started) || Number.isNaN(ended) ? null : ended - started;
|
|
20
|
+
const passing = grading?.assertions.filter((a) => a.passed).length ?? 0;
|
|
21
|
+
return {
|
|
22
|
+
caseSlug,
|
|
23
|
+
driverName,
|
|
24
|
+
ts,
|
|
25
|
+
runDir,
|
|
26
|
+
exitReason: meta?.exitReason ?? null,
|
|
27
|
+
driverAgentId: meta?.driver.agentId ?? null,
|
|
28
|
+
startedAt,
|
|
29
|
+
endedAt,
|
|
30
|
+
durationMs,
|
|
31
|
+
casePassed: grading?.casePassed ?? null,
|
|
32
|
+
caseScore: grading?.caseScore ?? null,
|
|
33
|
+
assertionPassed: passing,
|
|
34
|
+
assertionTotal: grading?.assertions.length ?? 0,
|
|
35
|
+
};
|
|
36
|
+
};
|
|
37
|
+
const listForCaseDriver = async (driverPath, caseSlug, driverName) => {
|
|
38
|
+
// Directories only, as in collectRuns: a timestamp-named *file* dropped
|
|
39
|
+
// beside real run dirs must not fabricate a phantom row.
|
|
40
|
+
const entries = (await readdir(driverPath, { withFileTypes: true }))
|
|
41
|
+
.filter((entry) => entry.isDirectory() && TS_DIR_RE.test(entry.name))
|
|
42
|
+
.map((entry) => entry.name);
|
|
43
|
+
const runs = await Promise.all(entries.map((ts) => summarizeRun(caseSlug, driverName, ts, join(driverPath, ts))));
|
|
44
|
+
return runs.sort((a, b) => b.ts.localeCompare(a.ts));
|
|
45
|
+
};
|
|
46
|
+
// Case slugs keep their slashes, so a run nests at a variable depth. Recurse to
|
|
47
|
+
// the directory whose children are timestamps (the driver dir) rather than
|
|
48
|
+
// assuming a fixed two levels, which misses every case in a subdirectory.
|
|
49
|
+
const collectRuns = async (runsRoot, dir, out) => {
|
|
50
|
+
// Only directories count: a stray timestamp-named file must not make this
|
|
51
|
+
// dir read as a driver dir, and a dangling symlink (or a racing delete)
|
|
52
|
+
// must not reject the listing.
|
|
53
|
+
const subdirs = (await readdir(dir, { withFileTypes: true }))
|
|
54
|
+
.filter((entry) => entry.isDirectory())
|
|
55
|
+
.map((entry) => entry.name);
|
|
56
|
+
if (subdirs.some((n) => TS_DIR_RE.test(n))) {
|
|
57
|
+
const caseDir = join(dir, "..");
|
|
58
|
+
const caseSlug = relative(runsRoot, caseDir).split(sep).join("/");
|
|
59
|
+
// A run layout is at least runsRoot/<case>/<driver>/<ts>; timestamp dirs
|
|
60
|
+
// higher up (a stray archived run copied into the tree) have no case
|
|
61
|
+
// level above them, so a slug of "" or ".." marks them as noise, not
|
|
62
|
+
// runs. Either way the non-timestamp siblings may still hold real runs —
|
|
63
|
+
// classify per child and keep descending instead of returning early.
|
|
64
|
+
if (caseSlug !== "" && caseSlug !== ".." && !caseSlug.startsWith("../")) {
|
|
65
|
+
out.push(...(await listForCaseDriver(dir, caseSlug, basename(dir))));
|
|
66
|
+
}
|
|
67
|
+
}
|
|
68
|
+
for (const name of subdirs.filter((n) => !TS_DIR_RE.test(n))) {
|
|
69
|
+
await collectRuns(runsRoot, join(dir, name), out);
|
|
70
|
+
}
|
|
71
|
+
};
|
|
72
|
+
export const listRuns = async (runsRoot) => {
|
|
73
|
+
if (!existsSync(runsRoot))
|
|
74
|
+
return [];
|
|
75
|
+
const out = [];
|
|
76
|
+
await collectRuns(runsRoot, runsRoot, out);
|
|
77
|
+
return out.sort((a, b) => {
|
|
78
|
+
if (a.caseSlug !== b.caseSlug)
|
|
79
|
+
return a.caseSlug.localeCompare(b.caseSlug);
|
|
80
|
+
if (a.driverName !== b.driverName)
|
|
81
|
+
return a.driverName.localeCompare(b.driverName);
|
|
82
|
+
return b.ts.localeCompare(a.ts);
|
|
83
|
+
});
|
|
84
|
+
};
|
|
85
|
+
export const formatRunsTable = (runs) => {
|
|
86
|
+
if (runs.length === 0)
|
|
87
|
+
return "(no runs found)";
|
|
88
|
+
const rows = [
|
|
89
|
+
["status", "case", "driver", "ts", "exit", "score", "duration"],
|
|
90
|
+
...runs.map((r) => {
|
|
91
|
+
let status = "ungraded";
|
|
92
|
+
if (r.casePassed === true)
|
|
93
|
+
status = "PASS";
|
|
94
|
+
else if (r.casePassed === false)
|
|
95
|
+
status = "FAIL";
|
|
96
|
+
const score = r.caseScore !== null
|
|
97
|
+
? `${(r.caseScore * 100).toFixed(0)}% (${r.assertionPassed}/${r.assertionTotal})`
|
|
98
|
+
: "—";
|
|
99
|
+
const duration = r.durationMs !== null ? `${(r.durationMs / 1000).toFixed(1)}s` : "—";
|
|
100
|
+
return [status, r.caseSlug, r.driverName, r.ts, r.exitReason ?? "?", score, duration];
|
|
101
|
+
}),
|
|
102
|
+
];
|
|
103
|
+
const header = rows[0] ?? [];
|
|
104
|
+
const widths = header.map((_, col) => Math.max(...rows.map((r) => r[col]?.length ?? 0)));
|
|
105
|
+
return rows
|
|
106
|
+
.map((row) => row.map((cell, col) => cell.padEnd(widths[col] ?? 0)).join(" "))
|
|
107
|
+
.join("\n");
|
|
108
|
+
};
|
|
109
|
+
const readEventsJsonl = async (path) => {
|
|
110
|
+
if (!existsSync(path))
|
|
111
|
+
return [];
|
|
112
|
+
if ((await lstat(path)).isSymbolicLink()) {
|
|
113
|
+
throw new Error(`run events cannot be a symlink: ${path}`);
|
|
114
|
+
}
|
|
115
|
+
let handle;
|
|
116
|
+
try {
|
|
117
|
+
handle = await open(path, constants.O_RDONLY | constants.O_NOFOLLOW);
|
|
118
|
+
}
|
|
119
|
+
catch (error) {
|
|
120
|
+
if (error.code === "ENOENT")
|
|
121
|
+
return [];
|
|
122
|
+
if (error.code === "ELOOP") {
|
|
123
|
+
throw new Error(`run events cannot be a symlink: ${path}`);
|
|
124
|
+
}
|
|
125
|
+
throw error;
|
|
126
|
+
}
|
|
127
|
+
let text;
|
|
128
|
+
try {
|
|
129
|
+
const metadata = await handle.stat();
|
|
130
|
+
if (!metadata.isFile() || metadata.size > MAX_RUN_TRACE_BYTES) {
|
|
131
|
+
throw new Error(`run events must be a regular file no larger than ${MAX_RUN_TRACE_BYTES} bytes: ${path}`);
|
|
132
|
+
}
|
|
133
|
+
text = (await readBoundedHandle(handle, MAX_RUN_TRACE_BYTES, "run events")).toString("utf8");
|
|
134
|
+
}
|
|
135
|
+
finally {
|
|
136
|
+
await handle.close();
|
|
137
|
+
}
|
|
138
|
+
const lines = text.split("\n");
|
|
139
|
+
if (lines.at(-1) === "")
|
|
140
|
+
lines.pop();
|
|
141
|
+
if (lines.length > MAX_RUN_TRACE_EVENTS) {
|
|
142
|
+
throw new Error(`run events exceed ${MAX_RUN_TRACE_EVENTS} lines: ${path}`);
|
|
143
|
+
}
|
|
144
|
+
return lines.map((line, index) => {
|
|
145
|
+
let raw;
|
|
146
|
+
try {
|
|
147
|
+
raw = JSON.parse(line);
|
|
148
|
+
}
|
|
149
|
+
catch {
|
|
150
|
+
throw new Error(`corrupt run events at ${path}:${index + 1}: invalid JSON`);
|
|
151
|
+
}
|
|
152
|
+
const parsed = RunEventSchema.safeParse(raw);
|
|
153
|
+
if (!parsed.success) {
|
|
154
|
+
throw new Error(`corrupt run events at ${path}:${index + 1}: invalid event`);
|
|
155
|
+
}
|
|
156
|
+
return parsed.data;
|
|
157
|
+
});
|
|
158
|
+
};
|
|
159
|
+
export const loadRunDetail = async (runDir, options = {}) => {
|
|
160
|
+
const [caseJson, metadata, grading, events, artifacts] = await Promise.all([
|
|
161
|
+
options.skipJson
|
|
162
|
+
? Promise.resolve(null)
|
|
163
|
+
: safeReadJson(join(runDir, "case.json"), PersistedCaseSchema),
|
|
164
|
+
options.skipJson
|
|
165
|
+
? Promise.resolve(null)
|
|
166
|
+
: safeReadJson(join(runDir, "run.json"), RunMetadataSchema),
