@post-print/agent-test 1.1.9 → 2.0.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +18 -66
- package/dist/cli.d.ts +2 -64
- package/dist/cli.d.ts.map +1 -1
- package/dist/cli.js +26 -691
- package/dist/cli.js.map +1 -1
- package/dist/compare-scenario.d.ts +1 -1
- package/dist/compare-scenario.d.ts.map +1 -1
- package/dist/compare-scenario.js +72 -75
- package/dist/compare-scenario.js.map +1 -1
- package/dist/failures.d.ts +5 -1
- package/dist/failures.d.ts.map +1 -1
- package/dist/failures.js +1 -1
- package/dist/failures.js.map +1 -1
- package/dist/host-login.d.ts +1 -26
- package/dist/host-login.d.ts.map +1 -1
- package/dist/host-login.js +1 -92
- package/dist/host-login.js.map +1 -1
- package/dist/index.d.ts +6 -28
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +4 -26
- package/dist/index.js.map +1 -1
- package/dist/scenario-story.d.ts.map +1 -1
- package/dist/scenario-story.js +57 -39
- package/dist/scenario-story.js.map +1 -1
- package/dist/sdk/cli.d.ts +3 -0
- package/dist/sdk/cli.d.ts.map +1 -0
- package/dist/sdk/cli.js +20 -0
- package/dist/sdk/cli.js.map +1 -0
- package/dist/sdk/definitions.d.ts +22 -0
- package/dist/sdk/definitions.d.ts.map +1 -0
- package/dist/sdk/definitions.js +58 -0
- package/dist/sdk/definitions.js.map +1 -0
- package/dist/sdk/expect.d.ts +31 -0
- package/dist/sdk/expect.d.ts.map +1 -0
- package/dist/sdk/expect.js +68 -0
- package/dist/sdk/expect.js.map +1 -0
- package/dist/sdk/judge.d.ts +16 -0
- package/dist/sdk/judge.d.ts.map +1 -0
- package/dist/sdk/judge.js +82 -0
- package/dist/sdk/judge.js.map +1 -0
- package/dist/sdk/metrics.d.ts +3 -0
- package/dist/sdk/metrics.d.ts.map +1 -0
- package/dist/sdk/metrics.js +24 -0
- package/dist/sdk/metrics.js.map +1 -0
- package/dist/sdk/reporter.d.ts +12 -0
- package/dist/sdk/reporter.d.ts.map +1 -0
- package/dist/sdk/reporter.js +58 -0
- package/dist/sdk/reporter.js.map +1 -0
- package/dist/sdk/runtime.d.ts +33 -0
- package/dist/sdk/runtime.d.ts.map +1 -0
- package/dist/sdk/runtime.js +188 -0
- package/dist/sdk/runtime.js.map +1 -0
- package/dist/sdk/test.d.ts +27 -0
- package/dist/sdk/test.d.ts.map +1 -0
- package/dist/sdk/test.js +92 -0
- package/dist/sdk/test.js.map +1 -0
- package/dist/sdk/types.d.ts +75 -0
- package/dist/sdk/types.d.ts.map +1 -0
- package/dist/sdk/types.js +2 -0
- package/dist/sdk/types.js.map +1 -0
- package/dist/sdk/usage.d.ts +10 -0
- package/dist/sdk/usage.d.ts.map +1 -0
- package/dist/sdk/usage.js +13 -0
- package/dist/sdk/usage.js.map +1 -0
- package/dist/sdk/viewer-results.d.ts +43 -0
- package/dist/sdk/viewer-results.d.ts.map +1 -0
- package/dist/sdk/viewer-results.js +126 -0
- package/dist/sdk/viewer-results.js.map +1 -0
- package/dist/sdk/viewer.d.ts +11 -0
- package/dist/sdk/viewer.d.ts.map +1 -0
- package/dist/sdk/viewer.js +160 -0
- package/dist/sdk/viewer.js.map +1 -0
- package/dist/sdk/workspace.d.ts +33 -0
- package/dist/sdk/workspace.d.ts.map +1 -0
- package/dist/sdk/workspace.js +100 -0
- package/dist/sdk/workspace.js.map +1 -0
- package/dist/types.d.ts +4 -40
- package/dist/types.d.ts.map +1 -1
- package/dist/viewer/catalog.d.ts +2 -8
- package/dist/viewer/catalog.d.ts.map +1 -1
- package/dist/viewer/catalog.js +23 -128
- package/dist/viewer/catalog.js.map +1 -1
- package/dist/viewer/client.d.ts.map +1 -1
- package/dist/viewer/client.js +12 -12
- package/dist/viewer/client.js.map +1 -1
- package/dist/viewer/run-controller.d.ts +4 -2
- package/dist/viewer/run-controller.d.ts.map +1 -1
- package/dist/viewer/run-controller.js +297 -248
- package/dist/viewer/run-controller.js.map +1 -1
- package/dist/viewer/server.d.ts +4 -3
- package/dist/viewer/server.d.ts.map +1 -1
- package/dist/viewer/server.js +149 -112
- package/dist/viewer/server.js.map +1 -1
- package/dist/viewer/styles.d.ts +2 -0
- package/dist/viewer/styles.d.ts.map +1 -0
- package/dist/viewer/styles.js +444 -0
- package/dist/viewer/styles.js.map +1 -0
- package/dist/worker-pool.d.ts +5 -1
- package/dist/worker-pool.d.ts.map +1 -1
- package/dist/worker-pool.js +1 -1
- package/dist/worker-pool.js.map +1 -1
- package/package.json +7 -7
- package/dist/check.d.ts +0 -38
- package/dist/check.d.ts.map +0 -1
