@xoxoai/checkmate 0.6.7 → 0.7.0
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/dist/checkmate.js +13375 -0
- package/package.json +14 -100
- package/plugin/.claude-plugin/plugin.json +10 -0
- package/plugin/bin/checkmate.ts +3 -0
- package/plugin/contracts/finding.schema.json +78 -0
- package/plugin/contracts/gap.schema.json +56 -0
- package/plugin/contracts/plan-candidate.schema.json +103 -0
- package/plugin/contracts/plan-delta.schema.json +63 -0
- package/plugin/contracts/plan-patch.schema.json +116 -0
- package/plugin/contracts/plan-request.schema.json +108 -0
- package/plugin/contracts/plan.schema.json +313 -0
- package/plugin/contracts/profile.schema.json +154 -0
- package/plugin/contracts/response.schema.json +25 -0
- package/plugin/contracts/result.schema.json +410 -0
- package/plugin/dist/checkmate.js +13012 -0
- package/plugin/package.json +4 -0
- package/plugin/references/browser.md +35 -0
- package/plugin/references/http.md +30 -0
- package/plugin/references/workflow.md +122 -0
- package/plugin/skills/execute/SKILL.md +92 -0
- package/plugin/skills/plan/SKILL.md +94 -0
- package/plugin/skills/report/SKILL.md +28 -0
- package/plugin/skills/revise/SKILL.md +43 -0
- package/plugin/skills/setup/SKILL.md +82 -0
- package/plugin/skills/status/SKILL.md +20 -0
- package/plugin/skills/ui/SKILL.md +26 -0
- package/plugin/src/actor.ts +21 -0
- package/plugin/src/cli.ts +1133 -0
- package/plugin/src/contracts.ts +342 -0
- package/plugin/src/index.ts +52 -0
- package/plugin/src/plans.ts +1028 -0
- package/plugin/src/render.ts +544 -0
- package/plugin/src/runs.ts +868 -0
- package/plugin/src/store.ts +384 -0
- package/plugin/src/targets.ts +205 -0
- package/plugin/src/types.ts +319 -0
- package/plugin/src/ui/assets/client.js +501 -0
- package/plugin/src/ui/assets/favicon.svg +1 -0
- package/plugin/src/ui/assets/fonts/inter-OFL.txt +93 -0
- package/plugin/src/ui/assets/fonts/inter.woff2 +0 -0
- package/plugin/src/ui/assets/fonts/jetbrains-mono-OFL.txt +93 -0
- package/plugin/src/ui/assets/fonts/jetbrains-mono.woff2 +0 -0
- package/plugin/src/ui/assets/fonts/space-grotesk-OFL.txt +93 -0
- package/plugin/src/ui/assets/fonts/space-grotesk.woff2 +0 -0
- package/plugin/src/ui/assets/style.css +1926 -0
- package/plugin/src/ui/edit.ts +305 -0
- package/plugin/src/ui/forms.ts +209 -0
- package/plugin/src/ui/html.ts +130 -0
- package/plugin/src/ui/layout.ts +198 -0
- package/plugin/src/ui/lists.ts +645 -0
- package/plugin/src/ui/plan.ts +577 -0
- package/plugin/src/ui/run.ts +489 -0
- package/plugin/src/ui/server.ts +684 -0
- package/plugin/src/validators.d.ts +13 -0
- package/plugin/src/validators.js +1 -0
- package/plugin/src/views.ts +839 -0
- package/LICENSE +0 -21
- package/README.md +0 -294
- package/bin/checkmate.js +0 -13
- package/dist/ai/client.d.ts +0 -64
- package/dist/ai/client.d.ts.map +0 -1
- package/dist/ai/client.js +0 -313
- package/dist/ai/client.js.map +0 -1
- package/dist/ai/message-handler.d.ts +0 -9
- package/dist/ai/message-handler.d.ts.map +0 -1
- package/dist/ai/message-handler.js +0 -50
- package/dist/ai/message-handler.js.map +0 -1
- package/dist/ai/message-history.d.ts +0 -17
- package/dist/ai/message-history.d.ts.map +0 -1
- package/dist/ai/message-history.js +0 -45
- package/dist/ai/message-history.js.map +0 -1
- package/dist/ai/prompts.d.ts +0 -4
- package/dist/ai/prompts.d.ts.map +0 -1
- package/dist/ai/prompts.js +0 -24
- package/dist/ai/prompts.js.map +0 -1
- package/dist/ai/rate-limit-policy.d.ts +0 -9
- package/dist/ai/rate-limit-policy.d.ts.map +0 -1
- package/dist/ai/rate-limit-policy.js +0 -34
- package/dist/ai/rate-limit-policy.js.map +0 -1
- package/dist/ai/tool-response-handler.d.ts +0 -16
- package/dist/ai/tool-response-handler.d.ts.map +0 -1
- package/dist/ai/tool-response-handler.js +0 -85
- package/dist/ai/tool-response-handler.js.map +0 -1
- package/dist/ai/turn-processor.d.ts +0 -34
- package/dist/ai/turn-processor.d.ts.map +0 -1
- package/dist/ai/turn-processor.js +0 -107
- package/dist/ai/turn-processor.js.map +0 -1
- package/dist/api/allocate-run-identity.d.ts +0 -8
- package/dist/api/allocate-run-identity.d.ts.map +0 -1
- package/dist/api/allocate-run-identity.js +0 -10
- package/dist/api/allocate-run-identity.js.map +0 -1
- package/dist/api/execute-prepared-run.d.ts +0 -22
- package/dist/api/execute-prepared-run.d.ts.map +0 -1
- package/dist/api/execute-prepared-run.js +0 -184
- package/dist/api/execute-prepared-run.js.map +0 -1
- package/dist/api/index.d.ts +0 -25
- package/dist/api/index.d.ts.map +0 -1
- package/dist/api/index.js +0 -69
- package/dist/api/index.js.map +0 -1
- package/dist/api/prepare-run.d.ts +0 -44
- package/dist/api/prepare-run.d.ts.map +0 -1
- package/dist/api/prepare-run.js +0 -261
- package/dist/api/prepare-run.js.map +0 -1
- package/dist/cli/diagnostic-relay.d.ts +0 -6
- package/dist/cli/diagnostic-relay.d.ts.map +0 -1
- package/dist/cli/diagnostic-relay.js +0 -6
- package/dist/cli/diagnostic-relay.js.map +0 -1
- package/dist/cli/main.d.ts +0 -21
- package/dist/cli/main.d.ts.map +0 -1
- package/dist/cli/main.js +0 -166
- package/dist/cli/main.js.map +0 -1
- package/dist/cli/protocol.d.ts +0 -57
- package/dist/cli/protocol.d.ts.map +0 -1
- package/dist/cli/protocol.js +0 -116
- package/dist/cli/protocol.js.map +0 -1
- package/dist/cli/reconcile-result.d.ts +0 -13
- package/dist/cli/reconcile-result.d.ts.map +0 -1
- package/dist/cli/reconcile-result.js +0 -300
- package/dist/cli/reconcile-result.js.map +0 -1
- package/dist/cli/run-parent.d.ts +0 -37
- package/dist/cli/run-parent.d.ts.map +0 -1
- package/dist/cli/run-parent.js +0 -491
- package/dist/cli/run-parent.js.map +0 -1
- package/dist/cli/signals.d.ts +0 -36
- package/dist/cli/signals.d.ts.map +0 -1
- package/dist/cli/signals.js +0 -85
- package/dist/cli/signals.js.map +0 -1
- package/dist/cli/static-commands.d.ts +0 -11
- package/dist/cli/static-commands.d.ts.map +0 -1
- package/dist/cli/static-commands.js +0 -83
- package/dist/cli/static-commands.js.map +0 -1
- package/dist/cli/worker-stream-guard.d.ts +0 -20
- package/dist/cli/worker-stream-guard.d.ts.map +0 -1
- package/dist/cli/worker-stream-guard.js +0 -41
- package/dist/cli/worker-stream-guard.js.map +0 -1
- package/dist/cli/worker.d.ts +0 -3
- package/dist/cli/worker.d.ts.map +0 -1
- package/dist/cli/worker.js +0 -111
- package/dist/cli/worker.js.map +0 -1
- package/dist/config/ingestion.d.ts +0 -58
- package/dist/config/ingestion.d.ts.map +0 -1
- package/dist/config/ingestion.js +0 -283
- package/dist/config/ingestion.js.map +0 -1
- package/dist/config/manifest.d.ts +0 -20
- package/dist/config/manifest.d.ts.map +0 -1
- package/dist/config/manifest.js +0 -91
- package/dist/config/manifest.js.map +0 -1
- package/dist/config/model-egress.d.ts +0 -10
- package/dist/config/model-egress.d.ts.map +0 -1
- package/dist/config/model-egress.js +0 -62
- package/dist/config/model-egress.js.map +0 -1
- package/dist/config/package-resolution.d.ts +0 -4
- package/dist/config/package-resolution.d.ts.map +0 -1
- package/dist/config/package-resolution.js +0 -24
- package/dist/config/package-resolution.js.map +0 -1
- package/dist/config/policy.d.ts +0 -44
- package/dist/config/policy.d.ts.map +0 -1
- package/dist/config/policy.js +0 -141
- package/dist/config/policy.js.map +0 -1
- package/dist/config/record.d.ts +0 -4
- package/dist/config/record.d.ts.map +0 -1
- package/dist/config/record.js +0 -10
- package/dist/config/record.js.map +0 -1
- package/dist/config/secrets.d.ts +0 -11
- package/dist/config/secrets.d.ts.map +0 -1
- package/dist/config/secrets.js +0 -42
- package/dist/config/secrets.js.map +0 -1
- package/dist/contracts/diagnostics.d.ts +0 -4
- package/dist/contracts/diagnostics.d.ts.map +0 -1
- package/dist/contracts/diagnostics.js +0 -31
- package/dist/contracts/diagnostics.js.map +0 -1
- package/dist/contracts/serialize.d.ts +0 -2
- package/dist/contracts/serialize.d.ts.map +0 -1
- package/dist/contracts/serialize.js +0 -90
- package/dist/contracts/serialize.js.map +0 -1
- package/dist/contracts/types.d.ts +0 -316
- package/dist/contracts/types.d.ts.map +0 -1
- package/dist/contracts/types.js +0 -74
- package/dist/contracts/types.js.map +0 -1
- package/dist/contracts/validator.d.ts +0 -8
- package/dist/contracts/validator.d.ts.map +0 -1
- package/dist/contracts/validator.js +0 -71
- package/dist/contracts/validator.js.map +0 -1
- package/dist/driver.d.ts +0 -97
- package/dist/driver.d.ts.map +0 -1
- package/dist/driver.js +0 -24
- package/dist/driver.js.map +0 -1
- package/dist/drivers/descriptor.d.ts +0 -15
- package/dist/drivers/descriptor.d.ts.map +0 -1
- package/dist/drivers/descriptor.js +0 -122
- package/dist/drivers/descriptor.js.map +0 -1
- package/dist/drivers/loader.d.ts +0 -18
- package/dist/drivers/loader.d.ts.map +0 -1
- package/dist/drivers/loader.js +0 -97
- package/dist/drivers/loader.js.map +0 -1
- package/dist/drivers/web/checkmate-driver.json +0 -47
- package/dist/drivers/web/index.d.ts +0 -6
- package/dist/drivers/web/index.d.ts.map +0 -1
- package/dist/drivers/web/index.js +0 -27
- package/dist/drivers/web/index.js.map +0 -1
- package/dist/drivers/web/session.d.ts +0 -25
- package/dist/drivers/web/session.d.ts.map +0 -1
- package/dist/drivers/web/session.js +0 -132
- package/dist/drivers/web/session.js.map +0 -1