|
|
167
|
+
options.skipJson
|
|
168
|
+
? Promise.resolve(null)
|
|
169
|
+
: safeReadJson(join(runDir, "grading.json"), GradingReportSchema),
|
|
170
|
+
readEventsJsonl(join(runDir, "events.jsonl")),
|
|
171
|
+
options.skipArtifactsWalk
|
|
172
|
+
? Promise.resolve([])
|
|
173
|
+
: walkFiles(join(runDir, "artifacts"), options.maxArtifacts),
|
|
174
|
+
]);
|
|
175
|
+
return { runDir, case: caseJson, metadata, grading, events, artifacts };
|
|
176
|
+
};
|
|
177
|
+
const appendCaseDetail = (lines, detail) => {
|
|
178
|
+
if (detail.case) {
|
|
179
|
+
lines.push(`id: ${detail.case.id ?? "(no id)"}`);
|
|
180
|
+
if (detail.case.prompt) {
|
|
181
|
+
lines.push(`prompt: ${detail.case.prompt.slice(0, 120).replace(/\n/g, " ")}`);
|
|
182
|
+
}
|
|
183
|
+
if (detail.case.persona) {
|
|
184
|
+
lines.push(`persona: ${detail.case.persona.slice(0, 120).replace(/\n/g, " ")}`);
|
|
185
|
+
}
|
|
186
|
+
}
|
|
187
|
+
};
|
|
188
|
+
const appendMetadataDetail = (lines, detail) => {
|
|
189
|
+
if (detail.metadata) {
|
|
190
|
+
lines.push("");
|
|
191
|
+
lines.push(`exit: ${detail.metadata.exitReason}`);
|
|
192
|
+
lines.push(`driver: ${detail.metadata.driver.name} (agent=${detail.metadata.driver.agentId}${detail.metadata.driver.model ? `, model=${detail.metadata.driver.model}` : ""})`);
|
|
193
|
+
const answerer = detail.metadata.answerer;
|
|
194
|
+
lines.push(`answerer: ${answerer.name}${answerer.provider ? ` (${answerer.provider})` : ""}${answerer.model ? `, model=${answerer.model}` : ""}`);
|
|
195
|
+
const provenance = detail.metadata.provenance;
|
|
196
|
+
if (provenance?.source) {
|
|
197
|
+
const git = provenance.source.git;
|
|
198
|
+
lines.push(`source: ${provenance.source.casePath}${git ? ` @ ${git.commit.slice(0, 12)}${git.dirty ? " (dirty)" : ""}` : ""}`);
|
|
199
|
+
}
|
|
200
|
+
if (provenance) {
|
|
201
|
+
const runtime = provenance.runtime;
|
|
202
|
+
lines.push(`runtime: ${runtime.platform}/${runtime.arch} ${runtime.node}${runtime.bun ? ` bun/${runtime.bun}` : ""}`);
|
|
203
|
+
}
|
|
204
|
+
const c = detail.metadata.interruptCounts;
|
|
205
|
+
lines.push(`interrupts: ask=${c.ask} approve=${c.approve} custom=${c.custom}`);
|
|
206
|
+
const tools = detail.metadata.toolOutcomes;
|
|
207
|
+
if (tools) {
|
|
208
|
+
lines.push(`tools: reported-calls=${tools.reportedCalls} valid-calls=${tools.total} succeeded=${tools.succeeded} failed=${tools.failed} unresolved=${tools.unresolved} unmatched-results=${tools.unmatchedResults} malformed=${tools.malformedEvents}`);
|
|
209
|
+
}
|
|
210
|
+
else if (detail.metadata.toolCallCount !== undefined) {
|
|
211
|
+
lines.push(`tools: reported-calls=${detail.metadata.toolCallCount}`);
|
|
212
|
+
}
|
|
213
|
+
const usage = detail.metadata.usage;
|
|
214
|
+
lines.push(`tokens: agent=${totalTokens(usage.agent)} harness=${totalTokens(harnessUsage(usage))}` +
|
|
215
|
+
` (answerer=${totalTokens(usage.answerer)} judge=${totalTokens(usage.grading)})`);
|
|
216
|
+
const phases = detail.metadata.phaseDurationMs;
|
|
217
|
+
if (phases) {
|
|
218
|
+
const secs = (ms) => `${(ms / 1000).toFixed(1)}s`;
|
|
219
|
+
lines.push(`phases: agent=${secs(phases.agent)} collect=${secs(phases.collect)} grading=${secs(phases.grading)}`);
|
|
220
|
+
}
|
|
221
|
+
}
|
|
222
|
+
};
|
|
223
|
+
const appendGradingDetail = (lines, detail) => {
|
|
224
|
+
if (detail.grading) {
|
|
225
|
+
lines.push("");
|
|
226
|
+
let status = "ungraded";
|
|
227
|
+
if (detail.grading.casePassed === true)
|
|
228
|
+
status = "PASS";
|
|
229
|
+
else if (detail.grading.casePassed === false)
|
|
230
|
+
status = "FAIL";
|
|
231
|
+
const score = detail.grading.caseScore !== null ? `${(detail.grading.caseScore * 100).toFixed(0)}%` : "—";
|
|
232
|
+
lines.push(`grading: ${status} (${score}, ${detail.grading.assertions.filter((a) => a.passed).length}/${detail.grading.assertions.length})`);
|
|
233
|
+
for (const a of detail.grading.assertions) {
|
|
234
|
+
const mark = a.passed ? "✓" : "✗";
|
|
235
|
+
const took = ` (${(a.durationMs / 1000).toFixed(1)}s)`;
|
|
236
|
+
lines.push(` ${mark} ${a.name}${took}${a.comment ? ` — ${a.comment}` : ""}`);
|
|
237
|
+
}
|
|
238
|
+
}
|
|
239
|
+
};
|
|
240
|
+
const appendEventDetail = (lines, detail) => {
|
|
241
|
+
if (detail.events.length > 0) {
|
|
242
|
+
lines.push("");
|
|
243
|
+
lines.push(`events: ${detail.events.length} total`);
|
|
244
|
+
const tail = detail.events.slice(-10);
|
|
245
|
+
for (const e of tail) {
|
|
246
|
+
lines.push(` ${formatEventLine(e)}`);
|
|
247
|
+
}
|
|
248
|
+
}
|
|
249
|
+
};
|
|
250
|
+
const appendArtifactDetail = (lines, detail) => {
|
|
251
|
+
if (detail.artifacts.length > 0) {
|
|
252
|
+
lines.push("");
|
|
253
|
+
lines.push(`artifacts: ${detail.artifacts.length} files`);
|
|
254
|
+
for (const a of detail.artifacts.slice(0, 20)) {
|
|
255
|
+
lines.push(` ${a}`);
|
|
256
|
+
}
|
|
257
|
+
if (detail.artifacts.length > 20) {
|
|
258
|
+
lines.push(` ... and ${detail.artifacts.length - 20} more`);
|
|
259
|
+
}
|
|
260
|
+
}
|
|
261
|
+
};
|
|
262
|
+
export const formatRunDetail = (detail) => {
|
|
263
|
+
const lines = [`runDir: ${detail.runDir}`];
|
|
264
|
+
appendCaseDetail(lines, detail);
|
|
265
|
+
appendMetadataDetail(lines, detail);
|
|
266
|
+
appendGradingDetail(lines, detail);
|
|
267
|
+
appendEventDetail(lines, detail);
|
|
268
|
+
appendArtifactDetail(lines, detail);
|
|
269
|
+
return lines.join("\n");
|
|
270
|
+
};
|
|
271
|
+
const formatEventLine = (event) => {
|
|
272
|
+
switch (event.type) {
|
|
273
|
+
case "interrupt":
|
|
274
|
+
if (event.kind === "ask")
|
|
275
|
+
return `interrupt(ask): ${event.question.slice(0, 80)}`;
|
|
276
|
+
if (event.kind === "approve")
|
|
277
|
+
return `interrupt(approve): ${event.request.slice(0, 80)}`;
|
|
278
|
+
return `interrupt(custom): ${JSON.stringify(event.payload).slice(0, 80)}`;
|
|
279
|
+
case "responded":
|
|
280
|
+
if (event.kind === "ask")
|
|
281
|
+
return `responded(ask): ${event.text.slice(0, 80)}`;
|
|
282
|
+
if (event.kind === "approve") {
|
|
283
|
+
return `responded(approve): ${event.approved ? "yes" : "no"}${event.reason ? ` — ${event.reason.slice(0, 60)}` : ""}`;
|
|
284
|
+
}
|
|
285
|
+
return `responded(custom): ${JSON.stringify(event.payload).slice(0, 80)}`;
|
|
286
|
+
case "progress":
|
|
287
|
+
if (event.kind === "tool-call") {
|
|
288
|
+
const payload = ToolCallPayloadSchema.safeParse(event.payload).data;
|
|
289
|
+
return payload ? `tool-call: ${payload.name}` : "tool-call: malformed";
|
|
290
|
+
}
|
|
291
|
+
if (event.kind === "tool-result") {
|
|
292
|
+
const payload = ToolResultPayloadSchema.safeParse(event.payload).data;
|
|
293
|
+
if (!payload)
|
|
294
|
+
return "tool-result: malformed";
|
|
295
|
+
return `tool-result: ${payload.name} (${payload.isError ? "failed" : "succeeded"})`;
|
|
296
|
+
}
|
|
297
|
+
return `progress(${event.kind})`;
|
|