- package/dist/check.js +0 -169
- package/dist/check.js.map +0 -1
- package/dist/command-allowlist.d.ts +0 -9
- package/dist/command-allowlist.d.ts.map +0 -1
- package/dist/command-allowlist.js +0 -29
- package/dist/command-allowlist.js.map +0 -1
- package/dist/debug-bundle.d.ts +0 -81
- package/dist/debug-bundle.d.ts.map +0 -1
- package/dist/debug-bundle.js +0 -392
- package/dist/debug-bundle.js.map +0 -1
- package/dist/discover-suites.d.ts +0 -2
- package/dist/discover-suites.d.ts.map +0 -1
- package/dist/discover-suites.js +0 -19
- package/dist/discover-suites.js.map +0 -1
- package/dist/doctor.d.ts +0 -24
- package/dist/doctor.d.ts.map +0 -1
- package/dist/doctor.js +0 -272
- package/dist/doctor.js.map +0 -1
- package/dist/expect.d.ts +0 -59
- package/dist/expect.d.ts.map +0 -1
- package/dist/expect.js +0 -518
- package/dist/expect.js.map +0 -1
- package/dist/host-log.d.ts +0 -17
- package/dist/host-log.d.ts.map +0 -1
- package/dist/host-log.js +0 -154
- package/dist/host-log.js.map +0 -1
- package/dist/hosts.d.ts +0 -22
- package/dist/hosts.d.ts.map +0 -1
- package/dist/hosts.js +0 -84
- package/dist/hosts.js.map +0 -1
- package/dist/html-report.d.ts +0 -18
- package/dist/html-report.d.ts.map +0 -1
- package/dist/html-report.js +0 -1276
- package/dist/html-report.js.map +0 -1
- package/dist/live-isolation.d.ts +0 -82
- package/dist/live-isolation.d.ts.map +0 -1
- package/dist/live-isolation.js +0 -399
- package/dist/live-isolation.js.map +0 -1
- package/dist/live-timeout.d.ts +0 -11
- package/dist/live-timeout.d.ts.map +0 -1
- package/dist/live-timeout.js +0 -29
- package/dist/live-timeout.js.map +0 -1
- package/dist/load-adapters.d.ts +0 -13
- package/dist/load-adapters.d.ts.map +0 -1
- package/dist/load-adapters.js +0 -76
- package/dist/load-adapters.js.map +0 -1
- package/dist/load-suite.d.ts +0 -22
- package/dist/load-suite.d.ts.map +0 -1
- package/dist/load-suite.js +0 -154
- package/dist/load-suite.js.map +0 -1
- package/dist/mcp-config.d.ts +0 -13
- package/dist/mcp-config.d.ts.map +0 -1
- package/dist/mcp-config.js +0 -121
- package/dist/mcp-config.js.map +0 -1
- package/dist/preflight.d.ts +0 -3
- package/dist/preflight.d.ts.map +0 -1
- package/dist/preflight.js +0 -17
- package/dist/preflight.js.map +0 -1
- package/dist/progress.d.ts +0 -24
- package/dist/progress.d.ts.map +0 -1
- package/dist/progress.js +0 -111
- package/dist/progress.js.map +0 -1
- package/dist/record-trace.d.ts +0 -64
- package/dist/record-trace.d.ts.map +0 -1
- package/dist/record-trace.js +0 -124
- package/dist/record-trace.js.map +0 -1
- package/dist/report-preview-server.d.ts +0 -2
- package/dist/report-preview-server.d.ts.map +0 -1
- package/dist/report-preview-server.js +0 -24
- package/dist/report-preview-server.js.map +0 -1
- package/dist/report-preview.d.ts +0 -35
- package/dist/report-preview.d.ts.map +0 -1
- package/dist/report-preview.js +0 -205
- package/dist/report-preview.js.map +0 -1
- package/dist/run-suite.d.ts +0 -139
- package/dist/run-suite.d.ts.map +0 -1
- package/dist/run-suite.js +0 -1688
- package/dist/run-suite.js.map +0 -1
- package/dist/scenario-finalizer.d.ts +0 -27
- package/dist/scenario-finalizer.d.ts.map +0 -1
- package/dist/scenario-finalizer.js +0 -39
- package/dist/scenario-finalizer.js.map +0 -1
- package/dist/scenario-retry.d.ts +0 -15
- package/dist/scenario-retry.d.ts.map +0 -1
- package/dist/scenario-retry.js +0 -52
- package/dist/scenario-retry.js.map +0 -1
- package/dist/scenario-seed.d.ts +0 -13
- package/dist/scenario-seed.d.ts.map +0 -1
- package/dist/scenario-seed.js +0 -72
- package/dist/scenario-seed.js.map +0 -1
- package/dist/suite-summary.d.ts +0 -16
- package/dist/suite-summary.d.ts.map +0 -1
- package/dist/suite-summary.js +0 -218
- package/dist/suite-summary.js.map +0 -1
- package/dist/theme.d.ts +0 -96
- package/dist/theme.d.ts.map +0 -1
- package/dist/theme.js +0 -446
- package/dist/theme.js.map +0 -1
- package/dist/validate-seeds.d.ts +0 -20
- package/dist/validate-seeds.d.ts.map +0 -1
- package/dist/validate-seeds.js +0 -72
- package/dist/validate-seeds.js.map +0 -1
- package/dist/validate-suite.d.ts +0 -24
- package/dist/validate-suite.d.ts.map +0 -1