- package/dist/drivers/web/tools/network-request-recorder.d.ts +0 -46
- package/dist/drivers/web/tools/network-request-recorder.d.ts.map +0 -1
- package/dist/drivers/web/tools/network-request-recorder.js +0 -164
- package/dist/drivers/web/tools/network-request-recorder.js.map +0 -1
- package/dist/drivers/web/tools/screenshot-service.d.ts +0 -10
- package/dist/drivers/web/tools/screenshot-service.d.ts.map +0 -1
- package/dist/drivers/web/tools/screenshot-service.js +0 -23
- package/dist/drivers/web/tools/screenshot-service.js.map +0 -1
- package/dist/drivers/web/tools/snapshot-filter/index.d.ts +0 -4
- package/dist/drivers/web/tools/snapshot-filter/index.d.ts.map +0 -1
- package/dist/drivers/web/tools/snapshot-filter/index.js +0 -4
- package/dist/drivers/web/tools/snapshot-filter/index.js.map +0 -1
- package/dist/drivers/web/tools/snapshot-filter/semantic-scorer.d.ts +0 -15
- package/dist/drivers/web/tools/snapshot-filter/semantic-scorer.d.ts.map +0 -1
- package/dist/drivers/web/tools/snapshot-filter/semantic-scorer.js +0 -99
- package/dist/drivers/web/tools/snapshot-filter/semantic-scorer.js.map +0 -1
- package/dist/drivers/web/tools/snapshot-filter/snapshot-filter.d.ts +0 -5
- package/dist/drivers/web/tools/snapshot-filter/snapshot-filter.d.ts.map +0 -1
- package/dist/drivers/web/tools/snapshot-filter/snapshot-filter.js +0 -56
- package/dist/drivers/web/tools/snapshot-filter/snapshot-filter.js.map +0 -1
- package/dist/drivers/web/tools/snapshot-filter/tree-reconstructor.d.ts +0 -3
- package/dist/drivers/web/tools/snapshot-filter/tree-reconstructor.d.ts.map +0 -1
- package/dist/drivers/web/tools/snapshot-filter/tree-reconstructor.js +0 -85
- package/dist/drivers/web/tools/snapshot-filter/tree-reconstructor.js.map +0 -1
- package/dist/drivers/web/tools/snapshot-service.d.ts +0 -31
- package/dist/drivers/web/tools/snapshot-service.d.ts.map +0 -1
- package/dist/drivers/web/tools/snapshot-service.js +0 -96
- package/dist/drivers/web/tools/snapshot-service.js.map +0 -1
- package/dist/drivers/web/tools/tool.d.ts +0 -76
- package/dist/drivers/web/tools/tool.d.ts.map +0 -1
- package/dist/drivers/web/tools/tool.js +0 -562
- package/dist/drivers/web/tools/tool.js.map +0 -1
- package/dist/drivers/web/tools/transient-state-tracker.d.ts +0 -26
- package/dist/drivers/web/tools/transient-state-tracker.d.ts.map +0 -1
- package/dist/drivers/web/tools/transient-state-tracker.js +0 -301
- package/dist/drivers/web/tools/transient-state-tracker.js.map +0 -1
- package/dist/drivers/web/tools/types.d.ts +0 -11
- package/dist/drivers/web/tools/types.d.ts.map +0 -1
- package/dist/drivers/web/tools/types.js +0 -2
- package/dist/drivers/web/tools/types.js.map +0 -1
- package/dist/evidence/atomic-file.d.ts +0 -23
- package/dist/evidence/atomic-file.d.ts.map +0 -1
- package/dist/evidence/atomic-file.js +0 -104
- package/dist/evidence/atomic-file.js.map +0 -1
- package/dist/evidence/layout.d.ts +0 -30
- package/dist/evidence/layout.d.ts.map +0 -1
- package/dist/evidence/layout.js +0 -152
- package/dist/evidence/layout.js.map +0 -1
- package/dist/evidence/retention.d.ts +0 -5
- package/dist/evidence/retention.d.ts.map +0 -1
- package/dist/evidence/retention.js +0 -10
- package/dist/evidence/retention.js.map +0 -1
- package/dist/evidence/store.d.ts +0 -101
- package/dist/evidence/store.d.ts.map +0 -1
- package/dist/evidence/store.js +0 -375
- package/dist/evidence/store.js.map +0 -1
- package/dist/evidence/terminal-finalizer.d.ts +0 -19
- package/dist/evidence/terminal-finalizer.d.ts.map +0 -1
- package/dist/evidence/terminal-finalizer.js +0 -56
- package/dist/evidence/terminal-finalizer.js.map +0 -1
- package/dist/index.d.ts +0 -9
- package/dist/index.d.ts.map +0 -1
- package/dist/index.js +0 -12
- package/dist/index.js.map +0 -1
- package/dist/logging/invocation-logger.d.ts +0 -4
- package/dist/logging/invocation-logger.d.ts.map +0 -1
- package/dist/logging/invocation-logger.js +0 -25
- package/dist/logging/invocation-logger.js.map +0 -1
- package/dist/logging/level-logger.d.ts +0 -11
- package/dist/logging/level-logger.d.ts.map +0 -1
- package/dist/logging/level-logger.js +0 -82
- package/dist/logging/level-logger.js.map +0 -1
- package/dist/logging/types.d.ts +0 -8
- package/dist/logging/types.d.ts.map +0 -1
- package/dist/logging/types.js +0 -7
- package/dist/logging/types.js.map +0 -1
- package/dist/redaction/diagnostic-sanitizer.d.ts +0 -7
- package/dist/redaction/diagnostic-sanitizer.d.ts.map +0 -1
- package/dist/redaction/diagnostic-sanitizer.js +0 -42
- package/dist/redaction/diagnostic-sanitizer.js.map +0 -1
- package/dist/redaction/redactor.d.ts +0 -40
- package/dist/redaction/redactor.d.ts.map +0 -1
- package/dist/redaction/redactor.js +0 -176
- package/dist/redaction/redactor.js.map +0 -1
- package/dist/redaction/scrub.d.ts +0 -29
- package/dist/redaction/scrub.d.ts.map +0 -1
- package/dist/redaction/scrub.js +0 -54
- package/dist/redaction/scrub.js.map +0 -1
- package/dist/runtime/config.d.ts +0 -16
- package/dist/runtime/config.d.ts.map +0 -1
- package/dist/runtime/config.js +0 -2
- package/dist/runtime/config.js.map +0 -1
- package/dist/runtime/driver-boundary.d.ts +0 -16
- package/dist/runtime/driver-boundary.d.ts.map +0 -1
- package/dist/runtime/driver-boundary.js +0 -71
- package/dist/runtime/driver-boundary.js.map +0 -1
- package/dist/runtime/driver-session.d.ts +0 -9
- package/dist/runtime/driver-session.d.ts.map +0 -1
- package/dist/runtime/driver-session.js +0 -64
- package/dist/runtime/driver-session.js.map +0 -1
- package/dist/runtime/internal-step-evidence.d.ts +0 -30
- package/dist/runtime/internal-step-evidence.d.ts.map +0 -1
- package/dist/runtime/internal-step-evidence.js +0 -82
- package/dist/runtime/internal-step-evidence.js.map +0 -1
- package/dist/runtime/result-adapter.d.ts +0 -9
- package/dist/runtime/result-adapter.d.ts.map +0 -1
- package/dist/runtime/result-adapter.js +0 -77
- package/dist/runtime/result-adapter.js.map +0 -1
- package/dist/runtime/runner.d.ts +0 -43
- package/dist/runtime/runner.d.ts.map +0 -1
- package/dist/runtime/runner.js +0 -72
- package/dist/runtime/runner.js.map +0 -1
- package/dist/runtime/scenario-control.d.ts +0 -40
- package/dist/runtime/scenario-control.d.ts.map +0 -1
- package/dist/runtime/scenario-control.js +0 -94
- package/dist/runtime/scenario-control.js.map +0 -1
- package/dist/runtime/scenario-runner.d.ts +0 -41
- package/dist/runtime/scenario-runner.d.ts.map +0 -1
- package/dist/runtime/scenario-runner.js +0 -274
- package/dist/runtime/scenario-runner.js.map +0 -1
- package/dist/runtime/scenario-state.d.ts +0 -55
- package/dist/runtime/scenario-state.d.ts.map +0 -1
- package/dist/runtime/scenario-state.js +0 -128
- package/dist/runtime/scenario-state.js.map +0 -1
- package/dist/runtime/step-execution.d.ts +0 -27
- package/dist/runtime/step-execution.d.ts.map +0 -1
- package/dist/runtime/step-execution.js +0 -143
- package/dist/runtime/step-execution.js.map +0 -1
- package/dist/runtime/text.d.ts +0 -10
- package/dist/runtime/text.d.ts.map +0 -1
- package/dist/runtime/text.js +0 -25
- package/dist/runtime/text.js.map +0 -1
- package/dist/runtime/types.d.ts +0 -53
- package/dist/runtime/types.d.ts.map +0 -1
- package/dist/runtime/types.js +0 -2
- package/dist/runtime/types.js.map +0 -1
- package/dist/runtime/usage-tracker.d.ts +0 -37
- package/dist/runtime/usage-tracker.d.ts.map +0 -1
- package/dist/runtime/usage-tracker.js +0 -110
- package/dist/runtime/usage-tracker.js.map +0 -1
- package/dist/tools/define-agent-tool.d.ts +0 -20
- package/dist/tools/define-agent-tool.d.ts.map +0 -1
- package/dist/tools/define-agent-tool.js +0 -30
- package/dist/tools/define-agent-tool.js.map +0 -1
- package/dist/tools/dispatcher.d.ts +0 -19
- package/dist/tools/dispatcher.d.ts.map +0 -1
- package/dist/tools/dispatcher.js +0 -91
- package/dist/tools/dispatcher.js.map +0 -1
- package/dist/tools/loop-detector.d.ts +0 -27
- package/dist/tools/loop-detector.d.ts.map +0 -1
- package/dist/tools/loop-detector.js +0 -76
- package/dist/tools/loop-detector.js.map +0 -1
- package/dist/tools/registry.d.ts +0 -19
- package/dist/tools/registry.d.ts.map +0 -1
- package/dist/tools/registry.js +0 -54
- package/dist/tools/registry.js.map +0 -1
- package/dist/tools/step/result-tool.d.ts +0 -7
- package/dist/tools/step/result-tool.d.ts.map +0 -1
- package/dist/tools/step/result-tool.js +0 -34
- package/dist/tools/step/result-tool.js.map +0 -1
- package/dist/tools/tool-contract.d.ts +0 -3
- package/dist/tools/tool-contract.d.ts.map +0 -1
- package/dist/tools/tool-contract.js +0 -2
- package/dist/tools/tool-contract.js.map +0 -1
- package/dist/tools/types.d.ts +0 -210
- package/dist/tools/types.d.ts.map +0 -1
- package/dist/tools/types.js +0 -4
- package/dist/tools/types.js.map +0 -1
- package/docs/CLI.md +0 -59
- package/docs/CONFIGURATION.md +0 -105
- package/docs/DRIVERS.md +0 -79
- package/docs/EVIDENCE.md +0 -64
- package/schemas/checkmate-config.v1.json +0 -176
- package/schemas/describe-result.v1.json +0 -414
- package/schemas/driver-descriptor.v1.json +0 -66
- package/schemas/run-request.v1.json +0 -67
- package/schemas/run-result.v1.json +0 -333
- package/schemas/validation-result.v1.json +0 -86
|
@@ -1,29 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* The single place that decides what counts as a secret.