298
|
+
case "lifecycle":
|
|
299
|
+
return `lifecycle: ${event.stage}`;
|
|
300
|
+
case "ready":
|
|
301
|
+
return `ready: agentId=${event.agentId}`;
|
|
302
|
+
case "done":
|
|
303
|
+
return "done";
|
|
304
|
+
case "error":
|
|
305
|
+
return `error: ${event.reason}`;
|
|
306
|
+
}
|
|
307
|
+
};
|
|
308
|
+
//# sourceMappingURL=list-runs.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"list-runs.js","sourceRoot":"","sources":["../../src/orchestrator/list-runs.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AACrC,OAAO,EAAE,SAAS,EAAE,KAAK,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,kBAAkB,CAAC;AACnE,OAAO,EAAE,QAAQ,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,EAAE,MAAM,WAAW,CAAC;AAC1D,OAAO,EAEL,mBAAmB,EACnB,YAAY,EAEZ,mBAAmB,EAEnB,cAAc,EAEd,iBAAiB,EACjB,qBAAqB,EACrB,uBAAuB,EACvB,WAAW,GACZ,MAAM,kBAAkB,CAAC;AAC1B,OAAO,EAAE,iBAAiB,EAAE,YAAY,EAAE,SAAS,EAAE,MAAM,cAAc,CAAC;AAC1E,OAAO,EAAE,mBAAmB,EAAE,oBAAoB,EAAE,MAAM,mBAAmB,CAAC;AAE9E,6EAA6E;AAC7E,sDAAsD;AACtD,MAAM,SAAS,GAAG,sCAAsC,CAAC;AAiBzD,MAAM,YAAY,GAAG,KAAK,EACxB,QAAgB,EAChB,UAAkB,EAClB,EAAU,EACV,MAAc,EACM,EAAE;IACtB,MAAM,CAAC,IAAI,EAAE,OAAO,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;QACxC,YAAY,CAAC,IAAI,CAAC,MAAM,EAAE,UAAU,CAAC,EAAE,iBAAiB,CAAC;QACzD,YAAY,CAAC,IAAI,CAAC,MAAM,EAAE,cAAc,CAAC,EAAE,mBAAmB,CAAC;KAChE,CAAC,CAAC;IACH,MAAM,SAAS,GAAG,IAAI,EAAE,SAAS,IAAI,IAAI,CAAC;IAC1C,MAAM,OAAO,GAAG,IAAI,EAAE,OAAO,IAAI,IAAI,CAAC;IACtC,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,IAAI,EAAE,CAAC,CAAC;IAC5C,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC;IACxC,MAAM,UAAU,GAAG,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,KAAK,GAAG,OAAO,CAAC;IACzF,MAAM,OAAO,GAAG,OAAO,EAAE,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC;IACxE,OAAO;QACL,QAAQ;QACR,UAAU;QACV,EAAE;QACF,MAAM;QACN,UAAU,EAAE,IAAI,EAAE,UAAU,IAAI,IAAI;QACpC,aAAa,EAAE,IAAI,EAAE,MAAM,CAAC,OAAO,IAAI,IAAI;QAC3C,SAAS;QACT,OAAO;QACP,UAAU;QACV,UAAU,EAAE,OAAO,EAAE,UAAU,IAAI,IAAI;QACvC,SAAS,EAAE,OAAO,EAAE,SAAS,IAAI,IAAI;QACrC,eAAe,EAAE,OAAO;QACxB,cAAc,EAAE,OAAO,EAAE,UAAU,CAAC,MAAM,IAAI,CAAC;KAChD,CAAC;AACJ,CAAC,CAAC;AAEF,MAAM,iBAAiB,GAAG,KAAK,EAC7B,UAAkB,EAClB,QAAgB,EAChB,UAAkB,EACI,EAAE;IACxB,wEAAwE;IACxE,yDAAyD;IACzD,MAAM,OAAO,GAAG,CAAC,MAAM,OAAO,CAAC,UAAU,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;SACjE,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,WAAW,EAAE,IAAI,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;SACpE,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAC9B,MAAM,IAAI,GAAG,MAAM,OAAO,CAAC,GAAG,CAC5B,OAAO,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,YAAY,CAAC,QAAQ,EAAE,UAAU,EAAE,EAAE,EAAE,IAAI,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC,CAAC,CAClF,CAAC;IACF,OAAO,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;AACvD,CAAC,CAAC;AAEF,gFAAgF;AAChF,2EAA2E;AAC3E,0EAA0E;AAC1E,MAAM,WAAW,GAAG,KAAK,EAAE,QAAgB,EAAE,GAAW,EAAE,GAAgB,EAAiB,EAAE;IAC3F,0EAA0E;IAC1E,wEAAwE;IACxE,+BAA+B;IAC/B,MAAM,OAAO,GAAG,CAAC,MAAM,OAAO,CAAC,GAAG,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;SAC1D,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,WAAW,EAAE,CAAC;SACtC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAC9B,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;QAC3C,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;QAChC,MAAM,QAAQ,GAAG,QAAQ,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAClE,yEAAyE;QACzE,qEAAqE;QACrE,qEAAqE;QACrE,yEAAyE;QACzE,qEAAqE;QACrE,IAAI,QAAQ,KAAK,EAAE,IAAI,QAAQ,KAAK,IAAI,IAAI,CAAC,QAAQ,CAAC,UAAU,CAAC,KAAK,CAAC,EAAE,CAAC;YACxE,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,MAAM,iBAAiB,CAAC,GAAG,EAAE,QAAQ,EAAE,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;QACvE,CAAC;IACH,CAAC;IACD,KAAK,MAAM,IAAI,IAAI,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;QAC7D,MAAM,WAAW,CAAC,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,IAAI,CAAC,EAAE,GAAG,CAAC,CAAC;IACpD,CAAC;AACH,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,QAAQ,GAAG,KAAK,EAAE,QAAgB,EAAwB,EAAE;IACvE,IAAI,CAAC,UAAU,CAAC,QAAQ,CAAC;QAAE,OAAO,EAAE,CAAC;IACrC,MAAM,GAAG,GAAgB,EAAE,CAAC;IAC5B,MAAM,WAAW,CAAC,QAAQ,EAAE,QAAQ,EAAE,GAAG,CAAC,CAAC;IAC3C,OAAO,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;QACvB,IAAI,CAAC,CAAC,QAAQ,KAAK,CAAC,CAAC,QAAQ;YAAE,OAAO,CAAC,CAAC,QAAQ,CAAC,aAAa,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC;QAC3E,IAAI,CAAC,CAAC,UAAU,KAAK,CAAC,CAAC,UAAU;YAAE,OAAO,CAAC,CAAC,UAAU,CAAC,aAAa,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC;QACnF,OAAO,CAAC,CAAC,EAAE,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;IAClC,CAAC,CAAC,CAAC;AACL,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,eAAe,GAAG,CAAC,IAAiB,EAAU,EAAE;IAC3D,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,iBAAiB,CAAC;IAChD,MAAM,IAAI,GAAe;QACvB,CAAC,QAAQ,EAAE,MAAM,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,UAAU,CAAC;QAC/D,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE;YAChB,IAAI,MAAM,GAAG,UAAU,CAAC;YACxB,IAAI,CAAC,CAAC,UAAU,KAAK,IAAI;gBAAE,MAAM,GAAG,MAAM,CAAC;iBACtC,IAAI,CAAC,CAAC,UAAU,KAAK,KAAK;gBAAE,MAAM,GAAG,MAAM,CAAC;YACjD,MAAM,KAAK,GACT,CAAC,CAAC,SAAS,KAAK,IAAI;gBAClB,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,SAAS,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,eAAe,IAAI,CAAC,CAAC,cAAc,GAAG;gBACjF,CAAC,CAAC,GAAG,CAAC;YACV,MAAM,QAAQ,GAAG,CAAC,CAAC,UAAU,KAAK,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,UAAU,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;YACtF,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,UAAU,IAAI,GAAG,EAAE,KAAK,EAAE,QAAQ,CAAC,CAAC;QACxF,CAAC,CAAC;KACH,CAAC;IACF,MAAM,MAAM,GAAG,IAAI,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IAC7B,MAAM,MAAM,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,EAAE,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,MAAM,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC;IACzF,OAAO,IAAI;SACR,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,GAAG,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;SAC9E,IAAI,CAAC,IAAI,CAAC,CAAC;AAChB,CAAC,CAAC;AAWF,MAAM,eAAe,GAAG,KAAK,EAAE,IAAY,EAAuB,EAAE;IAClE,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC;QAAE,OAAO,EAAE,CAAC;IACjC,IAAI,CAAC,MAAM,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,cAAc,EAAE,EAAE,CAAC;QACzC,MAAM,IAAI,KAAK,CAAC,mCAAmC,IAAI,EAAE,CAAC,CAAC;IAC7D,CAAC;IACD,IAAI,MAAwC,CAAC;IAC7C,IAAI,CAAC;QACH,MAAM,GAAG,MAAM,IAAI,CAAC,IAAI,EAAE,SAAS,CAAC,QAAQ,GAAG,SAAS,CAAC,UAAU,CAAC,CAAC;IACvE,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;YAAE,OAAO,EAAE,CAAC;QAClE,IAAK,KAA+B,CAAC,IAAI,KAAK,OAAO,EAAE,CAAC;YACtD,MAAM,IAAI,KAAK,CAAC,mCAAmC,IAAI,EAAE,CAAC,CAAC;QAC7D,CAAC;QACD,MAAM,KAAK,CAAC;IACd,CAAC;IACD,IAAI,IAAY,CAAC;IACjB,IAAI,CAAC;QACH,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,IAAI,EAAE,CAAC;QACrC,IAAI,CAAC,QAAQ,CAAC,MAAM,EAAE,IAAI,QAAQ,CAAC,IAAI,GAAG,mBAAmB,EAAE,CAAC;YAC9D,MAAM,IAAI,KAAK,CACb,oDAAoD,mBAAmB,WAAW,IAAI,EAAE,CACzF,CAAC;QACJ,CAAC;QACD,IAAI,GAAG,CAAC,MAAM,iBAAiB,CAAC,MAAM,EAAE,mBAAmB,EAAE,YAAY,CAAC,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;IAC/F,CAAC;YAAS,CAAC;QACT,MAAM,MAAM,CAAC,KAAK,EAAE,CAAC;IACvB,CAAC;IACD,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAC/B,IAAI,KAAK,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,KAAK,EAAE;QAAE,KAAK,CAAC,GAAG,EAAE,CAAC;IACrC,IAAI,KAAK,CAAC,MAAM,GAAG,oBAAoB,EAAE,CAAC;QACxC,MAAM,IAAI,KAAK,CAAC,qBAAqB,oBAAoB,WAAW,IAAI,EAAE,CAAC,CAAC;IAC9E,CAAC;IACD,OAAO,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE;QAC/B,IAAI,GAAY,CAAC;QACjB,IAAI,CAAC;YACH,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;QACzB,CAAC;QAAC,MAAM,CAAC;YACP,MAAM,IAAI,KAAK,CAAC,yBAAyB,IAAI,IAAI,KAAK,GAAG,CAAC,gBAAgB,CAAC,CAAC;QAC9E,CAAC;QACD,MAAM,MAAM,GAAG,cAAc,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC;QAC7C,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;YACpB,MAAM,IAAI,KAAK,CAAC,yBAAyB,IAAI,IAAI,KAAK,GAAG,CAAC,iBAAiB,CAAC,CAAC;QAC/E,CAAC;QACD,OAAO,MAAM,CAAC,IAAI,CAAC;IACrB,CAAC,CAAC,CAAC;AACL,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,aAAa,GAAG,KAAK,EAChC,MAAc,EACd,OAAO,GAA+E,EAAE,EACpE,EAAE;IACtB,MAAM,CAAC,QAAQ,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,SAAS,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;QACzE,OAAO,CAAC,QAAQ;YACd,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,CAAC;YACvB,CAAC,CAAC,YAAY,CAAC,IAAI,CAAC,MAAM,EAAE,WAAW,CAAC,EAAE,mBAAmB,CAAC;QAChE,OAAO,CAAC,QAAQ;YACd,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,CAAC;YACvB,CAAC,CAAC,YAAY,CAAC,IAAI,CAAC,MAAM,EAAE,UAAU,CAAC,EAAE,iBAAiB,CAAC;QAC7D,OAAO,CAAC,QAAQ;YACd,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,CAAC;YACvB,CAAC,CAAC,YAAY,CAAC,IAAI,CAAC,MAAM,EAAE,cAAc,CAAC,EAAE,mBAAmB,CAAC;QACnE,eAAe,CAAC,IAAI,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;QAC7C,OAAO,CAAC,iBAAiB;YACvB,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;YACrB,CAAC,CAAC,SAAS,CAAC,IAAI,CAAC,MAAM,EAAE,WAAW,CAAC,EAAE,OAAO,CAAC,YAAY,CAAC;KAC/D,CAAC,CAAC;IACH,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,QAAQ,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC;AAC1E,CAAC,CAAC;AAEF,MAAM,gBAAgB,GAAG,CAAC,KAAe,EAAE,MAAiB,EAAQ,EAAE;IACpE,IAAI,MAAM,CAAC,IAAI,EAAE,CAAC;QAChB,KAAK,CAAC,IAAI,CAAC,WAAW,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,SAAS,EAAE,CAAC,CAAC;QACrD,IAAI,MAAM,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC;YACvB,KAAK,CAAC,IAAI,CAAC,WAAW,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;QAChF,CAAC;QACD,IAAI,MAAM,CAAC,IAAI,CAAC,OAAO,EAAE,CAAC;YACxB,KAAK,CAAC,IAAI,CAAC,YAAY,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;QAClF,CAAC;IACH,CAAC;AACH,CAAC,CAAC;AAEF,MAAM,oBAAoB,GAAG,CAAC,KAAe,EAAE,MAAiB,EAAQ,EAAE;IACxE,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;QACpB,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACf,KAAK,CAAC,IAAI,CAAC,YAAY,MAAM,CAAC,QAAQ,CAAC,UAAU,EAAE,CAAC,CAAC;QACrD,KAAK,CAAC,IAAI,CACR,YAAY,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,WAAW,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,OAAO,GAC9E,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,WAAW,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,EAC7E,GAAG,CACJ,CAAC;QACF,MAAM,QAAQ,GAAG,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAC;QAC1C,KAAK,CAAC,IAAI,CACR,aAAa,QAAQ,CAAC,IAAI,GAAG,QAAQ,CAAC,QAAQ,CAAC,CAAC,CAAC,KAAK,QAAQ,CAAC,QAAQ,GAAG,CAAC,CAAC,CAAC,EAAE,GAC7E,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,WAAW,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,EACjD,EAAE,CACH,CAAC;QACF,MAAM,UAAU,GAAG,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC;QAC9C,IAAI,UAAU,EAAE,MAAM,EAAE,CAAC;YACvB,MAAM,GAAG,GAAG,UAAU,CAAC,MAAM,CAAC,GAAG,CAAC;YAClC,KAAK,CAAC,IAAI,CACR,aAAa,UAAU,CAAC,MAAM,CAAC,QAAQ,GACrC,GAAG,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EACxE,EAAE,CACH,CAAC;QACJ,CAAC;QACD,IAAI,UAAU,EAAE,CAAC;YACf,MAAM,OAAO,GAAG,UAAU,CAAC,OAAO,CAAC;YACnC,KAAK,CAAC,IAAI,CACR,aAAa,OAAO,CAAC,QAAQ,IAAI,OAAO,CAAC,IAAI,IAAI,OAAO,CAAC,IAAI,GAC3D,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,QAAQ,OAAO,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,EACxC,EAAE,CACH,CAAC;QACJ,CAAC;QACD,MAAM,CAAC,GAAG,MAAM,CAAC,QAAQ,CAAC,eAAe,CAAC;QAC1C,KAAK,CAAC,IAAI,CAAC,mBAAmB,CAAC,CAAC,GAAG,YAAY,CAAC,CAAC,OAAO,WAAW,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC;QAC/E,MAAM,KAAK,GAAG,MAAM,CAAC,QAAQ,CAAC,YAAY,CAAC;QAC3C,IAAI,KAAK,EAAE,CAAC;YACV,KAAK,CAAC,IAAI,CACR,yBAAyB,KAAK,CAAC,aAAa,gBAAgB,KAAK,CAAC,KAAK,cAAc,KAAK,CAAC,SAAS,WAAW,KAAK,CAAC,MAAM,eAAe,KAAK,CAAC,UAAU,sBAAsB,KAAK,CAAC,gBAAgB,cAAc,KAAK,CAAC,eAAe,EAAE,CAC5O,CAAC;QACJ,CAAC;aAAM,IAAI,MAAM,CAAC,QAAQ,CAAC,aAAa,KAAK,SAAS,EAAE,CAAC;YACvD,KAAK,CAAC,IAAI,CAAC,yBAAyB,MAAM,CAAC,QAAQ,CAAC,aAAa,EAAE,CAAC,CAAC;QACvE,CAAC;QACD,MAAM,KAAK,GAAG,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC;QACpC,KAAK,CAAC,IAAI,CACR,iBAAiB,WAAW,CAAC,KAAK,CAAC,KAAK,CAAC,YAAY,WAAW,CAAC,YAAY,CAAC,KAAK,CAAC,CAAC,EAAE;YACrF,cAAc,WAAW,CAAC,KAAK,CAAC,QAAQ,CAAC,UAAU,WAAW,CAAC,KAAK,CAAC,OAAO,CAAC,GAAG,CACnF,CAAC;QACF,MAAM,MAAM,GAAG,MAAM,CAAC,QAAQ,CAAC,eAAe,CAAC;QAC/C,IAAI,MAAM,EAAE,CAAC;YACX,MAAM,IAAI,GAAG,CAAC,EAAU,EAAU,EAAE,CAAC,GAAG,CAAC,EAAE,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;YAClE,KAAK,CAAC,IAAI,CACR,iBAAiB,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,YAAY,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,YAAY,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,EAAE,CACtG,CAAC;QACJ,CAAC;IACH,CAAC;AACH,CAAC,CAAC;AAEF,MAAM,mBAAmB,GAAG,CAAC,KAAe,EAAE,MAAiB,EAAQ,EAAE;IACvE,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;QACnB,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACf,IAAI,MAAM,GAAG,UAAU,CAAC;QACxB,IAAI,MAAM,CAAC,OAAO,CAAC,UAAU,KAAK,IAAI;YAAE,MAAM,GAAG,MAAM,CAAC;aACnD,IAAI,MAAM,CAAC,OAAO,CAAC,UAAU,KAAK,KAAK;YAAE,MAAM,GAAG,MAAM,CAAC;QAC9D,MAAM,KAAK,GACT,MAAM,CAAC,OAAO,CAAC,SAAS,KAAK,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,OAAO,CAAC,SAAS,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;QAC9F,KAAK,CAAC,IAAI,CACR,YAAY,MAAM,KAAK,KAAK,KAAK,MAAM,CAAC,OAAO,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,MAAM,IAAI,MAAM,CAAC,OAAO,CAAC,UAAU,CAAC,MAAM,GAAG,CACjI,CAAC;QACF,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC;YAC1C,MAAM,IAAI,GAAG,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;YAClC,MAAM,IAAI,GAAG,KAAK,CAAC,CAAC,CAAC,UAAU,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC;YACvD,KAAK,CAAC,IAAI,CAAC,KAAK,IAAI,IAAI,CAAC,CAAC,IAAI,GAAG,IAAI,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;QAChF,CAAC;IACH,CAAC;AACH,CAAC,CAAC;AAEF,MAAM,iBAAiB,GAAG,CAAC,KAAe,EAAE,MAAiB,EAAQ,EAAE;IACrE,IAAI,MAAM,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC7B,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACf,KAAK,CAAC,IAAI,CAAC,WAAW,MAAM,CAAC,MAAM,CAAC,MAAM,QAAQ,CAAC,CAAC;QACpD,MAAM,IAAI,GAAG,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC;QACtC,KAAK,MAAM,CAAC,IAAI,IAAI,EAAE,CAAC;YACrB,KAAK,CAAC,IAAI,CAAC,KAAK,eAAe,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;QACxC,CAAC;IACH,CAAC;AACH,CAAC,CAAC;AAEF,MAAM,oBAAoB,GAAG,CAAC,KAAe,EAAE,MAAiB,EAAQ,EAAE;IACxE,IAAI,MAAM,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAChC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACf,KAAK,CAAC,IAAI,CAAC,cAAc,MAAM,CAAC,SAAS,CAAC,MAAM,QAAQ,CAAC,CAAC;QAC1D,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,EAAE,CAAC;YAC9C,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QACvB,CAAC;QACD,IAAI,MAAM,CAAC,SAAS,CAAC,MAAM,GAAG,EAAE,EAAE,CAAC;YACjC,KAAK,CAAC,IAAI,CAAC,aAAa,MAAM,CAAC,SAAS,CAAC,MAAM,GAAG,EAAE,OAAO,CAAC,CAAC;QAC/D,CAAC;IACH,CAAC;AACH,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,eAAe,GAAG,CAAC,MAAiB,EAAU,EAAE;IAC3D,MAAM,KAAK,GAAG,CAAC,WAAW,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;IAC3C,gBAAgB,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;IAChC,oBAAoB,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;IACpC,mBAAmB,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;IACnC,iBAAiB,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;IACjC,oBAAoB,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;IACpC,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC,CAAC;AAEF,MAAM,eAAe,GAAG,CAAC,KAAe,EAAU,EAAE;IAClD,QAAQ,KAAK,CAAC,IAAI,EAAE,CAAC;QACnB,KAAK,WAAW;YACd,IAAI,KAAK,CAAC,IAAI,KAAK,KAAK;gBAAE,OAAO,mBAAmB,KAAK,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,EAAE,CAAC;YAClF,IAAI,KAAK,CAAC,IAAI,KAAK,SAAS;gBAAE,OAAO,uBAAuB,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,EAAE,CAAC;YACzF,OAAO,sBAAsB,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,EAAE,CAAC;QAC5E,KAAK,WAAW;YACd,IAAI,KAAK,CAAC,IAAI,KAAK,KAAK;gBAAE,OAAO,mBAAmB,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,EAAE,CAAC;YAC9E,IAAI,KAAK,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;gBAC7B,OAAO,uBAAuB,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,GACzD,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC,EACrD,EAAE,CAAC;YACL,CAAC;YACD,OAAO,sBAAsB,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,EAAE,CAAC;QAC5E,KAAK,UAAU;YACb,IAAI,KAAK,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;gBAC/B,MAAM,OAAO,GAAG,qBAAqB,CAAC,SAAS,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC;gBACpE,OAAO,OAAO,CAAC,CAAC,CAAC,cAAc,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,sBAAsB,CAAC;YACzE,CAAC;YACD,IAAI,KAAK,CAAC,IAAI,KAAK,aAAa,EAAE,CAAC;gBACjC,MAAM,OAAO,GAAG,uBAAuB,CAAC,SAAS,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC;gBACtE,IAAI,CAAC,OAAO;oBAAE,OAAO,wBAAwB,CAAC;gBAC9C,OAAO,gBAAgB,OAAO,CAAC,IAAI,KAAK,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,WAAW,GAAG,CAAC;YACtF,CAAC;YACD,OAAO,YAAY,KAAK,CAAC,IAAI,GAAG,CAAC;QACnC,KAAK,WAAW;YACd,OAAO,cAAc,KAAK,CAAC,KAAK,EAAE,CAAC;QACrC,KAAK,OAAO;YACV,OAAO,kBAAkB,KAAK,CAAC,OAAO,EAAE,CAAC;QAC3C,KAAK,MAAM;YACT,OAAO,MAAM,CAAC;QAChB,KAAK,OAAO;YACV,OAAO,UAAU,KAAK,CAAC,MAAM,EAAE,CAAC;IACpC,CAAC;AACH,CAAC,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"loader.d.ts","sourceRoot":"","sources":["../../src/orchestrator/loader.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,KAAK,QAAQ,EAAmC,MAAM,kBAAkB,CAAC;AAElF,eAAO,MAAM,YAAY,aAAoB,MAAM,KAAG,OAAO,CAAC,QAAQ,CAcrE,CAAC"}
|
|
@@ -0,0 +1,17 @@
|
|
|
1
|
+
import { resolve } from "node:path";
|
|
2
|
+
import { pathToFileURL } from "node:url";
|
|
3
|
+
import { EvalCaseSchema, formatZodIssues } from "../core/index.js";
|
|
4
|
+
export const loadEvalCase = async (filePath) => {
|
|
5
|
+
const absolute = resolve(filePath);
|
|
6
|
+
// A file URL, not the raw path: `import("C:\\...")` misparses on Windows.
|
|
7
|
+
const mod = (await import(pathToFileURL(absolute).href));
|
|
8
|
+
if (mod.default === undefined) {
|
|
9
|
+
throw new Error(`Eval case at ${absolute} has no default export. Use \`export default defineEvalCase({ ... })\`.`);
|
|
10
|
+
}
|
|
11
|
+
const result = EvalCaseSchema.safeParse(mod.default);
|
|
12
|
+
if (!result.success) {
|
|
13
|
+
throw new Error(`Invalid eval case at ${absolute}: ${formatZodIssues(result.error)}`);
|
|
14
|
+
}
|
|
15
|
+
return result.data;
|
|
16
|
+
};
|
|
17
|
+
//# sourceMappingURL=loader.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"loader.js","sourceRoot":"","sources":["../../src/orchestrator/loader.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AACzC,OAAO,EAAiB,cAAc,EAAE,eAAe,EAAE,MAAM,kBAAkB,CAAC;AAElF,MAAM,CAAC,MAAM,YAAY,GAAG,KAAK,EAAE,QAAgB,EAAqB,EAAE;IACxE,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,CAAC,CAAC;IACnC,0EAA0E;IAC1E,MAAM,GAAG,GAAG,CAAC,MAAM,MAAM,CAAC,aAAa,CAAC,QAAQ,CAAC,CAAC,IAAI,CAAC,CAA0B,CAAC;IAClF,IAAI,GAAG,CAAC,OAAO,KAAK,SAAS,EAAE,CAAC;QAC9B,MAAM,IAAI,KAAK,CACb,gBAAgB,QAAQ,yEAAyE,CAClG,CAAC;IACJ,CAAC;IACD,MAAM,MAAM,GAAG,cAAc,CAAC,SAAS,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;IACrD,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;QACpB,MAAM,IAAI,KAAK,CAAC,wBAAwB,QAAQ,KAAK,eAAe,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;IACxF,CAAC;IACD,OAAO,MAAM,CAAC,IAAI,CAAC;AACrB,CAAC,CAAC"}
|
|
@@ -0,0 +1,12 @@
|
|
|
1
|
+
export type OptimizationLifecycleEffects<Search, Deployment, Result> = {
|
|
2
|
+
initialSearch: () => Search;
|
|
3
|
+
evaluateSearchSeeds: () => Promise<void>;
|
|
4
|
+
searchCanContinue: (search: Search) => boolean;
|
|
5
|
+
runSearchIteration: (search: Search) => Promise<Search>;
|
|
6
|
+
promote: () => Promise<Deployment>;
|
|
7
|
+
test: (deployment: Deployment) => Promise<Result>;
|
|
8
|
+
};
|
|
9
|
+
type OptimizationPhase = "evaluatingSearchSeeds" | "searching" | "promoting" | "testing" | "succeeded" | "failed";
|
|
10
|
+
export declare const executeOptimizationLifecycle: <Search, Deployment, Result>(effects: OptimizationLifecycleEffects<Search, Deployment, Result>, observe?: (phase: OptimizationPhase) => void) => Promise<Result>;
|
|
11
|
+
export {};
|
|
12
|
+
//# sourceMappingURL=optimization-lifecycle.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"optimization-lifecycle.d.ts","sourceRoot":"","sources":["../../src/orchestrator/optimization-lifecycle.ts"],"names":[],"mappings":"AAEA,MAAM,MAAM,4BAA4B,CAAC,MAAM,EAAE,UAAU,EAAE,MAAM,IAAI;IACrE,aAAa,EAAE,MAAM,MAAM,CAAC;IAC5B,mBAAmB,EAAE,MAAM,OAAO,CAAC,IAAI,CAAC,CAAC;IACzC,iBAAiB,EAAE,CAAC,MAAM,EAAE,MAAM,KAAK,OAAO,CAAC;IAC/C,kBAAkB,EAAE,CAAC,MAAM,EAAE,MAAM,KAAK,OAAO,CAAC,MAAM,CAAC,CAAC;IACxD,OAAO,EAAE,MAAM,OAAO,CAAC,UAAU,CAAC,CAAC;IACnC,IAAI,EAAE,CAAC,UAAU,EAAE,UAAU,KAAK,OAAO,CAAC,MAAM,CAAC,CAAC;CACnD,CAAC;AAEF,KAAK,iBAAiB,GAClB,uBAAuB,GACvB,WAAW,GACX,WAAW,GACX,SAAS,GACT,WAAW,GACX,QAAQ,CAAC;AAEb,eAAO,MAAM,4BAA4B,GAAU,MAAM,EAAE,UAAU,EAAE,MAAM,WAClE,4BAA4B,CAAC,MAAM,EAAE,UAAU,EAAE,MAAM,CAAC,YACvD,CAAC,KAAK,EAAE,iBAAiB,KAAK,IAAI,KAC3C,OAAO,CAAC,MAAM,CAuGhB,CAAC"}
|
|
@@ -0,0 +1,92 @@
|
|
|
1
|
+
import { assign, createActor, fromPromise, setup, toPromise } from "xstate";
|
|
2
|
+