- package/dist/validate-suite.js +0 -542
- package/dist/validate-suite.js.map +0 -1
- package/dist/viewer/cli.d.ts +0 -17
- package/dist/viewer/cli.d.ts.map +0 -1
- package/dist/viewer/cli.js +0 -38
- package/dist/viewer/cli.js.map +0 -1
- package/dist/viewer/compare-metrics.d.ts +0 -30
- package/dist/viewer/compare-metrics.d.ts.map +0 -1
- package/dist/viewer/compare-metrics.js +0 -101
- package/dist/viewer/compare-metrics.js.map +0 -1
- package/dist/viewer/context-files.d.ts +0 -8
- package/dist/viewer/context-files.d.ts.map +0 -1
- package/dist/viewer/context-files.js +0 -57
- package/dist/viewer/context-files.js.map +0 -1
- package/dist/viewer/emit.d.ts +0 -7
- package/dist/viewer/emit.d.ts.map +0 -1
- package/dist/viewer/emit.js +0 -25
- package/dist/viewer/emit.js.map +0 -1
- package/dist/viewer/live-runner.d.ts +0 -27
- package/dist/viewer/live-runner.d.ts.map +0 -1
- package/dist/viewer/live-runner.js +0 -166
- package/dist/viewer/live-runner.js.map +0 -1
- package/dist/viewer/session-server.d.ts +0 -2
- package/dist/viewer/session-server.d.ts.map +0 -1
- package/dist/viewer/session-server.js +0 -71
- package/dist/viewer/session-server.js.map +0 -1
- package/dist/warnings.d.ts +0 -3
- package/dist/warnings.d.ts.map +0 -1
- package/dist/warnings.js +0 -13
- package/dist/warnings.js.map +0 -1
package/dist/types.d.ts
CHANGED
|
@@ -99,7 +99,7 @@ export interface CompareArm {
|
|
|
99
99
|
skills?: SkillContextSetting;
|
|
100
100
|
contextSources?: string[];
|
|
101
101
|
mcpServers?: Record<string, McpServerConfig>;
|
|
102
|
-
|
|
102
|
+
includeGlobalSkills?: boolean;
|
|
103
103
|
seedPatch?: string;
|
|
104
104
|
seedStageOnly?: boolean;
|
|
105
105
|
/**
|
|
@@ -184,54 +184,17 @@ export interface AgentScenario {
|
|
|
184
184
|
* Load host-global user skills from the developer machine.
|
|
185
185
|
* Default false. Keep false for a custom `workspace` fixture.
|
|
186
186
|
*/
|
|
187
|
-
|
|
187
|
+
includeGlobalSkills?: boolean;
|
|
188
188
|
/** OpenAI only: allow network access in the Codex workspace-write sandbox. Default false. */
|
|
189
189
|
networkAccess?: boolean;
|
|
190
190
|
/** Live-only: apply patch + commit in worktree so pr-mode branch diff exists. */
|
|
191
191
|
seedPatch?: string;
|
|
192
192
|
/** Live-only: with seedPatch, stage changes without committing (staged review mode). */
|
|
193
193
|
seedStageOnly?: boolean;
|
|
194
|
-
/**
|
|
195
|
-
* Assertions / judge criteria. May be omitted in scenarios.json when supplied via
|
|
196
|
-
* sibling `rubrics.json` / `scenarios.rubric.json` or `--rubrics-dir` (harness-only);
|
|
197
|
-
* `loadSuiteFile` always normalizes to an object.
|
|
198
|
-
*/
|
|
194
|
+
/** Captured rubric metadata retained for imported viewer results. */
|
|
199
195
|
rubric: ScenarioRubric;
|
|
200
196
|
skip?: boolean;
|
|
201
197
|
}
|
|
202
|
-
export interface AgentSuiteDefaults {
|
|
203
|
-
host?: AgentHost;
|
|
204
|
-
/** Default context delivery for scenarios in this suite. */
|
|
205
|
-
contextMode?: ContextMode;
|
|
206
|
-
profile?: ContextProfile;
|
|
207
|
-
/** Extra skill folders to overlay, or `"none"`. */
|
|
208
|
-
skills?: SkillContextSetting;
|
|
209
|
-
/** Additive repo-relative context paths or `.skeleton/customize/` basenames. */
|
|
210
|
-
contextSources?: string[];
|
|
211
|
-
/** Inline MCP servers for direct agent runs. */
|
|
212
|
-
mcpServers?: Record<string, McpServerConfig>;
|
|
213
|
-
/**
|
|
214
|
-
* Caller-relative folder that becomes the sealed repo.
|
|
215
|
-
* Scenario `workspace` wins when both are set.
|
|
216
|
-
*/
|
|
217
|
-
workspace?: string;
|
|
218
|
-
/**
|
|
219
|
-
* Load host-global user skills from the developer machine.
|
|
220
|
-
* Default false. Scenario `allowUserSkills` wins when both are set.
|
|
221
|
-
*/
|
|
222
|
-
allowUserSkills?: boolean;
|
|
223
|
-
}
|
|
224
|
-
export interface AgentSuiteFile {
|
|
225
|
-
name: string;
|
|
226
|
-
description?: string;
|
|
227
|
-
/**
|
|
228
|
-
* Host matrix for this suite (Playwright-style projects).
|
|
229
|
-
* When set, a run expands once per host. `--host` then filters the list.