|
|
3
|
-
*
|
|
4
|
-
* `StepEvidence` calls this where the loop *records* evidence, so a credential the model typed
|
|
5
|
-
* into the page never reaches `StepReport` in the first place. The logging path (`ToolDispatcher`,
|
|
6
|
-
* `ToolResponseHandler`) consumes the same function instead of keeping its own copy of the rules,
|
|
7
|
-
* so there is exactly one definition of "looks like a secret" in the codebase.
|
|
8
|
-
*
|
|
9
|
-
* @example
|
|
10
|
-
* ```ts
|
|
11
|
-
* scrub('Authorization: Bearer sk-abc123') // 'Authorization: [secret omitted]'
|
|
12
|
-
* ```
|
|
13
|
-
*/
|
|
14
|
-
export declare function scrub(value: string): string;
|
|
15
|
-
/**
|
|
16
|
-
* Recursively applies {@link scrub} to every string value in an arbitrary JSON-like value,
|
|
17
|
-
* preserving its shape.
|
|
18
|
-
*
|
|
19
|
-
* Tool call arguments are kept as a structured object rather than a serialized string, so
|
|
20
|
-
* scrubbing them string-by-string is what keeps the report's `toolCalls` array both readable
|
|
21
|
-
* and free of secrets.
|
|
22
|
-
*
|
|
23
|
-
* @example
|
|
24
|
-
* ```ts
|
|
25
|
-
* scrubValue({ ref: 'e17', text: 'sk-abc123' }) // { ref: 'e17', text: '[secret omitted]' }
|
|
26
|
-
* ```
|
|
27
|
-
*/
|
|
28
|
-
export declare function scrubValue<T>(value: T): T;
|
|
29
|
-
//# sourceMappingURL=scrub.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"scrub.d.ts","sourceRoot":"","sources":["../../src/redaction/scrub.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AACH,wBAAgB,KAAK,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,CAQ3C;AAED;;;;;;;;;;;;GAYG;AACH,wBAAgB,UAAU,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC,GAAG,CAAC,CAczC"}
|
package/dist/redaction/scrub.js
DELETED
|
@@ -1,54 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* The single place that decides what counts as a secret.
|
|
3
|
-
*
|
|
4
|
-
* `StepEvidence` calls this where the loop *records* evidence, so a credential the model typed
|
|
5
|
-
* into the page never reaches `StepReport` in the first place. The logging path (`ToolDispatcher`,
|
|
6
|
-
* `ToolResponseHandler`) consumes the same function instead of keeping its own copy of the rules,
|
|
7
|
-
* so there is exactly one definition of "looks like a secret" in the codebase.
|
|
8
|
-
*
|
|
9
|
-
* @example
|
|
10
|
-
* ```ts
|
|
11
|
-
* scrub('Authorization: Bearer sk-abc123') // 'Authorization: [secret omitted]'
|
|
12
|
-
* ```
|
|
13
|
-
*/
|
|
14
|
-
export function scrub(value) {
|
|
15
|
-
return scrubSecretFields(value)
|
|
16
|
-
.replace(/data:image\/[a-zA-Z0-9.+-]+;base64,[A-Za-z0-9+/=]+/g, '[image omitted]')
|
|
17
|
-
.replace(/[A-Za-z0-9+/]{200,}={0,2}/g, '[base64 omitted]')
|
|
18
|
-
.replace(/sk-[A-Za-z0-9_-]+/g, '[secret omitted]')
|
|
19
|
-
.replace(/Bearer\s+[A-Za-z0-9._~+/=-]+/gi, 'Bearer [secret omitted]')
|
|
20
|
-
.replace(/Authorization\s*[:=]\s*[^\s,;}]+/gi, '[secret omitted]')
|
|
21
|
-
.replace(/Cookie\s*[:=]\s*[^\n,;}]+/gi, '[secret omitted]');
|
|
22
|
-
}
|
|
23
|
-
/**
|
|
24
|
-
* Recursively applies {@link scrub} to every string value in an arbitrary JSON-like value,
|
|
25
|
-
* preserving its shape.
|
|
26
|
-
*
|
|
27
|
-
* Tool call arguments are kept as a structured object rather than a serialized string, so
|
|
28
|
-
* scrubbing them string-by-string is what keeps the report's `toolCalls` array both readable
|
|
29
|
-
* and free of secrets.
|
|
30
|
-
*
|
|
31
|
-
* @example
|
|
32
|
-
* ```ts
|
|
33
|
-
* scrubValue({ ref: 'e17', text: 'sk-abc123' }) // { ref: 'e17', text: '[secret omitted]' }
|
|
34
|
-
* ```
|
|
35
|
-
*/
|
|
36
|
-
export function scrubValue(value) {
|
|
37
|
-
if (typeof value === 'string') {
|
|
38
|
-
return scrub(value);
|
|
39
|
-
}
|
|
40
|
-
if (Array.isArray(value)) {
|
|
41
|
-
return value.map((item) => scrubValue(item));
|
|
42
|
-
}
|
|
43
|
-
if (value && typeof value === 'object') {
|
|
44
|
-
return Object.fromEntries(Object.entries(value).map(([key, entry]) => [key, scrubValue(entry)]));
|
|
45
|
-
}
|
|
46
|
-
return value;
|
|
47
|
-
}
|
|
48
|
-
function scrubSecretFields(value) {
|
|
49
|
-
const secretKey = String.raw `(?:CHECKMATE_OPENAI_API_KEY|OPENAI_API_KEY|api[_-]?key|apikey|authorization|cookie)`;
|
|
50
|
-
return value
|
|
51
|
-
.replace(new RegExp(String.raw `(["'])${secretKey}\1\s*:\s*(["'])(?:\\.|(?!\2).)*\2\s*,?`, 'gi'), '[secret omitted]')
|
|
52
|
-
.replace(new RegExp(String.raw `\b${secretKey}\b\s*[:=]\s*(?:Bearer\s+[^\s,;}]+|"[^"]*"|'[^']*'|[^\s,;}]+)`, 'gi'), '[secret omitted]');
|
|
53
|
-
}
|
|
54
|
-
//# sourceMappingURL=scrub.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"scrub.js","sourceRoot":"","sources":["../../src/redaction/scrub.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AACH,MAAM,UAAU,KAAK,CAAC,KAAa;IAClC,OAAO,iBAAiB,CAAC,KAAK,CAAC;SAC7B,OAAO,CAAC,qDAAqD,EAAE,iBAAiB,CAAC;SACjF,OAAO,CAAC,4BAA4B,EAAE,kBAAkB,CAAC;SACzD,OAAO,CAAC,oBAAoB,EAAE,kBAAkB,CAAC;SACjD,OAAO,CAAC,gCAAgC,EAAE,yBAAyB,CAAC;SACpE,OAAO,CAAC,oCAAoC,EAAE,kBAAkB,CAAC;SACjE,OAAO,CAAC,6BAA6B,EAAE,kBAAkB,CAAC,CAAA;AAC7D,CAAC;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,UAAU,UAAU,CAAI,KAAQ;IACrC,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;QAC/B,OAAO,KAAK,CAAC,KAAK,CAAM,CAAA;IACzB,CAAC;IAED,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC;QAC1B,OAAO,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,UAAU,CAAC,IAAI,CAAC,CAAM,CAAA;IAClD,CAAC;IAED,IAAI,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;QACxC,OAAO,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,UAAU,CAAC,KAAK,CAAC,CAAC,CAAC,CAAM,CAAA;IACtG,CAAC;IAED,OAAO,KAAK,CAAA;AACb,CAAC;AAED,SAAS,iBAAiB,CAAC,KAAa;IACvC,MAAM,SAAS,GAAG,MAAM,CAAC,GAAG,CAAA,qFAAqF,CAAA;IACjH,OAAO,KAAK;SACV,OAAO,CACP,IAAI,MAAM,CAAC,MAAM,CAAC,GAAG,CAAA,SAAS,SAAS,wCAAwC,EAAE,IAAI,CAAC,EACtF,kBAAkB,CAClB;SACA,OAAO,CACP,IAAI,MAAM,CAAC,MAAM,CAAC,GAAG,CAAA,KAAK,SAAS,8DAA8D,EAAE,IAAI,CAAC,EACxG,kBAAkB,CAClB,CAAA;AACH,CAAC"}
|
package/dist/runtime/config.d.ts
DELETED
|
@@ -1,16 +0,0 @@
|
|
|
1
|
-
import type { LogLevel } from '../logging/level-logger.js';
|
|
2
|
-
export type RuntimeConfig = Readonly<{
|
|
3
|
-
model: string;
|
|
4
|
-
baseUrl?: string;
|
|
5
|
-
reasoningEffort?: 'low' | 'medium' | 'high';
|
|
6
|
-
temperature: number;
|
|
7
|
-
turnCap: number;
|
|
8
|
-
requestTimeout: number;
|
|
9
|
-
maxRetries: number;
|
|
10
|
-
loopMaxRepetitions: number;
|
|
11
|
-
redact: boolean;
|
|
12
|
-
toolChoice: 'required' | 'auto';
|
|
13
|
-
rateLimitDelay: number;
|
|
14
|
-
logLevel: LogLevel;
|
|
15
|
-
}>;
|
|
16
|
-
//# sourceMappingURL=config.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"config.d.ts","sourceRoot":"","sources":["../../src/runtime/config.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,4BAA4B,CAAA;AAE1D,MAAM,MAAM,aAAa,GAAG,QAAQ,CAAC;IACpC,KAAK,EAAE,MAAM,CAAA;IACb,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,eAAe,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,CAAA;IAC3C,WAAW,EAAE,MAAM,CAAA;IACnB,OAAO,EAAE,MAAM,CAAA;IACf,cAAc,EAAE,MAAM,CAAA;IACtB,UAAU,EAAE,MAAM,CAAA;IAClB,kBAAkB,EAAE,MAAM,CAAA;IAC1B,MAAM,EAAE,OAAO,CAAA;IACf,UAAU,EAAE,UAAU,GAAG,MAAM,CAAA;IAC/B,cAAc,EAAE,MAAM,CAAA;IACtB,QAAQ,EAAE,QAAQ,CAAA;CAClB,CAAC,CAAA"}