export const executeOptimizationLifecycle = async (effects, observe) => {
|
|
3
|
+
const machine = setup({
|
|
4
|
+
types: {
|
|
5
|
+
context: {},
|
|
6
|
+
output: {},
|
|
7
|
+
},
|
|
8
|
+
actors: {
|
|
9
|
+
evaluateSearchSeeds: fromPromise(() => effects.evaluateSearchSeeds()),
|
|
10
|
+
runSearchIteration: fromPromise(({ input }) => effects.runSearchIteration(input.search)),
|
|
11
|
+
promote: fromPromise(() => effects.promote()),
|
|
12
|
+
test: fromPromise(({ input }) => effects.test(input.deployment)),
|
|
13
|
+
},
|
|
14
|
+
guards: {
|
|
15
|
+
searchCanContinue: ({ context }) => effects.searchCanContinue(context.search),
|
|
16
|
+
},
|
|
17
|
+
actions: {
|
|
18
|
+
advanceSearch: assign({
|
|
19
|
+
search: ({ event }) => ("output" in event ? event.output : effects.initialSearch()),
|
|
20
|
+
}),
|
|
21
|
+
captureDeployment: assign({
|
|
22
|
+
deployment: ({ event }) => ("output" in event ? event.output : undefined),
|
|
23
|
+
}),
|
|
24
|
+
captureResult: assign({
|
|
25
|
+
result: ({ event }) => ("output" in event ? event.output : undefined),
|
|
26
|
+
}),
|
|
27
|
+
captureError: assign({
|
|
28
|
+
error: ({ event }) => "error" in event ? event.error : new Error("optimization lifecycle failed"),
|
|
29
|
+
}),
|
|
30
|
+
},
|
|
31
|
+
}).createMachine({
|
|
32
|
+
id: "lux-optimization",
|
|
33
|
+
context: {
|
|
34
|
+
search: effects.initialSearch(),
|
|
35
|
+
deployment: undefined,
|
|
36
|
+
result: undefined,
|
|
37
|
+
error: undefined,
|
|
38
|
+
},
|
|
39
|
+
initial: "evaluatingSearchSeeds",
|
|
40
|
+
states: {
|
|
41
|
+
evaluatingSearchSeeds: {
|
|
42
|
+
invoke: {
|
|
43
|
+
src: "evaluateSearchSeeds",
|
|
44
|
+
input: ({ context }) => context,
|
|
45
|
+
onDone: { target: "decidingSearch" },
|
|
46
|
+
onError: { target: "failed", actions: "captureError" },
|
|
47
|
+
},
|
|
48
|
+
},
|
|
49
|
+
decidingSearch: {
|
|
50
|
+
always: [{ guard: "searchCanContinue", target: "searching" }, { target: "promoting" }],
|
|
51
|
+
},
|
|
52
|
+
searching: {
|
|
53
|
+
invoke: {
|
|
54
|
+
src: "runSearchIteration",
|
|
55
|
+
input: ({ context }) => context,
|
|
56
|
+
onDone: { target: "decidingSearch", actions: "advanceSearch" },
|
|
57
|
+
onError: { target: "failed", actions: "captureError" },
|
|
58
|
+
},
|
|
59
|
+
},
|
|
60
|
+
promoting: {
|
|
61
|
+
invoke: {
|
|
62
|
+
src: "promote",
|
|
63
|
+
input: ({ context }) => context,
|
|
64
|
+
onDone: { target: "testing", actions: "captureDeployment" },
|
|
65
|
+
onError: { target: "failed", actions: "captureError" },
|
|
66
|
+
},
|
|
67
|
+
},
|
|
68
|
+
testing: {
|
|
69
|
+
invoke: {
|
|
70
|
+
src: "test",
|
|
71
|
+
input: ({ context }) => context,
|
|
72
|
+
onDone: { target: "succeeded", actions: "captureResult" },
|
|
73
|
+
onError: { target: "failed", actions: "captureError" },
|
|
74
|
+
},
|
|
75
|
+
},
|
|
76
|
+
succeeded: { type: "final" },
|
|
77
|
+
failed: { type: "final" },
|
|
78
|
+
},
|
|
79
|
+
output: ({ context }) => context.error === undefined
|
|
80
|
+
? { status: "succeeded", result: context.result }
|
|
81
|
+
: { status: "failed", error: context.error },
|
|
82
|
+
});
|
|
83
|
+
const actor = createActor(machine);
|
|
84
|
+
if (observe)
|
|
85
|
+
actor.subscribe((snapshot) => observe(String(snapshot.value)));
|
|
86
|
+
actor.start();
|
|
87
|
+
const outcome = await toPromise(actor);
|
|
88
|
+
if (outcome.status === "failed")
|
|
89
|
+
throw outcome.error;
|
|
90
|
+
return outcome.result;
|
|
91
|
+
};
|
|
92
|
+
//# sourceMappingURL=optimization-lifecycle.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"optimization-lifecycle.js","sourceRoot":"","sources":["../../src/orchestrator/optimization-lifecycle.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,WAAW,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,QAAQ,CAAC;AAmB5E,MAAM,CAAC,MAAM,4BAA4B,GAAG,KAAK,EAC/C,OAAiE,EACjE,OAA4C,EAC3B,EAAE;IAQnB,MAAM,OAAO,GAAG,KAAK,CAAC;QACpB,KAAK,EAAE;YACL,OAAO,EAAE,EAAa;YACtB,MAAM,EAAE,EAEgC;SACzC;QACD,MAAM,EAAE;YACN,mBAAmB,EAAE,WAAW,CAAgB,GAAG,EAAE,CAAC,OAAO,CAAC,mBAAmB,EAAE,CAAC;YACpF,kBAAkB,EAAE,WAAW,CAAkB,CAAC,EAAE,KAAK,EAAE,EAAE,EAAE,CAC7D,OAAO,CAAC,kBAAkB,CAAC,KAAK,CAAC,MAAM,CAAC,CACzC;YACD,OAAO,EAAE,WAAW,CAAsB,GAAG,EAAE,CAAC,OAAO,CAAC,OAAO,EAAE,CAAC;YAClE,IAAI,EAAE,WAAW,CAAkB,CAAC,EAAE,KAAK,EAAE,EAAE,EAAE,CAC/C,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,UAAwB,CAAC,CAC7C;SACF;QACD,MAAM,EAAE;YACN,iBAAiB,EAAE,CAAC,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC,OAAO,CAAC,iBAAiB,CAAC,OAAO,CAAC,MAAM,CAAC;SAC9E;QACD,OAAO,EAAE;YACP,aAAa,EAAE,MAAM,CAAC;gBACpB,MAAM,EAAE,CAAC,EAAE,KAAK,EAAE,EAAE,EAAE,CAAC,CAAC,QAAQ,IAAI,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,aAAa,EAAE,CAAC;aACpF,CAAC;YACF,iBAAiB,EAAE,MAAM,CAAC;gBACxB,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,EAAE,EAAE,CAAC,CAAC,QAAQ,IAAI,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC;aAC1E,CAAC;YACF,aAAa,EAAE,MAAM,CAAC;gBACpB,MAAM,EAAE,CAAC,EAAE,KAAK,EAAE,EAAE,EAAE,CAAC,CAAC,QAAQ,IAAI,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC;aACtE,CAAC;YACF,YAAY,EAAE,MAAM,CAAC;gBACnB,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,EAAE,EAAE,CACnB,OAAO,IAAI,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,+BAA+B,CAAC;aAC9E,CAAC;SACH;KACF,CAAC,CAAC,aAAa,CAAC;QACf,EAAE,EAAE,kBAAkB;QACtB,OAAO,EAAE;YACP,MAAM,EAAE,OAAO,CAAC,aAAa,EAAE;YAC/B,UAAU,EAAE,SAAS;YACrB,MAAM,EAAE,SAAS;YACjB,KAAK,EAAE,SAAS;SACjB;QACD,OAAO,EAAE,uBAAuB;QAChC,MAAM,EAAE;YACN,qBAAqB,EAAE;gBACrB,MAAM,EAAE;oBACN,GAAG,EAAE,qBAAqB;oBAC1B,KAAK,EAAE,CAAC,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC,OAAO;oBAC/B,MAAM,EAAE,EAAE,MAAM,EAAE,gBAAgB,EAAE;oBACpC,OAAO,EAAE,EAAE,MAAM,EAAE,QAAQ,EAAE,OAAO,EAAE,cAAc,EAAE;iBACvD;aACF;YACD,cAAc,EAAE;gBACd,MAAM,EAAE,CAAC,EAAE,KAAK,EAAE,mBAAmB,EAAE,MAAM,EAAE,WAAW,EAAE,EAAE,EAAE,MAAM,EAAE,WAAW,EAAE,CAAC;aACvF;YACD,SAAS,EAAE;gBACT,MAAM,EAAE;oBACN,GAAG,EAAE,oBAAoB;oBACzB,KAAK,EAAE,CAAC,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC,OAAO;oBAC/B,MAAM,EAAE,EAAE,MAAM,EAAE,gBAAgB,EAAE,OAAO,EAAE,eAAe,EAAE;oBAC9D,OAAO,EAAE,EAAE,MAAM,EAAE,QAAQ,EAAE,OAAO,EAAE,cAAc,EAAE;iBACvD;aACF;YACD,SAAS,EAAE;gBACT,MAAM,EAAE;oBACN,GAAG,EAAE,SAAS;oBACd,KAAK,EAAE,CAAC,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC,OAAO;oBAC/B,MAAM,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,OAAO,EAAE,mBAAmB,EAAE;oBAC3D,OAAO,EAAE,EAAE,MAAM,EAAE,QAAQ,EAAE,OAAO,EAAE,cAAc,EAAE;iBACvD;aACF;YACD,OAAO,EAAE;gBACP,MAAM,EAAE;oBACN,GAAG,EAAE,MAAM;oBACX,KAAK,EAAE,CAAC,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC,OAAO;oBAC/B,MAAM,EAAE,EAAE,MAAM,EAAE,WAAW,EAAE,OAAO,EAAE,eAAe,EAAE;oBACzD,OAAO,EAAE,EAAE,MAAM,EAAE,QAAQ,EAAE,OAAO,EAAE,cAAc,EAAE;iBACvD;aACF;YACD,SAAS,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE;YAC5B,MAAM,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE;SAC1B;QACD,MAAM,EAAE,CAAC,EAAE,OAAO,EAAE,EAAE,EAAE,CACtB,OAAO,CAAC,KAAK,KAAK,SAAS;YACzB,CAAC,CAAC,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE;YACjD,CAAC,CAAC,EAAE,MAAM,EAAE,QAAQ,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE;KACjD,CAAC,CAAC;IAEH,MAAM,KAAK,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC;IACnC,IAAI,OAAO;QAAE,KAAK,CAAC,SAAS,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAsB,CAAC,CAAC,CAAC;IACjG,KAAK,CAAC,KAAK,EAAE,CAAC;IACd,MAAM,OAAO,GAAG,MAAM,SAAS,CAAC,KAAK,CAAC,CAAC;IACvC,IAAI,OAAO,CAAC,MAAM,KAAK,QAAQ;QAAE,MAAM,OAAO,CAAC,KAAK,CAAC;IACrD,OAAO,OAAO,CAAC,MAAgB,CAAC;AAClC,CAAC,CAAC"}