|
|
230
|
-
*/
|
|
231
|
-
hosts?: AgentHost[];
|
|
232
|
-
defaults?: AgentSuiteDefaults;
|
|
233
|
-
scenarios: AgentScenario[];
|
|
234
|
-
}
|
|
235
198
|
export interface AssertionFailure {
|
|
236
199
|
matcher: string;
|
|
237
200
|
message: string;
|
|
@@ -288,6 +251,7 @@ export interface ScenarioStory {
|
|
|
288
251
|
sections?: StorySection[];
|
|
289
252
|
}
|
|
290
253
|
export interface ScenarioResult {
|
|
254
|
+
authoring?: "typescript";
|
|
291
255
|
suite: string;
|
|
292
256
|
scenario: string;
|
|
293
257
|
passed: boolean;
|
package/dist/types.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACX,SAAS,EACT,UAAU,EACV,UAAU,EACV,WAAW,EACX,cAAc,EACd,eAAe,EACf,sBAAsB,EACtB,mBAAmB,EACnB,MAAM,2BAA2B,CAAC;AAEnC,YAAY,EACX,UAAU,EACV,WAAW,EACX,eAAe,EACf,sBAAsB,GACtB,MAAM,2BAA2B,CAAC;AAEnC,MAAM,MAAM,eAAe,GAAG,MAAM,GAAG;IAAE,EAAE,CAAC,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,CAAA;CAAE,CAAC;AAEzE,sDAAsD;AACtD,MAAM,MAAM,qBAAqB,GAAG,gBAAgB,GAAG,SAAS,GAAG,QAAQ,CAAC;AAE5E,qDAAqD;AACrD,MAAM,WAAW,mBAAmB;IACnC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,qBAAqB,CAAC;IAC9B,GAAG,EAAE,MAAM,CAAC;CACZ;AAED,MAAM,WAAW,cAAc;IAC9B,IAAI,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,CAAC;IACjC,kGAAkG;IAClG,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,0DAA0D;IAC1D,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,qHAAqH;IACrH,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;IACnB,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;IACnB,mGAAmG;IACnG,mBAAmB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC/B;;;;OAIG;IACH,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;IAC3B;;;OAGG;IACH,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB,6EAA6E;IAC7E,oBAAoB,CAAC,EAAE,MAAM,EAAE,CAAC;IAChC,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;IAC3B;;;OAGG;IACH,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB,mEAAmE;IACnE,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;IAC3B;;;OAGG;IACH,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;IAC3B,kBAAkB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC9B,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,WAAW,CAAC,EAAE,OAAO,GAAG,UAAU,GAAG,UAAU,GAAG,MAAM,CAAC;IACzD,0FAA0F;IAC1F,KAAK,CAAC,EAAE,eAAe,EAAE,CAAC;CAC1B;AAED,8EAA8E;AAC9E,MAAM,MAAM,YAAY,GAAG,MAAM,CAAC;AAElC,wDAAwD;AACxD,MAAM,MAAM,aAAa,GACtB,SAAS,GACT,OAAO,GACP,QAAQ,GACR,OAAO,GACP,YAAY,GACZ,SAAS,MAAM,EAAE,CAAC;AAErB,MAAM,MAAM,mBAAmB,GAAG,OAAO,GAAG,UAAU,GAAG,QAAQ,GAAG,SAAS,GAAG,aAAa,CAAC;AAE9F,MAAM,MAAM,WAAW,GACpB;IAAE,MAAM,EAAE,aAAa,CAAC;IAAC,MAAM,EAAE,YAAY,CAAC;IAAC,KAAK,EAAE,YAAY,CAAA;CAAE,GACpE;IACA,MAAM,EAAE,aAAa,CAAC;IACtB,GAAG,EAAE,YAAY,CAAC;IAClB,QAAQ,EAAE,mBAAmB,CAAC;IAC9B,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,MAAM,CAAC;CAC/B,CAAC;AAEL,MAAM,WAAW,kBAAkB;IAClC,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,iBAAiB;IACjC,IAAI,EAAE,WAAW,CAAC;IAClB,MAAM,EAAE,OAAO,CAAC;IAChB,IAAI,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,MAAM,CAAC;IAChC,KAAK,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,MAAM,CAAC;IACjC,OAAO,EAAE,MAAM,CAAC;CAChB;AAED,yFAAyF;AACzF,MAAM,WAAW,UAAU;IAC1B,0EAA0E;IAC1E,EAAE,CAAC,EAAE,YAAY,CAAC;IAClB,qFAAqF;IACrF,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,0CAA0C;IAC1C,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,SAAS,CAAC;IACjB,iDAAiD;IACjD,WAAW,CAAC,EAAE,WAAW,CAAC;IAC1B,OAAO,CAAC,EAAE,cAAc,CAAC;IACzB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,MAAM,CAAC,EAAE,mBAAmB,CAAC;IAC7B,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;IAC1B,UAAU,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,eAAe,CAAC,CAAC;IAC7C,
|
|
1
|
+
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACX,SAAS,EACT,UAAU,EACV,UAAU,EACV,WAAW,EACX,cAAc,EACd,eAAe,EACf,sBAAsB,EACtB,mBAAmB,EACnB,MAAM,2BAA2B,CAAC;AAEnC,YAAY,EACX,UAAU,EACV,WAAW,EACX,eAAe,EACf,sBAAsB,GACtB,MAAM,2BAA2B,CAAC;AAEnC,MAAM,MAAM,eAAe,GAAG,MAAM,GAAG;IAAE,EAAE,CAAC,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,CAAA;CAAE,CAAC;AAEzE,sDAAsD;AACtD,MAAM,MAAM,qBAAqB,GAAG,gBAAgB,GAAG,SAAS,GAAG,QAAQ,CAAC;AAE5E,qDAAqD;AACrD,MAAM,WAAW,mBAAmB;IACnC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,qBAAqB,CAAC;IAC9B,GAAG,EAAE,MAAM,CAAC;CACZ;AAED,MAAM,WAAW,cAAc;IAC9B,IAAI,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,CAAC;IACjC,kGAAkG;IAClG,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,0DAA0D;IAC1D,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,qHAAqH;IACrH,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;IACnB,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;IACnB,mGAAmG;IACnG,mBAAmB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC/B;;;;OAIG;IACH,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;IAC3B;;;OAGG;IACH,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB,6EAA6E;IAC7E,oBAAoB,CAAC,EAAE,MAAM,EAAE,CAAC;IAChC,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;IAC3B;;;OAGG;IACH,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB,mEAAmE;IACnE,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;IAC3B;;;OAGG;IACH,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;IAC3B,kBAAkB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC9B,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,WAAW,CAAC,EAAE,OAAO,GAAG,UAAU,GAAG,UAAU,GAAG,MAAM,CAAC;IACzD,0FAA0F;IAC1F,KAAK,CAAC,EAAE,eAAe,EAAE,CAAC;CAC1B;AAED,8EAA8E;AAC9E,MAAM,MAAM,YAAY,GAAG,MAAM,CAAC;AAElC,wDAAwD;AACxD,MAAM,MAAM,aAAa,GACtB,SAAS,GACT,OAAO,GACP,QAAQ,GACR,OAAO,GACP,YAAY,GACZ,SAAS,MAAM,EAAE,CAAC;AAErB,MAAM,MAAM,mBAAmB,GAAG,OAAO,GAAG,UAAU,GAAG,QAAQ,GAAG,SAAS,GAAG,aAAa,CAAC;AAE9F,MAAM,MAAM,WAAW,GACpB;IAAE,MAAM,EAAE,aAAa,CAAC;IAAC,MAAM,EAAE,YAAY,CAAC;IAAC,KAAK,EAAE,YAAY,CAAA;CAAE,GACpE;IACA,MAAM,EAAE,aAAa,CAAC;IACtB,GAAG,EAAE,YAAY,CAAC;IAClB,QAAQ,EAAE,mBAAmB,CAAC;IAC9B,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,MAAM,CAAC;CAC/B,CAAC;AAEL,MAAM,WAAW,kBAAkB;IAClC,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,iBAAiB;IACjC,IAAI,EAAE,WAAW,CAAC;IAClB,MAAM,EAAE,OAAO,CAAC;IAChB,IAAI,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,MAAM,CAAC;IAChC,KAAK,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,MAAM,CAAC;IACjC,OAAO,EAAE,MAAM,CAAC;CAChB;AAED,yFAAyF;AACzF,MAAM,WAAW,UAAU;IAC1B,0EAA0E;IAC1E,EAAE,CAAC,EAAE,YAAY,CAAC;IAClB,qFAAqF;IACrF,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,0CAA0C;IAC1C,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,SAAS,CAAC;IACjB,iDAAiD;IACjD,WAAW,CAAC,EAAE,WAAW,CAAC;IAC1B,OAAO,CAAC,EAAE,cAAc,CAAC;IACzB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,MAAM,CAAC,EAAE,mBAAmB,CAAC;IAC7B,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;IAC1B,UAAU,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,eAAe,CAAC,CAAC;IAC7C,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB;;;OAGG;IACH,MAAM,CAAC,EAAE,cAAc,CAAC;CACxB;AAED;;;;;GAKG;AACH,MAAM,WAAW,eAAe;IAC/B,CAAC,CAAC,EAAE,UAAU,CAAC;IACf,CAAC,CAAC,EAAE,UAAU,CAAC;IACf,sGAAsG;IACtG,IAAI,CAAC,EAAE,UAAU,EAAE,CAAC;IACpB,yDAAyD;IACzD,YAAY,CAAC,EAAE,kBAAkB,EAAE,CAAC;IACpC,wDAAwD;IACxD,KAAK,CAAC,EAAE,WAAW,EAAE,CAAC;CACtB;AAED,MAAM,WAAW,gBAAgB;IAChC,EAAE,EAAE,YAAY,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,wCAAwC;IACxC,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,KAAK,CAAC,EAAE,UAAU,CAAC;IACnB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,QAAQ,CAAC,EAAE,gBAAgB,EAAE,CAAC;IAC9B,aAAa,CAAC,EAAE,kBAAkB,EAAE,CAAC;IACrC,uEAAuE;IACvE,WAAW,CAAC,EAAE,WAAW,CAAC;IAC1B,gDAAgD;IAChD,YAAY,CAAC,EAAE,mBAAmB,EAAE,CAAC;IACrC,yEAAyE;IACzE,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,qDAAqD;IACrD,cAAc,CAAC,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,OAAO,CAAC,IAAI,CAAC,CAAA;KAAE,CAAC;CAC9E;AAED,MAAM,WAAW,qBAAqB;IACrC,iCAAiC;IACjC,IAAI,EAAE,gBAAgB,EAAE,CAAC;IACzB,qCAAqC;IACrC,CAAC,CAAC,EAAE,gBAAgB,CAAC;IACrB,qCAAqC;IACrC,CAAC,CAAC,EAAE,gBAAgB,CAAC;IACrB,KAAK,CAAC,EAAE,WAAW,EAAE,CAAC;IACtB,WAAW,CAAC,EAAE,iBAAiB,EAAE,CAAC;CAClC;AAED,MAAM,WAAW,aAAa;IAC7B,IAAI,EAAE,MAAM,CAAC;IACb,0DAA0D;IAC1D,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,gFAAgF;IAChF,OAAO,CAAC,EAAE,eAAe,CAAC;IAC1B,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,SAAS,CAAC;IACjB,oEAAoE;IACpE,WAAW,CAAC,EAAE,WAAW,CAAC;IAC1B,OAAO,CAAC,EAAE,cAAc,CAAC;IACzB,wFAAwF;IACxF,MAAM,CAAC,EAAE,mBAAmB,CAAC;IAC7B,2EAA2E;IAC3E,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;IAC1B,sFAAsF;IACtF,UAAU,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,eAAe,CAAC,CAAC;IAC7C;;;OAGG;IACH,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;OAGG;IACH,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,6FAA6F;IAC7F,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,iFAAiF;IACjF,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,wFAAwF;IACxF,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,qEAAqE;IACrE,MAAM,EAAE,cAAc,CAAC;IACvB,IAAI,CAAC,EAAE,OAAO,CAAC;CACf;AAED,MAAM,WAAW,gBAAgB;IAChC,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,EAAE,MAAM,CAAC;IAChB,uEAAuE;IACvE,QAAQ,EAAE,eAAe,CAAC;IAC1B,uDAAuD;IACvD,QAAQ,CAAC,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