|
package/dist/runtime/config.js
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"config.js","sourceRoot":"","sources":["../../src/runtime/config.ts"],"names":[],"mappings":""}
|
|
@@ -1,16 +0,0 @@
|
|
|
1
|
-
import type { ControlExpirationReason, StepControl } from './scenario-control.js';
|
|
2
|
-
export declare class DriverBoundaryError extends Error {
|
|
3
|
-
readonly operation: string;
|
|
4
|
-
readonly reason: ControlExpirationReason;
|
|
5
|
-
constructor(operation: string, reason: ControlExpirationReason);
|
|
6
|
-
}
|
|
7
|
-
export declare function awaitDriverBoundary<T>(input: {
|
|
8
|
-
operation: string;
|
|
9
|
-
signal: AbortSignal;
|
|
10
|
-
deadline: number;
|
|
11
|
-
call: (signal: AbortSignal) => Promise<T> | T;
|
|
12
|
-
now?: () => number;
|
|
13
|
-
reason?: () => ControlExpirationReason;
|
|
14
|
-
}): Promise<T>;
|
|
15
|
-
export declare function awaitStepDriverBoundary<T>(operation: string, control: StepControl, call: (signal: AbortSignal) => Promise<T> | T): Promise<T>;
|
|
16
|
-
//# sourceMappingURL=driver-boundary.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"driver-boundary.d.ts","sourceRoot":"","sources":["../../src/runtime/driver-boundary.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,uBAAuB,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAA;AAEjF,qBAAa,mBAAoB,SAAQ,KAAK;IAE5C,QAAQ,CAAC,SAAS,EAAE,MAAM;IAC1B,QAAQ,CAAC,MAAM,EAAE,uBAAuB;gBAD/B,SAAS,EAAE,MAAM,EACjB,MAAM,EAAE,uBAAuB;CAKzC;AAED,wBAAsB,mBAAmB,CAAC,CAAC,EAAE,KAAK,EAAE;IACnD,SAAS,EAAE,MAAM,CAAA;IACjB,MAAM,EAAE,WAAW,CAAA;IACnB,QAAQ,EAAE,MAAM,CAAA;IAChB,IAAI,EAAE,CAAC,MAAM,EAAE,WAAW,KAAK,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAA;IAC7C,GAAG,CAAC,EAAE,MAAM,MAAM,CAAA;IAClB,MAAM,CAAC,EAAE,MAAM,uBAAuB,CAAA;CACtC,GAAG,OAAO,CAAC,CAAC,CAAC,CAyCb;AAED,wBAAgB,uBAAuB,CAAC,CAAC,EACxC,SAAS,EAAE,MAAM,EACjB,OAAO,EAAE,WAAW,EACpB,IAAI,EAAE,CAAC,MAAM,EAAE,WAAW,KAAK,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,GAC3C,OAAO,CAAC,CAAC,CAAC,CAYZ"}
|
|
@@ -1,71 +0,0 @@
|
|
|
1
|
-
export class DriverBoundaryError extends Error {
|
|
2
|
-
operation;
|
|
3
|
-
reason;
|
|
4
|
-
constructor(operation, reason) {
|
|
5
|
-
super(`Driver boundary '${operation}' expired: ${reason}`);
|
|
6
|
-
this.operation = operation;
|
|
7
|
-
this.reason = reason;
|
|
8
|
-
this.name = 'DriverBoundaryError';
|
|
9
|
-
}
|
|
10
|
-
}
|
|
11
|
-
export async function awaitDriverBoundary(input) {
|
|
12
|
-
const now = input.now ?? Date.now;
|
|
13
|
-
if (input.signal.aborted || now() >= input.deadline) {
|
|
14
|
-
throw new DriverBoundaryError(input.operation, input.reason?.() ?? signalReason(input.signal));
|
|
15
|
-
}
|
|
16
|
-
const controller = new AbortController();
|
|
17
|
-
const reason = () => input.reason?.() ?? signalReason(input.signal);
|
|
18
|
-
const timeout = Math.max(0, input.deadline - now());
|
|
19
|
-
let timer;
|
|
20
|
-
let onAbort;
|
|
21
|
-
const boundaryPromise = new Promise((_, reject) => {
|
|
22
|
-
onAbort = () => {
|
|
23
|
-
const expiration = reason();
|
|
24
|
-
controller.abort(expiration);
|
|
25
|
-
reject(new DriverBoundaryError(input.operation, expiration));
|
|
26
|
-
};
|
|
27
|
-
if (input.signal.aborted)
|
|
28
|
-
queueMicrotask(onAbort);
|
|
29
|
-
else
|
|
30
|
-
input.signal.addEventListener('abort', onAbort, { once: true });
|
|
31
|
-
timer = setTimeout(onAbort, timeout);
|
|
32
|
-
});
|
|
33
|
-
let callPromise;
|
|
34
|
-
try {
|
|
35
|
-
callPromise = Promise.resolve(input.call(controller.signal));
|
|
36
|
-
}
|
|
37
|
-
catch (error) {
|
|
38
|
-
callPromise = Promise.reject(error);
|
|
39
|
-
}
|
|
40
|
-
void callPromise.catch(() => undefined);
|
|
41
|
-
try {
|
|
42
|
-
const value = await Promise.race([callPromise, boundaryPromise]);
|
|
43
|
-
if (input.signal.aborted || now() >= input.deadline) {
|
|
44
|
-
const expiration = reason();
|
|
45
|
-
controller.abort(expiration);
|
|
46
|
-
throw new DriverBoundaryError(input.operation, expiration);
|
|
47
|
-
}
|
|
48
|
-
return value;
|
|
49
|
-
}
|
|
50
|
-
finally {
|
|
51
|
-
clearTimeout(timer);
|
|
52
|
-
input.signal.removeEventListener('abort', onAbort);
|
|
53
|
-
}
|
|
54
|
-
}
|
|
55
|
-
export function awaitStepDriverBoundary(operation, control, call) {
|
|
56
|
-
return awaitDriverBoundary({
|
|
57
|
-
operation,
|
|
58
|
-
signal: control.signal,
|
|
59
|
-
deadline: control.deadline,
|
|
60
|
-
now: control.now,
|
|
61
|
-
reason: () => {
|
|
62
|
-
const state = control.poll();
|
|
63
|
-
return state.expired ? state.reason : 'step-timeout';
|
|
64
|
-
},
|
|
65
|
-
call,
|
|
66
|
-
});
|
|
67
|
-
}
|
|
68
|
-
function signalReason(signal) {
|
|
69
|
-
return signal.reason === 'interrupted' || signal.reason === 'scenario-timeout' ? signal.reason : 'step-timeout';
|
|
70
|
-
}
|
|
71
|
-
//# sourceMappingURL=driver-boundary.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"driver-boundary.js","sourceRoot":"","sources":["../../src/runtime/driver-boundary.ts"],"names":[],"mappings":"AAEA,MAAM,OAAO,mBAAoB,SAAQ,KAAK;IAEnC;IACA;IAFV,YACU,SAAiB,EACjB,MAA+B;QAExC,KAAK,CAAC,oBAAoB,SAAS,cAAc,MAAM,EAAE,CAAC,CAAA;QAHjD,cAAS,GAAT,SAAS,CAAQ;QACjB,WAAM,GAAN,MAAM,CAAyB;QAGxC,IAAI,CAAC,IAAI,GAAG,qBAAqB,CAAA;IAClC,CAAC;CACD;AAED,MAAM,CAAC,KAAK,UAAU,mBAAmB,CAAI,KAO5C;IACA,MAAM,GAAG,GAAG,KAAK,CAAC,GAAG,IAAI,IAAI,CAAC,GAAG,CAAA;IACjC,IAAI,KAAK,CAAC,MAAM,CAAC,OAAO,IAAI,GAAG,EAAE,IAAI,KAAK,CAAC,QAAQ,EAAE,CAAC;QACrD,MAAM,IAAI,mBAAmB,CAAC,KAAK,CAAC,SAAS,EAAE,KAAK,CAAC,MAAM,EAAE,EAAE,IAAI,YAAY,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAA;IAC/F,CAAC;IACD,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAA;IACxC,MAAM,MAAM,GAAG,GAAG,EAAE,CAAC,KAAK,CAAC,MAAM,EAAE,EAAE,IAAI,YAAY,CAAC,KAAK,CAAC,MAAM,CAAC,CAAA;IACnE,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,CAAC,QAAQ,GAAG,GAAG,EAAE,CAAC,CAAA;IACnD,IAAI,KAAoC,CAAA;IACxC,IAAI,OAAmB,CAAA;IACvB,MAAM,eAAe,GAAG,IAAI,OAAO,CAAQ,CAAC,CAAC,EAAE,MAAM,EAAE,EAAE;QACxD,OAAO,GAAG,GAAG,EAAE;YACd,MAAM,UAAU,GAAG,MAAM,EAAE,CAAA;YAC3B,UAAU,CAAC,KAAK,CAAC,UAAU,CAAC,CAAA;YAC5B,MAAM,CAAC,IAAI,mBAAmB,CAAC,KAAK,CAAC,SAAS,EAAE,UAAU,CAAC,CAAC,CAAA;QAC7D,CAAC,CAAA;QACD,IAAI,KAAK,CAAC,MAAM,CAAC,OAAO;YAAE,cAAc,CAAC,OAAO,CAAC,CAAA;;YAC5C,KAAK,CAAC,MAAM,CAAC,gBAAgB,CAAC,OAAO,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAA;QACpE,KAAK,GAAG,UAAU,CAAC,OAAO,EAAE,OAAO,CAAC,CAAA;IACrC,CAAC,CAAC,CAAA;IAEF,IAAI,WAAuB,CAAA;IAC3B,IAAI,CAAC;QACJ,WAAW,GAAG,OAAO,CAAC,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,CAAA;IAC7D,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QAChB,WAAW,GAAG,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,CAAA;IACpC,CAAC;IACD,KAAK,WAAW,CAAC,KAAK,CAAC,GAAS,EAAE,CAAC,SAAS,CAAC,CAAA;IAE7C,IAAI,CAAC;QACJ,MAAM,KAAK,GAAG,MAAM,OAAO,CAAC,IAAI,CAAC,CAAC,WAAW,EAAE,eAAe,CAAC,CAAC,CAAA;QAChE,IAAI,KAAK,CAAC,MAAM,CAAC,OAAO,IAAI,GAAG,EAAE,IAAI,KAAK,CAAC,QAAQ,EAAE,CAAC;YACrD,MAAM,UAAU,GAAG,MAAM,EAAE,CAAA;YAC3B,UAAU,CAAC,KAAK,CAAC,UAAU,CAAC,CAAA;YAC5B,MAAM,IAAI,mBAAmB,CAAC,KAAK,CAAC,SAAS,EAAE,UAAU,CAAC,CAAA;QAC3D,CAAC;QACD,OAAO,KAAK,CAAA;IACb,CAAC;YAAS,CAAC;QACV,YAAY,CAAC,KAAM,CAAC,CAAA;QACpB,KAAK,CAAC,MAAM,CAAC,mBAAmB,CAAC,OAAO,EAAE,OAAQ,CAAC,CAAA;IACpD,CAAC;AACF,CAAC;AAED,MAAM,UAAU,uBAAuB,CACtC,SAAiB,EACjB,OAAoB,EACpB,IAA6C;IAE7C,OAAO,mBAAmB,CAAC;QAC1B,SAAS;QACT,MAAM,EAAE,OAAO,CAAC,MAAM;QACtB,QAAQ,EAAE,OAAO,CAAC,QAAQ;QAC1B,GAAG,EAAE,OAAO,CAAC,GAAG;QAChB,MAAM,EAAE,GAAG,EAAE;YACZ,MAAM,KAAK,GAAG,OAAO,CAAC,IAAI,EAAE,CAAA;YAC5B,OAAO,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,cAAc,CAAA;QACrD,CAAC;QACD,IAAI;KACJ,CAAC,CAAA;AACH,CAAC;AAED,SAAS,YAAY,CAAC,MAAmB;IACxC,OAAO,MAAM,CAAC,MAAM,KAAK,aAAa,IAAI,MAAM,CAAC,MAAM,KAAK,kBAAkB,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,cAAc,CAAA;AAChH,CAAC"}