|
|
@@ -0,0 +1,144 @@
|
|
|
1
|
+
import { type AnswererModelCallAccounting, type AnswererRef, type DriverRef, type EvalCase, type GradingAssertionEntry, type GradingReport, type HarnessModelConfig, type LifecycleHooksPlugin, type Registry, type RunMetadata, type RunObserver } from "../core/index.js";
|
|
2
|
+
import { closeRunDirectory, phaseDurations } from "./run-execution.js";
|
|
3
|
+
declare const serializeEffect: (effect: () => Promise<void>) => (() => Promise<void>);
|
|
4
|
+
export type OrchestratorOptions = {
|
|
5
|
+
registry: Registry;
|
|
6
|
+
/** Answerer used when a case names none. Defaults to "persona". */
|
|
7
|
+
defaultAnswerer?: AnswererRef;
|
|
8
|
+
/** Driver used when a case names none. */
|
|
9
|
+
defaultDriver?: DriverRef;
|
|
10
|
+
/** Shared provider/model defaults for built-in harness components. */
|
|
11
|
+
harness?: HarnessModelConfig;
|
|
12
|
+
/** Versioned project resource fixtures. */
|
|
13
|
+
lifecycleFixtures?: LifecycleHooksPlugin[] | undefined;
|
|
14
|
+
};
|
|
15
|
+
export type RunRequest = {
|
|
16
|
+
/** Path to a TS case file. Mutually exclusive with `case`. */
|
|
17
|
+
casePath?: string;
|
|
18
|
+
/** Pre-loaded EvalCase. Mutually exclusive with `casePath`. */
|
|
19
|
+
case?: EvalCase;
|
|
20
|
+
runsRoot: string;
|
|
21
|
+
/**
|
|
22
|
+
* Id to file the run under when the case declares none. Discovery derives
|
|
23
|
+
* it from the file's path relative to the cases root, so two id-less cases
|
|
24
|
+
* that share a basename (a/check.ts, b/check.ts) keep separate histories.
|
|
25
|
+
*/
|
|
26
|
+
caseId?: string | undefined;
|
|
27
|
+
/** Root that relative `fixtures` dir paths resolve against. */
|
|
28
|
+
fixturesRoot?: string | undefined;
|
|
29
|
+
answererOverride?: string | undefined;
|
|
30
|
+
startedAt?: Date | undefined;
|
|
31
|
+
/** The `runSuite` invocation this case attempt belongs to. */
|
|
32
|
+
suiteId?: string | undefined;
|
|
33
|
+
/** Root substituted into values created with `subjectPath()`. */
|
|
34
|
+
subjectRoot?: string | undefined;
|
|
35
|
+
subjectHash?: string | undefined;
|
|
36
|
+
profileId?: string | undefined;
|
|
37
|
+
experimentId?: string | undefined;
|
|
38
|
+
variantId?: string | undefined;
|
|
39
|
+
candidateId?: string | undefined;
|
|
40
|
+
recordGradingAssertion?: (entry: GradingAssertionEntry, index: number) => void | Promise<void>;
|
|
41
|
+
/** Persists subject/answerer spend before any potentially long grading call. */
|
|
42
|
+
recordPreGradingAccounting?: (metadata: RunMetadata) => void | Promise<void>;
|
|
43
|
+
/** Durable model-attempt markers used by experiment ledgers. */
|
|
44
|
+
recordModelAttemptStart?: (kind: "agent" | "answerer" | "judge") => string | Promise<string>;
|
|
45
|
+
recordModelAttemptComplete?: (attemptId: string, accounting: AnswererModelCallAccounting) => void | Promise<void>;
|
|
46
|
+
recordModelAttemptsComplete?: (attemptIds: readonly string[]) => void | Promise<void>;
|
|
47
|
+
abortSignal?: AbortSignal | undefined;
|
|
48
|
+
/** Watches the run as it happens. See {@link RunObserver}. */
|
|
49
|
+
observer?: RunObserver | undefined;
|
|
50
|
+
};
|
|
51
|
+
export type RunResult = {
|
|
52
|
+
runDir: string;
|
|
53
|
+
metadata: RunMetadata;
|
|
54
|
+
grading: GradingReport;
|
|
55
|
+
exitReason: RunMetadata["exitReason"];
|
|
56
|
+
};
|
|
57
|
+
export type SuiteCaseResult = {
|
|
58
|
+
caseId: string;
|
|
59
|
+
runDir: string | null;
|
|
60
|
+
exitReason: RunMetadata["exitReason"];
|
|
61
|
+
casePassed: boolean | null;
|
|
62
|
+
errorMessage?: string;
|
|
63
|
+
/** The full result, when the case got far enough to produce one. */
|
|
64
|
+
result?: RunResult;
|
|
65
|
+
};
|
|
66
|
+
/** Watches one case within a suite; returned by {@link SuiteObserver.caseStart}. */
|
|
67
|
+
export type CaseObserver = RunObserver & {
|
|
68
|
+
/** The case settled. Receives the same entry that lands in `results`. */
|
|
69
|
+
end?(result: SuiteCaseResult): void;
|
|
70
|
+
};
|
|
71
|
+
/** A {@link RunObserver} that also sees case boundaries within a suite run. */
|
|
72
|
+
export type SuiteObserver = RunObserver & {
|
|
73
|
+
/**
|
|
74
|
+
* Called when a worker picks the case up. Return a {@link CaseObserver} to
|
|
75
|
+
* receive that case's events under their own identity — the only way to
|
|
76
|
+
* tell cases apart when `concurrency > 1`. Returning nothing routes the
|
|
77
|
+
* case's events to this observer's own {@link RunObserver} methods.
|
|
78
|
+
*/
|
|
79
|
+
caseStart?(info: {
|
|
80
|
+
caseId: string;
|
|
81
|
+
index: number;
|
|
82
|
+
total: number;
|
|
83
|
+
}): CaseObserver | void;
|
|
84
|
+
};
|
|
85
|
+
/**
|
|
86
|
+
* One case to run in a suite: its file and the id to display it under.
|
|
87
|
+
* `caseId` is the canonical id runs are filed under when the case declares
|
|
88
|
+
* none — display decorations (a `--loop` suffix) must not fork run history.
|
|
89
|
+
*/
|
|
90
|
+
export type SuiteItem = {
|
|
91
|
+
path: string;
|
|
92
|
+
id: string;
|
|
93
|
+
caseId?: string | undefined;
|
|
94
|
+
};
|
|
95
|
+
export declare const resolveDriverReference: (evalCase: EvalCase, fallback: DriverRef | undefined) => DriverRef;
|
|
96
|
+
export declare const resolveAnswererReference: (evalCase: EvalCase, fallback: AnswererRef, harness: HarnessModelConfig | undefined, override?: string) => AnswererRef;
|
|
97
|
+
export declare const boundExperimentDriverConfig: (driverName: string, config: unknown, experimentId: string | undefined) => unknown;
|
|
98
|
+
declare const closeQuietly: (close: () => Promise<void>, label: string) => Promise<void>;
|
|
99
|
+
export declare class Orchestrator {
|
|
100
|
+
private evaluatorHash;
|
|
101
|
+
readonly registry: Registry;
|
|
102
|
+
private readonly defaultAnswerer;
|
|
103
|
+
private readonly defaultDriver;
|
|
104
|
+
private readonly harness;
|
|
105
|
+
private readonly lifecycleFixtures;
|
|
106
|
+
constructor(options: OrchestratorOptions);
|
|
107
|
+
private evaluatorImplementationIdentity;
|
|
108
|
+
run(request: RunRequest): Promise<RunResult>;
|
|
109
|
+
/**
|
|
110
|
+
* Run cases through a bounded pool of `concurrency` workers, continuing on
|
|
111
|
+
* per-case failures and aggregating results in input order. `items` is the
|
|
112
|
+
* exact run list — repetition (`--loop`) is the caller repeating entries.