,MAAM,eAAe,GACxB,aAAa,GACb,aAAa,GACb,aAAa,GACb,eAAe,GACf,eAAe,GACf,iBAAiB,CAAC;AAErB,MAAM,WAAW,kBAAkB;IAClC,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,OAAO,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,EAAE,CAAC;IACpB,iDAAiD;IACjD,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,mDAAmD;IACnD,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,EAAE;QAAE,OAAO,CAAC,EAAE,MAAM,CAAC;QAAC,IAAI,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;IAC/C,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,iBAAiB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC7B,KAAK,CAAC,EAAE,UAAU,CAAC;CACnB;AAED,MAAM,MAAM,gBAAgB,GAAG,MAAM,GAAG,MAAM,GAAG,MAAM,CAAC;AAExD,wCAAwC;AACxC,MAAM,WAAW,UAAU;IAC1B,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,gBAAgB,CAAC;CACzB;AAED,+DAA+D;AAC/D,MAAM,WAAW,YAAY;IAC5B,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,MAAM,EAAE,UAAU,EAAE,CAAC;IACrB,6CAA6C;IAC7C,KAAK,CAAC,EAAE,MAAM,EAAE,CAAC;CACjB;AAED,uEAAuE;AACvE,MAAM,WAAW,aAAa;IAC7B,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,oDAAoD;IACpD,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,yEAAyE;IACzE,QAAQ,CAAC,EAAE,YAAY,EAAE,CAAC;CAC1B;AAED,MAAM,WAAW,cAAc;IAC9B,SAAS,CAAC,EAAE,YAAY,CAAC;IACzB,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,OAAO,CAAC;IAChB,QAAQ,EAAE,gBAAgB,EAAE,CAAC;IAC7B,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,qFAAqF;IACrF,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,6DAA6D;IAC7D,aAAa,CAAC,EAAE,kBAAkB,EAAE,CAAC;IACrC,uDAAuD;IACvD,KAAK,CAAC,EAAE,sBAAsB,CAAC;IAC/B,oCAAoC;IACpC,UAAU,CAAC,EAAE,UAAU,CAAC;IACxB,kDAAkD;IAClD,UAAU,CAAC,EAAE,UAAU,CAAC;IACxB,gEAAgE;IAChE,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,qFAAqF;IACrF,WAAW,CAAC,EAAE,WAAW,CAAC;IAC1B,gDAAgD;IAChD,YAAY,CAAC,EAAE,mBAAmB,EAAE,CAAC;IACrC,yEAAyE;IACzE,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,0DAA0D;IAC1D,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,+EAA+E;IAC/E,KAAK,CAAC,EAAE,UAAU,CAAC;IACnB,wDAAwD;IACxD,OAAO,CAAC,EAAE,qBAAqB,CAAC;IAChC,2EAA2E;IAC3E,KAAK,CAAC,EAAE,aAAa,CAAC;IACtB,0EAA0E;IAC1E,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,+EAA+E;IAC/E,cAAc,CAAC,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,OAAO,CAAC,IAAI,CAAC,CAAA;KAAE,CAAC;CAC9E;AAED,+DAA+D;AAC/D,MAAM,WAAW,UAAU;IAC1B,kBAAkB,EAAE,MAAM,CAAC;IAC3B,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,eAAe,CAAC,EAAE,MAAM,CAAC;CACzB;AAED,MAAM,WAAW,UAAU;IAC1B,aAAa,EAAE,MAAM,CAAC;IACtB,cAAc,EAAE,MAAM,CAAC;IACvB,oBAAoB,EAAE,MAAM,CAAC;IAC7B,aAAa,EAAE,MAAM,CAAC;IACtB,eAAe,EAAE,MAAM,CAAC;IACxB,kBAAkB,EAAE,MAAM,CAAC;IAC3B,gEAAgE;IAChE,gBAAgB,EAAE,MAAM,CAAC;IACzB,qEAAqE;IACrE,wBAAwB,EAAE,MAAM,CAAC;IACjC,KAAK,CAAC,EAAE,UAAU,CAAC;CACnB;AAED,MAAM,WAAW,cAAc;IAC9B,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,SAAS,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,EAAE,cAAc,EAAE,CAAC;IAC1B,OAAO,CAAC,EAAE,UAAU,CAAC;CACrB"}
|
package/dist/viewer/catalog.d.ts
CHANGED
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import {
|
|
1
|
+
import type { AgentHost, ContextMode } from "@post-print/agent-harness";
|
|
2
2
|
import type { CompareGate, CompareJudgeMetric, ScenarioRubric } from "../types.js";
|
|
3
3
|
export interface ViewerCatalogArm {
|
|
4
4
|
id: string;
|
|
@@ -18,6 +18,7 @@ export interface ViewerSuppliedMcpServer {
|
|
|
18
18
|
tools: string[];
|
|
19
19
|
}
|
|
20
20
|
export interface ViewerCatalogScenario {
|
|
21
|
+
authoring?: "typescript";
|
|
21
22
|
name: string;
|
|
22
23
|
description?: string;
|
|
23
24
|
prompt: string;
|
|
@@ -45,11 +46,6 @@ export interface ViewerCatalog {
|
|
|
45
46
|
defaultSelectedHosts: AgentHost[];
|
|
46
47
|
suites: ViewerCatalogSuite[];
|
|
47
48
|
}
|
|
48
|
-
export interface LoadViewerCatalogOptions {
|
|
49
|
-
cwd: string;
|
|
50
|
-
suitesDir: string;
|
|
51
|
-
rubricsDir?: string;
|
|
52
|
-
}
|
|
53
49
|
export interface ViewerRunRequest {
|
|
54
50
|
suite?: string;
|
|
55
51
|
scenario?: string;
|
|
@@ -65,8 +61,6 @@ export interface ViewerJob {
|
|
|
65
61
|
arm?: string;
|
|
66
62
|
prompt: string;
|
|
67
63
|
}
|
|
68
|
-
/** Load suite JSON for the viewer. Omit MCP server env and other secrets. */
|
|
69
|
-
export declare function loadViewerCatalog(options: LoadViewerCatalogOptions): Promise<ViewerCatalog>;
|
|
70
64
|
/** Expand a viewer run into one job per host cell, or per compare arm. */
|
|
71
65
|
export declare function expandViewerJobs(catalog: ViewerCatalog, request: ViewerRunRequest): ViewerJob[];
|
|
72
66