|
|
@@ -1,9 +0,0 @@
|
|
|
1
|
-
import type { ChatCompletionMessageParam } from 'openai/resources/chat/completions';
|
|
2
|
-
import type { DriverSession, DriverTool, StepIntent } from '../driver.js';
|
|
3
|
-
import type { AgentTool, ToolExecution } from '../tools/types.js';
|
|
4
|
-
import type { StepControl } from './scenario-control.js';
|
|
5
|
-
export declare function adaptDriverTool(driverId: string, tool: DriverTool): AgentTool;
|
|
6
|
-
export declare function buildInitialDriverContext(session: DriverSession, step: StepIntent, control: StepControl): Promise<ChatCompletionMessageParam[]>;
|
|
7
|
-
export declare function buildPostToolDriverContext(session: DriverSession, step: StepIntent, turn: number, toolResponses: readonly ToolExecution[], control: StepControl): Promise<ChatCompletionMessageParam[]>;
|
|
8
|
-
export declare function isEphemeralDriverMessage(message: ChatCompletionMessageParam): boolean;
|
|
9
|
-
//# sourceMappingURL=driver-session.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"driver-session.d.ts","sourceRoot":"","sources":["../../src/runtime/driver-session.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,0BAA0B,EAAE,MAAM,mCAAmC,CAAA;AACnF,OAAO,KAAK,EAAwB,aAAa,EAAE,UAAU,EAAuB,UAAU,EAAE,MAAM,cAAc,CAAA;AACpH,OAAO,KAAK,EAAE,SAAS,EAAE,aAAa,EAAE,MAAM,mBAAmB,CAAA;AAEjE,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAA;AAExD,wBAAgB,eAAe,CAAC,QAAQ,EAAE,MAAM,EAAE,IAAI,EAAE,UAAU,GAAG,SAAS,CAiB7E;AAED,wBAAsB,yBAAyB,CAC9C,OAAO,EAAE,aAAa,EACtB,IAAI,EAAE,UAAU,EAChB,OAAO,EAAE,WAAW,GAClB,OAAO,CAAC,0BAA0B,EAAE,CAAC,CAKvC;AAED,wBAAsB,0BAA0B,CAC/C,OAAO,EAAE,aAAa,EACtB,IAAI,EAAE,UAAU,EAChB,IAAI,EAAE,MAAM,EACZ,aAAa,EAAE,SAAS,aAAa,EAAE,EACvC,OAAO,EAAE,WAAW,GAClB,OAAO,CAAC,0BAA0B,EAAE,CAAC,CAUvC;AAED,wBAAgB,wBAAwB,CAAC,OAAO,EAAE,0BAA0B,GAAG,OAAO,CAErF"}
|
|
@@ -1,64 +0,0 @@
|
|
|
1
|
-
import { awaitStepDriverBoundary } from './driver-boundary.js';
|
|
2
|
-
export function adaptDriverTool(driverId, tool) {
|
|
3
|
-
return {
|
|
4
|
-
driverId,
|
|
5
|
-
definition: { ...tool.definition, parameters: structuredClone(tool.definition.parameters) },
|
|
6
|
-
execute: async (args, context) => {
|
|
7
|
-
if (!context.control || context.turn === undefined || !context.signal) {
|
|
8
|
-
throw new Error(`Driver tool '${tool.definition.name}' requires an active step control`);
|
|
9
|
-
}
|
|
10
|
-
return awaitStepDriverBoundary(`tool:${tool.definition.name}`, context.control, (signal) => tool.execute(args, {
|
|
11
|
-
step: toStepIntent(context.step),
|
|
12
|
-
turn: context.turn,
|
|
13
|
-
signal,
|
|
14
|
-
}));
|
|
15
|
-
},
|
|
16
|
-
};
|
|
17
|
-
}
|
|
18
|
-
export async function buildInitialDriverContext(session, step, control) {
|
|
19
|
-
const context = await awaitStepDriverBoundary('build-initial-context', control, (signal) => session.buildInitialContext({ step, signal }));
|
|
20
|
-
return context.map(toProviderMessage);
|
|
21
|
-
}
|
|
22
|
-
export async function buildPostToolDriverContext(session, step, turn, toolResponses, control) {
|
|
23
|
-
const context = await awaitStepDriverBoundary('handle-tool-responses', control, (signal) => session.handleToolResponses({
|
|
24
|
-
step,
|
|
25
|
-
turn,
|
|
26
|
-
toolResponses: toolResponses.map(toDriverToolExecution),
|
|
27
|
-
signal,
|
|
28
|
-
}));
|
|
29
|
-
return context.map(toProviderMessage);
|
|
30
|
-
}
|
|
31
|
-
export function isEphemeralDriverMessage(message) {
|
|
32
|
-
return Boolean(message.__checkmateEphemeral);
|
|
33
|
-
}
|
|
34
|
-
function toProviderMessage(context) {
|
|
35
|
-
const content = typeof context.content === 'string'
|
|
36
|
-
? context.content
|
|
37
|
-
: context.content.map((part) => part.type === 'text'
|
|
38
|
-
? { type: 'text', text: part.text }
|
|
39
|
-
: {
|
|
40
|
-
type: 'image_url',
|
|
41
|
-
image_url: {
|
|
42
|
-
url: `data:${part.mediaType};base64,${part.data}`,
|
|
43
|
-
detail: 'high',
|
|
44
|
-
},
|
|
45
|
-
});
|
|
46
|
-
const message = { role: 'user', content };
|
|
47
|
-
if (context.ephemeral) {
|
|
48
|
-
Object.defineProperty(message, '__checkmateEphemeral', { value: true, enumerable: false });
|
|
49
|
-
}
|
|
50
|
-
return message;
|
|
51
|
-
}
|
|
52
|
-
function toDriverToolExecution(execution) {
|
|
53
|
-
return {
|
|
54
|
-
toolCallId: execution.toolCallId,
|
|
55
|
-
name: execution.toolCall.name,
|
|
56
|
-
arguments: structuredClone(execution.toolCall.arguments ?? {}),
|
|
57
|
-
response: execution.toolResponse.response,
|
|
58
|
-
status: execution.toolResponse.status,
|
|
59
|
-
};
|
|
60
|
-
}
|
|
61
|
-
function toStepIntent(step) {
|
|
62
|
-
return { id: step.id ?? 'legacy-step', action: step.action, expect: step.expect };
|
|
63
|
-
}
|
|
64
|
-
//# sourceMappingURL=driver-session.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"driver-session.js","sourceRoot":"","sources":["../../src/runtime/driver-session.ts"],"names":[],"mappings":"AAGA,OAAO,EAAE,uBAAuB,EAAE,MAAM,sBAAsB,CAAA;AAG9D,MAAM,UAAU,eAAe,CAAC,QAAgB,EAAE,IAAgB;IACjE,OAAO;QACN,QAAQ;QACR,UAAU,EAAE,EAAE,GAAG,IAAI,CAAC,UAAU,EAAE,UAAU,EAAE,eAAe,CAAC,IAAI,CAAC,UAAU,CAAC,UAAU,CAAC,EAAE;QAC3F,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,OAAO,EAAE,EAAE;YAChC,IAAI,CAAC,OAAO,CAAC,OAAO,IAAI,OAAO,CAAC,IAAI,KAAK,SAAS,IAAI,CAAC,OAAO,CAAC,MAAM,EAAE,CAAC;gBACvE,MAAM,IAAI,KAAK,CAAC,gBAAgB,IAAI,CAAC,UAAU,CAAC,IAAI,mCAAmC,CAAC,CAAA;YACzF,CAAC;YACD,OAAO,uBAAuB,CAAC,QAAQ,IAAI,CAAC,UAAU,CAAC,IAAI,EAAE,EAAE,OAAO,CAAC,OAAO,EAAE,CAAC,MAAM,EAAE,EAAE,CAC1F,IAAI,CAAC,OAAO,CAAC,IAAI,EAAE;gBAClB,IAAI,EAAE,YAAY,CAAC,OAAO,CAAC,IAAI,CAAC;gBAChC,IAAI,EAAE,OAAO,CAAC,IAAK;gBACnB,MAAM;aACN,CAAC,CACF,CAAA;QACF,CAAC;KACD,CAAA;AACF,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAC9C,OAAsB,EACtB,IAAgB,EAChB,OAAoB;IAEpB,MAAM,OAAO,GAAG,MAAM,uBAAuB,CAAC,uBAAuB,EAAE,OAAO,EAAE,CAAC,MAAM,EAAE,EAAE,CAC1F,OAAO,CAAC,mBAAmB,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC,CAC7C,CAAA;IACD,OAAO,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC,CAAA;AACtC,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,0BAA0B,CAC/C,OAAsB,EACtB,IAAgB,EAChB,IAAY,EACZ,aAAuC,EACvC,OAAoB;IAEpB,MAAM,OAAO,GAAG,MAAM,uBAAuB,CAAC,uBAAuB,EAAE,OAAO,EAAE,CAAC,MAAM,EAAE,EAAE,CAC1F,OAAO,CAAC,mBAAmB,CAAC;QAC3B,IAAI;QACJ,IAAI;QACJ,aAAa,EAAE,aAAa,CAAC,GAAG,CAAC,qBAAqB,CAAC;QACvD,MAAM;KACN,CAAC,CACF,CAAA;IACD,OAAO,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC,CAAA;AACtC,CAAC;AAED,MAAM,UAAU,wBAAwB,CAAC,OAAmC;IAC3E,OAAO,OAAO,CAAE,OAA2E,CAAC,oBAAoB,CAAC,CAAA;AAClH,CAAC;AAED,SAAS,iBAAiB,CAAC,OAA6B;IACvD