|
|
113
|
+
*/
|
|
114
|
+
runSuite(items: SuiteItem[], opts: {
|
|
115
|
+
runsRoot: string;
|
|
116
|
+
fixturesRoot?: string | undefined;
|
|
117
|
+
answererOverride?: string | undefined;
|
|
118
|
+
observer?: SuiteObserver | undefined;
|
|
119
|
+
/** Aborts the cases in flight and stops the pool pulling new ones. */
|
|
120
|
+
abortSignal?: AbortSignal | undefined;
|
|
121
|
+
/** How many cases may run at once. Defaults to 1 (sequential). */
|
|
122
|
+
concurrency?: number | undefined;
|
|
123
|
+
subjectRoot?: string | undefined;
|
|
124
|
+
subjectHash?: string | undefined;
|
|
125
|
+
profileId?: string | undefined;
|
|
126
|
+
experimentId?: string | undefined;
|
|
127
|
+
experimentUnitId?: string | undefined;
|
|
128
|
+
experimentAttemptId?: string | undefined;
|
|
129
|
+
variantId?: string | undefined;
|
|
130
|
+
candidateId?: string | undefined;
|
|
131
|
+
recordGradingAssertion?: (entry: GradingAssertionEntry, index: number) => void | Promise<void>;
|
|
132
|
+
recordPreGradingAccounting?: (metadata: RunMetadata) => void | Promise<void>;
|
|
133
|
+
recordModelAttemptStart?: (kind: "agent" | "answerer" | "judge") => string | Promise<string>;
|
|
134
|
+
recordModelAttemptComplete?: (attemptId: string, accounting: AnswererModelCallAccounting) => void | Promise<void>;
|
|
135
|
+
recordModelAttemptsComplete?: (attemptIds: readonly string[]) => void | Promise<void>;
|
|
136
|
+
}): Promise<{
|
|
137
|
+
results: SuiteCaseResult[];
|
|
138
|
+
aborted: boolean;
|
|
139
|
+
suiteDir: string;
|
|
140
|
+
}>;
|
|
141
|
+
private runWithPlugins;
|
|
142
|
+
}
|
|
143
|
+
export { closeQuietly, closeRunDirectory, phaseDurations, serializeEffect };
|
|
144
|
+
//# sourceMappingURL=orchestrator.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"orchestrator.d.ts","sourceRoot":"","sources":["../../src/orchestrator/orchestrator.ts"],"names":[],"mappings":"AACA,OAAO,EAEL,KAAK,2BAA2B,EAChC,KAAK,WAAW,EAEhB,KAAK,SAAS,EACd,KAAK,QAAQ,EACb,KAAK,qBAAqB,EAC1B,KAAK,aAAa,EAElB,KAAK,kBAAkB,EACvB,KAAK,oBAAoB,EAEzB,KAAK,QAAQ,EACb,KAAK,WAAW,EAChB,KAAK,WAAW,EAGjB,MAAM,kBAAkB,CAAC;AAW1B,OAAO,EACL,iBAAiB,EAGjB,cAAc,EAGf,MAAM,oBAAoB,CAAC;AAW5B,QAAA,MAAM,eAAe,WAAY,MAAM,OAAO,CAAC,IAAI,CAAC,KAAG,CAAC,MAAM,OAAO,CAAC,IAAI,CAAC,CAG1E,CAAC;AA8CF,MAAM,MAAM,mBAAmB,GAAG;IAChC,QAAQ,EAAE,QAAQ,CAAC;IACnB,mEAAmE;IACnE,eAAe,CAAC,EAAE,WAAW,CAAC;IAC9B,0CAA0C;IAC1C,aAAa,CAAC,EAAE,SAAS,CAAC;IAC1B,sEAAsE;IACtE,OAAO,CAAC,EAAE,kBAAkB,CAAC;IAC7B,2CAA2C;IAC3C,iBAAiB,CAAC,EAAE,oBAAoB,EAAE,GAAG,SAAS,CAAC;CACxD,CAAC;AAEF,MAAM,MAAM,UAAU,GAAG;IACvB,8DAA8D;IAC9D,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,+DAA+D;IAC/D,IAAI,CAAC,EAAE,QAAQ,CAAC;IAChB,QAAQ,EAAE,MAAM,CAAC;IACjB;;;;OAIG;IACH,MAAM,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC5B,+DAA+D;IAC/D,YAAY,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAClC,gBAAgB,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACtC,SAAS,CAAC,EAAE,IAAI,GAAG,SAAS,CAAC;IAC7B,8DAA8D;IAC9D,OAAO,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC7B,iEAAiE;IACjE,WAAW,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACjC,WAAW,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACjC,SAAS,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC/B,YAAY,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAClC,SAAS,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC/B,WAAW,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACjC,sBAAsB,CAAC,EAAE,CAAC,KAAK,EAAE,qBAAqB,EAAE,KAAK,EAAE,MAAM,KAAK,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC/F,gFAAgF;IAChF,0BAA0B,CAAC,EAAE,CAAC,QAAQ,EAAE,WAAW,KAAK,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC7E,gEAAgE;IAChE,uBAAuB,CAAC,EAAE,CAAC,IAAI,EAAE,OAAO,GAAG,UAAU,GAAG,OAAO,KAAK,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;IAC7F,0BAA0B,CAAC,EAAE,CAC3B,SAAS,EAAE,MAAM,EACjB,UAAU,EAAE,2BAA2B,KACpC,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC1B,2BAA2B,CAAC,EAAE,CAAC,UAAU,EAAE,SAAS,MAAM,EAAE,KAAK,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IACtF,WAAW,CAAC,EAAE,WAAW,GAAG,SAAS,CAAC;IACtC,8DAA8D;IAC9D,QAAQ,CAAC,EAAE,WAAW,GAAG,SAAS,CAAC;CACpC,CAAC;AAEF,MAAM,MAAM,SAAS,GAAG;IACtB,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,WAAW,CAAC;IACtB,OAAO,EAAE,aAAa,CAAC;IACvB,UAAU,EAAE,WAAW,CAAC,YAAY,CAAC,CAAC;CACvC,CAAC;AAEF,MAAM,MAAM,eAAe,GAAG;IAC5B,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,UAAU,EAAE,WAAW,CAAC,YAAY,CAAC,CAAC;IACtC,UAAU,EAAE,OAAO,GAAG,IAAI,CAAC;IAC3B,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,oEAAoE;IACpE,MAAM,CAAC,EAAE,SAAS,CAAC;CACpB,CAAC;AAEF,oFAAoF;AACpF,MAAM,MAAM,YAAY,GAAG,WAAW,GAAG;IACvC,yEAAyE;IACzE,GAAG,CAAC,CAAC,MAAM,EAAE,eAAe,GAAG,IAAI,CAAC;CACrC,CAAC;AAEF,+EAA+E;AAC/E,MAAM,MAAM,aAAa,GAAG,WAAW,GAAG;IACxC;;;;;OAKG;IAEH,SAAS,CAAC,CAAC,IAAI,EAAE;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,GAAG,YAAY,GAAG,IAAI,CAAC;CACzF,CAAC;AAEF;;;;GAIG;AACH,MAAM,MAAM,SAAS,GAAG;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,MAAM,CAAC,EAAE,MAAM,GAAG,SAAS,CAAA;CAAE,CAAC;AA0BlF,eAAO,MAAM,sBAAsB,aACvB,QAAQ,YACR,SAAS,GAAG,SAAS,KAC9B,SAgBF,CAAC;AAEF,eAAO,MAAM,wBAAwB,aACzB,QAAQ,YACR,WAAW,WACZ,kBAAkB,GAAG,SAAS,aAC5B,MAAM,KAChB,WAQF,CAAC;AAuLF,eAAO,MAAM,2BAA2B,eAC1B,MAAM,UACV,OAAO,gBACD,MAAM,GAAG,SAAS,KAC/B,OAQF,CAAC;AAEF,QAAA,MAAM,YAAY,UAAiB,MAAM,OAAO,CAAC,IAAI,CAAC,SAAS,MAAM,KAAG,OAAO,CAAC,IAAI,CAQnF,CAAC;AAEF,qBAAa,YAAY;IACvB,OAAO,CAAC,aAAa,CAA8B;IACnD,QAAQ,CAAC,QAAQ,EAAE,QAAQ,CAAC;IAC5B,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAc;IAC9C,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAwB;IACtD,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAiC;IACzD,OAAO,CAAC,QAAQ,CAAC,iBAAiB,CAAkC;IAEpE,YAAY,OAAO,EAAE,mBAAmB,EAMvC;IAED,OAAO,CAAC,+BAA+B;IAKjC,GAAG,CAAC,OAAO,EAAE,UAAU,GAAG,OAAO,CAAC,SAAS,CAAC,CA+GjD;IAED;;;;OAIG;IACG,QAAQ,CACZ,KAAK,EAAE,SAAS,EAAE,EAClB,IAAI,EAAE;QACJ,QAAQ,EAAE,MAAM,CAAC;QACjB,YAAY,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;QAClC,gBAAgB,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;QACtC,QAAQ,CAAC,EAAE,aAAa,GAAG,SAAS,CAAC;QACrC,sEAAsE;QACtE,WAAW,CAAC,EAAE,WAAW,GAAG,SAAS,CAAC;QACtC,kEAAkE;QAClE,WAAW,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;QACjC,WAAW,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;QACjC,WAAW,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;QACjC,SAAS,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;QAC/B,YAAY,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;QAClC,gBAAgB,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;QACtC,mBAAmB,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;QACzC,SAAS,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;QAC/B,WAAW,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;QACjC,sBAAsB,CAAC,EAAE,CACvB,KAAK,EAAE,qBAAqB,EAC5B,KAAK,EAAE,MAAM,KACV,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;QAC1B,0BAA0B,CAAC,EAAE,CAAC,QAAQ,EAAE,WAAW,KAAK,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;QAC7E,uBAAuB,CAAC,EAAE,CAAC,IAAI,EAAE,OAAO,GAAG,UAAU,GAAG,OAAO,KAAK,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;QAC7F,0BAA0B,CAAC,EAAE,CAC3B,SAAS,EAAE,MAAM,EACjB,UAAU,EAAE,2BAA2B,KACpC,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;QAC1B,2BAA2B,CAAC,EAAE,CAAC,UAAU,EAAE,SAAS,MAAM,EAAE,KAAK,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;KACvF,GACA,OAAO,CAAC;QAAE,OAAO,EAAE,eAAe,EAAE,CAAC;QAAC,OAAO,EAAE,OAAO,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAA;KAAE,CAAC,CAsR7E;IAED,OAAO,CAAC,cAAc;CASvB;AAGD,OAAO,EAAE,YAAY,EAAE,iBAAiB,EAAE,cAAc,EAAE,eAAe,EAAE,CAAC"}
|