|
//# sourceMappingURL=catalog.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"catalog.d.ts","sourceRoot":"","sources":["../../src/viewer/catalog.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"catalog.d.ts","sourceRoot":"","sources":["../../src/viewer/catalog.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,SAAS,EAAE,WAAW,EAAE,MAAM,2BAA2B,CAAC;AACxE,OAAO,KAAK,EAAE,WAAW,EAAE,kBAAkB,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AACnF,MAAM,WAAW,gBAAgB;IAChC,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,MAAM,CAAC;IACd,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,cAAc,CAAC;IACvB,WAAW,CAAC,EAAE,WAAW,CAAC;IAC1B,iFAAiF;IACjF,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;IAC1B,WAAW,CAAC,EAAE,uBAAuB,EAAE,CAAC;IACxC,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;CAC1B;AAED,MAAM,WAAW,uBAAuB;IACvC,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,EAAE,CAAC;CAChB;AAED,MAAM,WAAW,qBAAqB;IACrC,SAAS,CAAC,EAAE,YAAY,CAAC;IACzB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,IAAI,CAAC,EAAE,SAAS,CAAC;IACjB,WAAW,CAAC,EAAE,WAAW,CAAC;IAC1B,0EAA0E;IAC1E,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,cAAc,CAAC;IACvB,OAAO,CAAC,EAAE,gBAAgB,EAAE,CAAC;IAC7B,KAAK,CAAC,EAAE,WAAW,EAAE,CAAC;IACtB,YAAY,CAAC,EAAE,kBAAkB,EAAE,CAAC;IACpC,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;IAC1B,WAAW,CAAC,EAAE,uBAAuB,EAAE,CAAC;IACxC,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;CAC1B;AAED,MAAM,WAAW,kBAAkB;IAClC,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,KAAK,EAAE,SAAS,EAAE,CAAC;IACnB,SAAS,EAAE,qBAAqB,EAAE,CAAC;CACnC;AAED,MAAM,WAAW,aAAa;IAC7B,SAAS,EAAE,MAAM,CAAC;IAClB,oBAAoB,EAAE,SAAS,EAAE,CAAC;IAClC,MAAM,EAAE,kBAAkB,EAAE,CAAC;CAC7B;AAED,MAAM,WAAW,gBAAgB;IAChC,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,KAAK,CAAC,EAAE,SAAS,EAAE,CAAC;IACpB,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,kDAAkD;IAClD,OAAO,CAAC,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,SAAS;IACzB,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,SAAS,CAAC;IAChB,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,MAAM,CAAC;CACf;AAED,0EAA0E;AAC1E,wBAAgB,gBAAgB,CAAC,OAAO,EAAE,aAAa,EAAE,OAAO,EAAE,gBAAgB,GAAG,SAAS,EAAE,CAmB/F"}
|
package/dist/viewer/catalog.js
CHANGED
|
@@ -1,113 +1,3 @@
|
|
|
1
|
-
import { resolve } from "node:path";
|
|
2
|
-
import { mergeMcpServers, skillPathsFromSetting, } from "@post-print/agent-harness";
|
|
3
|
-
import { applyCompareArm, compareArmDescription, compareArmLabel, resolveCompareArms, } from "../compare-scenario.js";
|
|
4
|
-
import { discoverSuites } from "../discover-suites.js";
|
|
5
|
-
import { resolveSuiteHosts } from "../hosts.js";
|
|
6
|
-
import { loadSuiteFile } from "../load-suite.js";
|
|
7
|
-
function suppliedMcp(defaults, overrides) {
|
|
8
|
-
const servers = mergeMcpServers(defaults, overrides);
|
|
9
|
-
if (!servers)
|
|
10
|
-
return undefined;
|
|
11
|
-
return Object.entries(servers).map(([name, config]) => ({
|
|
12
|
-
name,
|
|
13
|
-
tools: [...(config.tools ?? [])],
|
|
14
|
-
}));
|
|
15
|
-
}
|
|
16
|
-
function suppliedSkills(setting) {
|
|
17
|
-
const paths = skillPathsFromSetting(setting);
|
|
18
|
-
return paths.length > 0 ? paths : undefined;
|
|
19
|
-
}
|
|
20
|
-
/** Load suite JSON for the viewer. Omit MCP server env and other secrets. */
|
|
21
|
-
export async function loadViewerCatalog(options) {
|
|
22
|
-
const suitesDir = resolve(options.cwd, options.suitesDir);
|
|
23
|
-
const suitePaths = await discoverSuites(suitesDir);
|
|
24
|
-
const suites = [];
|
|
25
|
-
for (const suitePath of suitePaths) {
|
|
26
|
-
const suite = await loadSuiteFile(suitePath, { rubricsDir: options.rubricsDir });
|
|
27
|
-
suites.push({
|
|
28
|
-
name: suite.name,
|
|
29
|
-
description: suite.description,
|
|
30
|
-
hosts: resolveSuiteHosts({
|
|
31
|
-
suiteHosts: suite.hosts,
|
|
32
|
-
defaultHost: suite.defaults?.host,
|
|
33
|
-
}),
|
|
34
|
-
scenarios: suite.scenarios.map((scenario) => {
|
|
35
|
-
const arms = resolveCompareArms(scenario.compare).map((entry) => {
|
|
36
|
-
const effective = applyCompareArm(scenario, entry.id);
|
|
37
|
-
const armRubric = { ...effective.rubric };
|
|
38
|
-
delete armRubric.judge;
|
|
39
|
-
const arm = {
|
|
40
|
-
id: entry.id,
|
|
41
|
-
label: compareArmLabel(entry.arm, entry.id),
|
|
42
|
-
prompt: effective.prompt,
|
|
43
|
-
// Compare judge questions are owned by the comparison, even though
|
|
44
|
-
// isolated arm execution temporarily merges them into this rubric.