,MAAM,OAAO,GACZ,OAAO,OAAO,CAAC,OAAO,KAAK,QAAQ;QAClC,CAAC,CAAC,OAAO,CAAC,OAAO;QACjB,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAC7B,IAAI,CAAC,IAAI,KAAK,MAAM;YACnB,CAAC,CAAC,EAAE,IAAI,EAAE,MAAe,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE;YAC5C,CAAC,CAAC;gBACA,IAAI,EAAE,WAAoB;gBAC1B,SAAS,EAAE;oBACV,GAAG,EAAE,QAAQ,IAAI,CAAC,SAAS,WAAW,IAAI,CAAC,IAAI,EAAE;oBACjD,MAAM,EAAE,MAAe;iBACvB;aACD,CACH,CAAA;IACJ,MAAM,OAAO,GAAG,EAAE,IAAI,EAAE,MAAe,EAAE,OAAO,EAE/C,CAAA;IACD,IAAI,OAAO,CAAC,SAAS,EAAE,CAAC;QACvB,MAAM,CAAC,cAAc,CAAC,OAAO,EAAE,sBAAsB,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,UAAU,EAAE,KAAK,EAAE,CAAC,CAAA;IAC3F,CAAC;IACD,OAAO,OAAO,CAAA;AACf,CAAC;AAED,SAAS,qBAAqB,CAAC,SAAwB;IACtD,OAAO;QACN,UAAU,EAAE,SAAS,CAAC,UAAU;QAChC,IAAI,EAAE,SAAS,CAAC,QAAQ,CAAC,IAAI;QAC7B,SAAS,EAAE,eAAe,CAAC,SAAS,CAAC,QAAQ,CAAC,SAAS,IAAI,EAAE,CAAC;QAC9D,QAAQ,EAAE,SAAS,CAAC,YAAY,CAAC,QAAQ;QACzC,MAAM,EAAE,SAAS,CAAC,YAAY,CAAC,MAAM;KACrC,CAAA;AACF,CAAC;AAED,SAAS,YAAY,CAAC,IAAqD;IAC1E,OAAO,EAAE,EAAE,EAAE,IAAI,CAAC,EAAE,IAAI,aAAa,EAAE,MAAM,EAAE,IAAI,CAAC,MAAM,EAAE,MAAM,EAAE,IAAI,CAAC,MAAM,EAAE,CAAA;AAClF,CAAC"}
|
|
@@ -1,30 +0,0 @@
|
|
|
1
|
-
import type { Diagnostic, Usage } from '../contracts/types.js';
|
|
2
|
-
import type { StepIntent } from '../driver.js';
|
|
3
|
-
import { DiagnosticSanitizer } from '../redaction/diagnostic-sanitizer.js';
|
|
4
|
-
import type { ToolCall, ToolResponse } from '../tools/types.js';
|
|
5
|
-
import type { InternalStepReport, InternalTerminationReason } from './types.js';
|
|
6
|
-
export type InternalStepTermination = Readonly<{
|
|
7
|
-
outcome: 'passed' | 'failed';
|
|
8
|
-
reason: InternalTerminationReason;
|
|
9
|
-
actual?: string;
|
|
10
|
-
turns: number;
|
|
11
|
-
}>;
|
|
12
|
-
export declare class InternalStepEvidence {
|
|
13
|
-
private readonly step;
|
|
14
|
-
private readonly driverId;
|
|
15
|
-
private readonly redact;
|
|
16
|
-
private readonly diagnosticSanitizer;
|
|
17
|
-
private readonly now;
|
|
18
|
-
private readonly startedAt;
|
|
19
|
-
private readonly toolCalls;
|
|
20
|
-
private readonly transcript;
|
|
21
|
-
private readonly diagnostics;
|
|
22
|
-
constructor(step: StepIntent, driverId: string, redact: boolean, diagnosticSanitizer: DiagnosticSanitizer, now?: () => number);
|
|
23
|
-
recordAssistantMessage(turn: number, content: string): void;
|
|
24
|
-
recordToolCall(turn: number, toolCall: ToolCall, toolResponse: ToolResponse): void;
|
|
25
|
-
recordDiagnostic(diagnostic: Diagnostic): void;
|
|
26
|
-
buildReport(termination: InternalStepTermination, usage: Usage): InternalStepReport;
|
|
27
|
-
private sanitize;
|
|
28
|
-
private sanitizeArguments;
|
|
29
|
-
}
|
|
30
|
-
//# sourceMappingURL=internal-step-evidence.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"internal-step-evidence.d.ts","sourceRoot":"","sources":["../../src/runtime/internal-step-evidence.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,KAAK,EAAE,MAAM,uBAAuB,CAAA;AAC9D,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,cAAc,CAAA;AAC9C,OAAO,EAAE,mBAAmB,EAAE,MAAM,sCAAsC,CAAA;AAE1E,OAAO,KAAK,EAAE,QAAQ,EAAE,YAAY,EAAE,MAAM,mBAAmB,CAAA;AAC/D,OAAO,KAAK,EACX,kBAAkB,EAElB,yBAAyB,EAGzB,MAAM,YAAY,CAAA;AAkBnB,MAAM,MAAM,uBAAuB,GAAG,QAAQ,CAAC;IAC9C,OAAO,EAAE,QAAQ,GAAG,QAAQ,CAAA;IAC5B,MAAM,EAAE,yBAAyB,CAAA;IACjC,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,KAAK,EAAE,MAAM,CAAA;CACb,CAAC,CAAA;AAEF,qBAAa,oBAAoB;IAO/B,OAAO,CAAC,QAAQ,CAAC,IAAI;IACrB,OAAO,CAAC,QAAQ,CAAC,QAAQ;IACzB,OAAO,CAAC,QAAQ,CAAC,MAAM;IACvB,OAAO,CAAC,QAAQ,CAAC,mBAAmB;IACpC,OAAO,CAAC,QAAQ,CAAC,GAAG;IAVrB,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAQ;IAClC,OAAO,CAAC,QAAQ,CAAC,SAAS,CAA6B;IACvD,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAwB;IACnD,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAmB;gBAG7B,IAAI,EAAE,UAAU,EAChB,QAAQ,EAAE,MAAM,EAChB,MAAM,EAAE,OAAO,EACf,mBAAmB,EAAE,mBAAmB,EACxC,GAAG,GAAE,MAAM,MAAiB;IAK9C,sBAAsB,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,IAAI;IAI3D,cAAc,CAAC,IAAI,EAAE,MAAM,EAAE,QAAQ,EAAE,QAAQ,EAAE,YAAY,EAAE,YAAY,GAAG,IAAI;IAelF,gBAAgB,CAAC,UAAU,EAAE,UAAU,GAAG,IAAI;IAI9C,WAAW,CAAC,WAAW,EAAE,uBAAuB,EAAE,KAAK,EAAE,KAAK,GAAG,kBAAkB;IAgBnF,OAAO,CAAC,QAAQ;IAIhB,OAAO,CAAC,iBAAiB;CAGzB"}
|
|
@@ -1,82 +0,0 @@
|
|
|
1
|
-
import { scrub, scrubValue } from '../redaction/scrub.js';
|
|
2
|
-
const TRANSCRIPT_CONTENT_LIMIT = 2_000;
|
|
3
|
-
const CATEGORY_BY_REASON = {
|
|
4
|
-
'met-expectation': 'app',
|
|
5
|
-
'failed-expectation': 'app',
|
|
6
|
-
'loop-detected': 'model',
|
|
7
|
-
'turn-cap-exceeded': 'model',
|
|
8
|
-
'step-timeout': 'model',
|
|
9
|
-
'scenario-timeout': 'infra',
|
|
10
|
-
'tool-error': 'infra',
|
|
11
|
-
'provider-error': 'infra',
|
|
12
|
-
'token-budget-exceeded': 'infra',
|
|
13
|
-
interrupted: 'infra',
|
|
14
|
-
'internal-error': 'infra',
|
|
15
|
-
};
|
|
16
|
-
export class InternalStepEvidence {
|
|
17
|
-
step;
|
|
18
|
-
driverId;
|
|
19
|
-
redact;
|
|
20
|
-
diagnosticSanitizer;
|
|
21
|
-
now;
|
|
22
|
-
startedAt;
|
|
23
|
-
toolCalls = [];
|
|
24
|
-
transcript = [];
|
|
25
|
-
diagnostics = [];
|
|
26
|
-
constructor(step, driverId, redact, diagnosticSanitizer, now = Date.now) {
|
|
27
|
-
this.step = step;
|
|
28
|
-
this.driverId = driverId;
|
|
29
|
-
this.redact = redact;
|
|
30
|
-
this.diagnosticSanitizer = diagnosticSanitizer;
|
|
31
|
-
this.now = now;
|
|
32
|
-
this.startedAt = now();
|
|
33
|
-
}
|
|
34
|
-
recordAssistantMessage(turn, content) {
|
|
35
|
-
this.transcript.push({ turn, role: 'assistant', content: truncate(this.sanitize(content)) });
|
|
36
|
-
}
|
|
37
|
-
recordToolCall(turn, toolCall, toolResponse) {
|
|
38
|
-
this.toolCalls.push({
|
|
39
|
-
turn,
|
|
40
|
-
driverId: isHarnessTool(toolCall.name) ? 'harness' : this.driverId,
|
|
41
|
-
name: toolCall.name,
|
|
42
|
-
arguments: this.sanitizeArguments(toolCall.arguments ?? {}),
|
|
43
|
-
status: toolResponse.status === 'error' ? 'error' : 'ok',
|
|
44
|
-
});
|
|
45
|
-
this.transcript.push({
|
|
46
|
-
turn,
|
|
47
|
-
role: 'tool',
|
|
48
|
-
content: truncate(this.sanitize(`${toolCall.name} -> ${toolResponse.response}`)),
|
|
49
|
-
});
|
|
50
|
-
}
|
|
51
|
-
recordDiagnostic(diagnostic) {
|
|
52
|
-
this.diagnostics.push({ ...diagnostic, message: this.diagnosticSanitizer.text(diagnostic.message) });
|
|
53
|
-
}
|
|
54
|
-
buildReport(termination, usage) {
|
|
55
|
-
return {
|
|
56
|
-
step: { ...this.step },
|
|
57
|
-
outcome: termination.outcome,
|
|
58
|
-
category: CATEGORY_BY_REASON[termination.reason],
|
|
59
|
-
reason: termination.reason,
|
|
60
|
-
...(termination.actual === undefined ? {} : { actual: this.sanitize(termination.actual) }),
|
|
61
|
-
turns: termination.turns,
|
|
62
|
-
durationMs: this.now() - this.startedAt,
|
|
63
|
-
usage: { ...usage },
|
|
64
|
-
toolCalls: this.toolCalls.map((call) => ({ ...call })),
|
|
65
|
-
transcript: this.transcript.map((entry) => ({ ...entry })),
|
|
66
|
-
diagnostics: this.diagnostics.map((diagnostic) => ({ ...diagnostic })),
|
|
67
|
-
};
|
|
68
|
-
}
|
|
69
|
-
sanitize(value) {
|
|
70
|
-
return this.redact ? scrub(value) : value;
|
|
71
|
-
}
|
|
72
|
-
sanitizeArguments(value) {
|
|
73
|
-
return this.redact ? scrubValue(value) : structuredClone(value);
|
|
74
|
-
}
|
|
75
|
-
}
|
|
76
|
-
function isHarnessTool(name) {
|
|
77
|
-
return name === 'pass_test_step' || name === 'fail_test_step';
|
|
78
|
-
}
|
|
79
|
-
function truncate(value) {
|
|
80
|
-
return value.length <= TRANSCRIPT_CONTENT_LIMIT ? value : `${value.slice(0, TRANSCRIPT_CONTENT_LIMIT - 3)}...`;
|
|
81
|
-
}
|
|
82
|
-
//# sourceMappingURL=internal-step-evidence.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"internal-step-evidence.js","sourceRoot":"","sources":["../../src/runtime/internal-step-evidence.ts"],"names":[],"mappings":"AAGA,OAAO,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,uBAAuB,CAAA;AAUzD,MAAM,wBAAwB,GAAG,KAAK,CAAA;AAEtC,MAAM,kBAAkB,GAAoD;IAC3E,iBAAiB,EAAE,KAAK;IACxB,oBAAoB,EAAE,KAAK;IAC3B,eAAe,EAAE,OAAO;IACxB,mBAAmB,EAAE,OAAO;IAC5B,cAAc,EAAE,OAAO;IACvB,kBAAkB,EAAE,OAAO;IAC3B,YAAY,EAAE,OAAO;IACrB,gBAAgB,EAAE,OAAO;IACzB,uBAAuB,EAAE,OAAO;IAChC,WAAW,EAAE,OAAO;IACpB,gBAAgB,EAAE,OAAO;CACzB,CAAA;AASD,MAAM,OAAO,oBAAoB;IAOd;IACA;IACA;IACA;IACA;IAVD,SAAS,CAAQ;IACjB,SAAS,GAA2B,EAAE,CAAA;IACtC,UAAU,GAAsB,EAAE,CAAA;IAClC,WAAW,GAAiB,EAAE,CAAA;IAE/C,YACkB,IAAgB,EAChB,QAAgB,EAChB,MAAe,EACf,mBAAwC,EACxC,MAAoB,IAAI,CAAC,GAAG;QAJ5B,SAAI,GAAJ,IAAI,CAAY;QAChB,aAAQ,GAAR,QAAQ,CAAQ;QAChB,WAAM,GAAN,MAAM,CAAS;QACf,wBAAmB,GAAnB,mBAAmB,CAAqB;QACxC,QAAG,GAAH,GAAG,CAAyB;QAE7C,IAAI,CAAC,SAAS,GAAG,GAAG,EAAE,CAAA;IACvB,CAAC;IAED,sBAAsB,CAAC,IAAY,EAAE,OAAe;QACnD,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,WAAW,EAAE,OAAO,EAAE,QAAQ,CAAC,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,EAAE,CAAC,CAAA;IAC7F,CAAC;IAED,cAAc,CAAC,IAAY,EAAE,QAAkB,EAAE,YAA0B;QAC1E,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC;YACnB,IAAI;YACJ,QAAQ,EAAE,aAAa,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ;YAClE,IAAI,EAAE,QAAQ,CAAC,IAAI;YACnB,SAAS,EAAE,IAAI,CAAC,iBAAiB,CAAC,QAAQ,CAAC,SAAS,IAAI,EAAE,CAAC;YAC3D,MAAM,EAAE,YAAY,CAAC,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI;SACxD,CAAC,CAAA;QACF,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC;YACpB,IAAI;YACJ,IAAI,EAAE,MAAM;YACZ,OAAO,EAAE,QAAQ,CAAC,IAAI,CAAC,QAAQ,CAAC,GAAG,QAAQ,CAAC,IAAI,OAAO,YAAY,CAAC,QAAQ,EAAE,CAAC,CAAC;SAChF,CAAC,CAAA;IACH,CAAC;IAED,gBAAgB,CAAC,UAAsB;QACtC,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,EAAE,GAAG,UAAU,EAAE,OAAO,EAAE,IAAI,CAAC,mBAAmB,CAAC,IAAI,CAAC,UAAU,CAAC,OAAO,CAAC,EAAE,CAAC,CAAA;IACrG,CAAC;IAED,WAAW,CAAC,WAAoC,EAAE,KAAY;QAC7D,OAAO;YACN,IAAI,EAAE,EAAE,GAAG,IAAI,CAAC,IAAI,EAAE;YACtB,OAAO,EAAE,WAAW,CAAC,OAAO;YAC5B,QAAQ,EAAE,kBAAkB,CAAC,WAAW,CAAC,MAAM,CAAC;YAChD,MAAM,EAAE,WAAW,CAAC,MAAM;YAC1B,GAAG,CAAC,WAAW,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,QAAQ,CAAC,WAAW,CAAC,MAAM,CAAC,EAAE,CAAC;YAC1F,KAAK,EAAE,WAAW,CAAC,KAAK;YACxB,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC,SAAS;YACvC,KAAK,EAAE,EAAE,GAAG,KAAK,EAAE;YACnB,SAAS,EAAE,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,IAAI,EAAE,CAAC,CAAC;YACtD,UAAU,EAAE,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,KAAK,EAAE,CAAC,CAAC;YAC1D,WAAW,EAAE,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,UAAU,EAAE,CAAC,CAAC;SACtE,CAAA;IACF,CAAC;IAEO,QAAQ,CAAC,KAAa;QAC7B,OAAO,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAA;IAC1C,CAAC;IAEO,iBAAiB,CAAC,KAAc;QACvC,OAAO,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,eAAe,CAAC,KAAK,CAAC,CAAA;IAChE,CAAC;CACD;AAED,SAAS,aAAa,CAAC,IAAY;IAClC,OAAO,IAAI,KAAK,gBAAgB,IAAI,IAAI,KAAK,gBAAgB,CAAA;AAC9D,CAAC;AAED,SAAS,QAAQ,CAAC,KAAa;IAC9B,OAAO,KAAK,CAAC,MAAM,IAAI,wBAAwB,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,wBAAwB,GAAG,CAAC,CAAC,KAAK,CAAA;AAC/G,CAAC"}
|
|
@@ -1,9 +0,0 @@
|
|
|
1
|
-
import type { ExecutedStepResult } from '../contracts/types.js';
|
|
2
|
-
import type { HarnessEvidenceCandidate } from '../evidence/store.js';
|
|
3
|
-
import type { InternalStepReport } from './types.js';
|
|
4
|
-
export type AdaptedStepResult = Readonly<{
|
|
5
|
-
result: ExecutedStepResult;
|
|
6
|
-
evidence: readonly HarnessEvidenceCandidate[];
|
|
7
|
-
}>;
|
|
8
|
-
export declare function adaptStepResult(report: InternalStepReport): AdaptedStepResult;
|
|
9
|
-
//# sourceMappingURL=result-adapter.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"result-adapter.d.ts","sourceRoot":"","sources":["../../src/runtime/result-adapter.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,kBAAkB,EAAyB,MAAM,uBAAuB,CAAA;AAEtF,OAAO,KAAK,EAAE,wBAAwB,EAAE,MAAM,sBAAsB,CAAA;AACpE,OAAO,KAAK,EAAE,kBAAkB,EAA2C,MAAM,YAAY,CAAA;AA8B7F,MAAM,MAAM,iBAAiB,GAAG,QAAQ,CAAC;IACxC,MAAM,EAAE,kBAAkB,CAAA;IAC1B,QAAQ,EAAE,SAAS,wBAAwB,EAAE,CAAA;CAC7C,CAAC,CAAA;AAEF,wBAAgB,eAAe,CAAC,MAAM,EAAE,kBAAkB,GAAG,iBAAiB,CAkC7E"}
|
|
@@ -1,77 +0,0 @@
|
|
|
1
|
-
import { serializeJson } from '../contracts/serialize.js';
|
|
2
|
-
const PUBLIC_REASON_BY_INTERNAL = {
|
|
3
|
-
'met-expectation': 'met-expectation',
|
|
4
|
-
'failed-expectation': 'failed-expectation',
|
|
5
|
-
'loop-detected': 'loop-detected',
|
|
6
|
-
'turn-cap-exceeded': 'turn-cap-exceeded',
|
|
7
|
-
'step-timeout': 'step-timeout',
|
|
8
|
-
'scenario-timeout': 'scenario-timeout',
|
|
9
|
-
'tool-error': 'tool-error',
|
|
10
|
-
'provider-error': 'provider-error',
|
|
11
|
-
'token-budget-exceeded': 'token-budget-exceeded',
|
|
12
|
-
interrupted: 'interrupted',
|
|
13
|
-
'internal-error': 'internal-error',
|
|
14
|
-
};
|
|
15
|
-
const CATEGORY_BY_INTERNAL = {
|
|
16
|
-
'met-expectation': 'app',
|
|
17
|
-
'failed-expectation': 'app',
|
|
18
|
-
'loop-detected': 'model',
|
|
19
|
-
'turn-cap-exceeded': 'model',
|
|
20
|
-
'step-timeout': 'model',
|
|
21
|
-
'scenario-timeout': 'infra',
|
|
22
|
-
'tool-error': 'infra',
|
|
23
|
-
'provider-error': 'infra',
|
|
24
|
-
'token-budget-exceeded': 'infra',
|
|
25
|
-
interrupted: 'infra',
|
|
26
|
-
'internal-error': 'infra',
|
|
27
|
-
};
|
|
28
|
-
export function adaptStepResult(report) {
|
|
29
|
-
const reason = PUBLIC_REASON_BY_INTERNAL[report.reason];
|
|
30
|
-
const status = report.reason === 'met-expectation' ? 'passed' : 'failed';
|
|
31
|
-
if (report.outcome !== status) {
|
|
32
|
-
throw new Error(`Internal step '${report.step.id}' has inconsistent outcome and reason`);
|
|
33
|
-
}
|
|
34
|
-
const result = {
|
|
35
|
-
id: report.step.id,
|
|
36
|
-
status,
|
|
37
|
-
category: CATEGORY_BY_INTERNAL[report.reason],
|
|
38
|
-
reason,
|
|
39
|
-
...(report.actual === undefined ? {} : { actual: report.actual }),
|
|
40
|
-
turns: report.turns,
|
|
41
|
-
durationMs: report.durationMs,
|
|
42
|
-
usage: { ...report.usage },
|
|
43
|
-
toolCalls: report.toolCalls.map((call) => ({
|
|
44
|
-
turn: call.turn,
|
|
45
|
-
driverId: call.driverId,
|
|
46
|
-
name: call.name,
|
|
47
|
-
arguments: copyJson(call.arguments),
|
|
48
|
-
status: call.status,
|
|
49
|
-
})),
|
|
50
|
-
};
|
|
51
|
-
const evidence = [];
|
|
52
|
-
if (report.transcript.length > 0) {
|
|
53
|
-
evidence.push({
|
|
54
|
-
stepId: report.step.id,
|
|
55
|
-
kind: 'transcript',
|
|
56
|
-
mediaType: 'text/markdown',
|
|
57
|
-
content: renderTranscript(report),
|
|
58
|
-
});
|
|
59
|
-
}
|
|
60
|
-
return { result, evidence };
|
|
61
|
-
}
|
|
62
|
-
function copyJson(value) {
|
|
63
|
-
serializeJson(value);
|
|
64
|
-
return structuredClone(value);
|
|
65
|
-
}
|
|
66
|
-
function renderTranscript(report) {
|
|
67
|
-
const lines = [
|
|
68
|
-
`# ${report.step.action}`,
|
|
69
|
-
'',
|
|
70
|
-
`**Outcome:** ${report.outcome} - ${CATEGORY_BY_INTERNAL[report.reason]} / ${PUBLIC_REASON_BY_INTERNAL[report.reason]}`,
|
|
71
|
-
'',
|
|
72
|
-
];
|
|
73
|
-
for (const entry of report.transcript)
|
|
74
|
-
lines.push(`### Turn ${entry.turn} - ${entry.role}`, '', entry.content, '');
|
|
75
|
-
return lines.join('\n');
|
|
76
|
-
}
|
|
77
|
-
//# sourceMappingURL=result-adapter.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"result-adapter.js","sourceRoot":"","sources":["../../src/runtime/result-adapter.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,aAAa,EAAE,MAAM,2BAA2B,CAAA;AAIzD,MAAM,yBAAyB,GAAkD;IAChF,iBAAiB,EAAE,iBAAiB;IACpC,oBAAoB,EAAE,oBAAoB;IAC1C,eAAe,EAAE,eAAe;IAChC,mBAAmB,EAAE,mBAAmB;IACxC,cAAc,EAAE,cAAc;IAC9B,kBAAkB,EAAE,kBAAkB;IACtC,YAAY,EAAE,YAAY;IAC1B,gBAAgB,EAAE,gBAAgB;IAClC,uBAAuB,EAAE,uBAAuB;IAChD,WAAW,EAAE,aAAa;IAC1B,gBAAgB,EAAE,gBAAgB;CAClC,CAAA;AAED,MAAM,oBAAoB,GAAoD;IAC7E,iBAAiB,EAAE,KAAK;IACxB,oBAAoB,EAAE,KAAK;IAC3B,eAAe,EAAE,OAAO;IACxB,mBAAmB,EAAE,OAAO;IAC5B,cAAc,EAAE,OAAO;IACvB,kBAAkB,EAAE,OAAO;IAC3B,YAAY,EAAE,OAAO;IACrB,gBAAgB,EAAE,OAAO;IACzB,uBAAuB,EAAE,OAAO;IAChC,WAAW,EAAE,OAAO;IACpB,gBAAgB,EAAE,OAAO;CACzB,CAAA;AAOD,MAAM,UAAU,eAAe,CAAC,MAA0B;IACzD,MAAM,MAAM,GAAG,yBAAyB,CAAC,MAAM,CAAC,MAAM,CAAC,CAAA;IACvD,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,KAAK,iBAAiB,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAA;IACxE,IAAI,MAAM,CAAC,OAAO,KAAK,MAAM,EAAE,CAAC;QAC/B,MAAM,IAAI,KAAK,CAAC,kBAAkB,MAAM,CAAC,IAAI,CAAC,EAAE,uCAAuC,CAAC,CAAA;IACzF,CAAC;IAED,MAAM,MAAM,GAAuB;QAClC,EAAE,EAAE,MAAM,CAAC,IAAI,CAAC,EAAE;QAClB,MAAM;QACN,QAAQ,EAAE,oBAAoB,CAAC,MAAM,CAAC,MAAM,CAAC;QAC7C,MAAM;QACN,GAAG,CAAC,MAAM,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,CAAC;QACjE,KAAK,EAAE,MAAM,CAAC,KAAK;QACnB,UAAU,EAAE,MAAM,CAAC,UAAU;QAC7B,KAAK,EAAE,EAAE,GAAG,MAAM,CAAC,KAAK,EAAE;QAC1B,SAAS,EAAE,MAAM,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC;YAC1C,IAAI,EAAE,IAAI,CAAC,IAAI;YACf,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,IAAI,EAAE,IAAI,CAAC,IAAI;YACf,SAAS,EAAE,QAAQ,CAAC,IAAI,CAAC,SAAS,CAAC;YACnC,MAAM,EAAE,IAAI,CAAC,MAAM;SACnB,CAAC,CAAC;KACH,CAAA;IACD,MAAM,QAAQ,GAA+B,EAAE,CAAA;IAC/C,IAAI,MAAM,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAClC,QAAQ,CAAC,IAAI,CAAC;YACb,MAAM,EAAE,MAAM,CAAC,IAAI,CAAC,EAAE;YACtB,IAAI,EAAE,YAAY;YAClB,SAAS,EAAE,eAAe;YAC1B,OAAO,EAAE,gBAAgB,CAAC,MAAM,CAAC;SACjC,CAAC,CAAA;IACH,CAAC;IACD,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAA;AAC5B,CAAC;AAED,SAAS,QAAQ,CAAC,KAAc;IAC/B,aAAa,CAAC,KAAK,CAAC,CAAA;IACpB,OAAO,eAAe,CAAC,KAAK,CAAc,CAAA;AAC3C,CAAC;AAED,SAAS,gBAAgB,CAAC,MAA0B;IACnD,MAAM,KAAK,GAAG;QACb,KAAK,MAAM,CAAC,IAAI,CAAC,MAAM,EAAE;QACzB,EAAE;QACF,gBAAgB,MAAM,CAAC,OAAO,MAAM,oBAAoB,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,yBAAyB,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE;QACvH,EAAE;KACF,CAAA;IACD,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,UAAU;QAAE,KAAK,CAAC,IAAI,CAAC,YAAY,KAAK,CAAC,IAAI,MAAM,KAAK,CAAC,IAAI,EAAE,EAAE,EAAE,EAAE,KAAK,CAAC,OAAO,EAAE,EAAE,CAAC,CAAA;IAClH,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;AACxB,CAAC"}
|
package/dist/runtime/runner.d.ts
DELETED
|
@@ -1,43 +0,0 @@
|
|
|
1
|
-
import type { DriverDescriptorV1, ModelEgressPolicyV1 } from '../contracts/types.js';
|
|
2
|
-
import type { DriverSession, StepIntent } from '../driver.js';
|
|
3
|
-
import type { StepControl } from './scenario-control.js';
|
|
4
|
-
import type { InternalStepReport } from './types.js';
|
|
5
|
-
import { AiClient } from '../ai/client.js';
|
|
6
|
-
import { ScenarioUsageTracker } from './usage-tracker.js';
|
|
7
|
-
import type { LogLevel } from '../logging/level-logger.js';
|
|
8
|
-
import { type RuntimeLogger } from '../logging/types.js';
|
|
9
|
-
export type DriverCheckmateRunnerOptions = {
|
|
10
|
-
driverId: string;
|
|
11
|
-
descriptor: DriverDescriptorV1;
|
|
12
|
-
session: DriverSession;
|
|
13
|
-
allowedTools: '*' | readonly string[];
|
|
14
|
-
modelEgress: ModelEgressPolicyV1;
|
|
15
|
-
limits: {
|
|
16
|
-
turnsPerStep: number;
|
|
17
|
-
stepTimeoutMs: number;
|
|
18
|
-
requestTimeoutMs: number;
|
|
19
|
-
maxRetries: number;
|
|
20
|
-
loopMaxRepetitions: number;
|
|
21
|
-
budgetTokens?: number;
|
|
22
|
-
};
|
|
23
|
-
apiKey: string;
|
|
24
|
-
exactSecrets?: Iterable<string>;
|
|
25
|
-
logger?: RuntimeLogger;
|
|
26
|
-
logLevel?: LogLevel;
|
|
27
|
-
usageTracker?: ScenarioUsageTracker;
|
|
28
|
-
aiClient?: AiClient;
|
|
29
|
-
};
|
|
30
|
-
export declare class CheckmateRunner {
|
|
31
|
-
private readonly options;
|
|
32
|
-
private readonly config;
|
|
33
|
-
private readonly toolRegistry;
|
|
34
|
-
private readonly aiClient;
|
|
35
|
-
private readonly usageTracker;
|
|
36
|
-
private readonly diagnosticSanitizer;
|
|
37
|
-
private readonly runtimeLogger;
|
|
38
|
-
constructor(options: DriverCheckmateRunnerOptions);
|
|
39
|
-
teardown(): Promise<void>;
|
|
40
|
-
run(step: StepIntent, control: StepControl): Promise<InternalStepReport>;
|
|
41
|
-
}
|
|
42
|
-
export declare function createDriverRunner(options: DriverCheckmateRunnerOptions): CheckmateRunner;
|
|
43
|
-
//# sourceMappingURL=runner.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../src/runtime/runner.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,kBAAkB,EAAE,mBAAmB,EAAE,MAAM,uBAAuB,CAAA;AACpF,OAAO,KAAK,EAAE,aAAa,EAAE,UAAU,EAAE,MAAM,cAAc,CAAA;AAG7D,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAA;AACxD,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,YAAY,CAAA;AAIpD,OAAO,EAAE,QAAQ,EAAE,MAAM,iBAAiB,CAAA;AAC1C,OAAO,EAAE,oBAAoB,EAAE,MAAM,oBAAoB,CAAA;AAGzD,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,4BAA4B,CAAA;AAC1D,OAAO,EAAgB,KAAK,aAAa,EAAE,MAAM,qBAAqB,CAAA;AAGtE,MAAM,MAAM,4BAA4B,GAAG;IAC1C,QAAQ,EAAE,MAAM,CAAA;IAChB,UAAU,EAAE,kBAAkB,CAAA;IAC9B,OAAO,EAAE,aAAa,CAAA;IACtB,YAAY,EAAE,GAAG,GAAG,SAAS,MAAM,EAAE,CAAA;IACrC,WAAW,EAAE,mBAAmB,CAAA;IAChC,MAAM,EAAE;QACP,YAAY,EAAE,MAAM,CAAA;QACpB,aAAa,EAAE,MAAM,CAAA;QACrB,gBAAgB,EAAE,MAAM,CAAA;QACxB,UAAU,EAAE,MAAM,CAAA;QAClB,kBAAkB,EAAE,MAAM,CAAA;QAC1B,YAAY,CAAC,EAAE,MAAM,CAAA;KACrB,CAAA;IACD,MAAM,EAAE,MAAM,CAAA;IACd,YAAY,CAAC,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAA;IAC/B,MAAM,CAAC,EAAE,aAAa,CAAA;IACtB,QAAQ,CAAC,EAAE,QAAQ,CAAA;IACnB,YAAY,CAAC,EAAE,oBAAoB,CAAA;IACnC,QAAQ,CAAC,EAAE,QAAQ,CAAA;CACnB,CAAA;AAED,qBAAa,eAAe;IAQf,OAAO,CAAC,QAAQ,CAAC,OAAO;IAPpC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAe;IACtC,OAAO,CAAC,QAAQ,CAAC,YAAY,CAAc;IAC3C,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAU;IACnC,OAAO,CAAC,QAAQ,CAAC,YAAY,CAAsB;IACnD,OAAO,CAAC,QAAQ,CAAC,mBAAmB,CAAqB;IACzD,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAe;gBAEhB,OAAO,EAAE,4BAA4B;IAmC5D,QAAQ,IAAI,OAAO,CAAC,IAAI,CAAC;IAEzB,GAAG,CAAC,IAAI,EAAE,UAAU,EAAE,OAAO,EAAE,WAAW,GAAG,OAAO,CAAC,kBAAkB,CAAC;CAa9E;AAED,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,4BAA4B,GAAG,eAAe,CAEzF"}
|