|
|
45
|
-
rubric: armRubric,
|
|
46
|
-
contextMode: entry.arm.contextMode ??
|
|
47
|
-
scenario.contextMode ??
|
|
48
|
-
suite.defaults?.contextMode ??
|
|
49
|
-
"harness-preamble",
|
|
50
|
-
workspace: effective.workspace ?? ".",
|
|
51
|
-
};
|
|
52
|
-
const description = compareArmDescription(entry.arm);
|
|
53
|
-
if (description) {
|
|
54
|
-
arm.description = description;
|
|
55
|
-
}
|
|
56
|
-
const contextSources = [
|
|
57
|
-
...(suite.defaults?.contextSources ?? []),
|
|
58
|
-
...(effective.contextSources ?? []),
|
|
59
|
-
].filter((value) => typeof value === "string" && value.trim().length > 0);
|
|
60
|
-
if (contextSources.length > 0) {
|
|
61
|
-
arm.contextSources = contextSources;
|
|
62
|
-
}
|
|
63
|
-
arm.suppliedMcp = suppliedMcp(suite.defaults?.mcpServers, effective.mcpServers);
|
|
64
|
-
arm.suppliedSkills = suppliedSkills(effective.skills ?? suite.defaults?.skills);
|
|
65
|
-
return arm;
|
|
66
|
-
});
|
|
67
|
-
const row = {
|
|
68
|
-
name: scenario.name,
|
|
69
|
-
prompt: scenario.prompt,
|
|
70
|
-
rubric: scenario.rubric,
|
|
71
|
-
contextMode: scenario.contextMode ?? suite.defaults?.contextMode ?? "harness-preamble",
|
|
72
|
-
workspace: scenario.workspace ?? suite.defaults?.workspace ?? ".",
|
|
73
|
-
};
|
|
74
|
-
if (scenario.description) {
|
|
75
|
-
row.description = scenario.description;
|
|
76
|
-
}
|
|
77
|
-
if (scenario.skip) {
|
|
78
|
-
row.skip = true;
|
|
79
|
-
}
|
|
80
|
-
if (scenario.host) {
|
|
81
|
-
row.host = scenario.host;
|
|
82
|
-
}
|
|
83
|
-
const contextSources = [
|
|
84
|
-
...(suite.defaults?.contextSources ?? []),
|
|
85
|
-
...(scenario.contextSources ?? []),
|
|
86
|
-
].filter((value) => typeof value === "string" && value.trim().length > 0);
|
|
87
|
-
if (contextSources.length > 0) {
|
|
88
|
-
row.contextSources = contextSources;
|
|
89
|
-
}
|
|
90
|
-
row.suppliedMcp = suppliedMcp(suite.defaults?.mcpServers, scenario.mcpServers);
|
|
91
|
-
row.suppliedSkills = suppliedSkills(scenario.skills ?? suite.defaults?.skills);
|
|
92
|
-
if (arms.length > 0) {
|
|
93
|
-
row.compare = arms;
|
|
94
|
-
if (scenario.compare?.gates?.length)
|
|
95
|
-
row.gates = scenario.compare.gates;
|
|
96
|
-
if (scenario.compare?.judgeMetrics?.length) {
|
|
97
|
-
row.judgeMetrics = scenario.compare.judgeMetrics;
|
|
98
|
-
}
|
|
99
|
-
}
|
|
100
|
-
return row;
|
|
101
|
-
}),
|
|
102
|
-
});
|
|
103
|
-
}
|
|
104
|
-
suites.sort((left, right) => left.name.localeCompare(right.name));
|
|
105
|
-
return {
|
|
106
|
-
suitesDir,
|
|
107
|
-
defaultSelectedHosts: [],
|
|
108
|
-
suites,
|
|
109
|
-
};
|
|
110
|
-
}
|
|
111
1
|
/** Expand a viewer run into one job per host cell, or per compare arm. */
|
|
112
2
|
export function expandViewerJobs(catalog, request) {
|
|
113
3
|
const selectedHosts = request.hosts ?? catalog.defaultSelectedHosts;
|
|
@@ -124,30 +14,35 @@ export function expandViewerJobs(catalog, request) {
|
|
|
124
14
|
if (scenario.skip) {
|
|
125
15
|
continue;
|
|
126
16
|
}
|
|
127
|
-
|
|
128
|
-
|
|
129
|
-
|
|
130
|
-
|
|
131
|
-
|
|
132
|
-
|
|
133
|
-
|
|
134
|
-
|
|
135
|
-
|
|
136
|
-
|
|
137
|
-
|
|
138
|
-
|
|
139
|
-
|
|
140
|
-
}
|
|
141
|
-
continue;
|
|
142
|
-
}
|
|
17
|
+
jobs.push(...scenarioJobs(suite.name, scenario, hosts));
|
|
18
|
+
}
|
|
19
|
+
}
|
|
20
|
+
return jobs;
|
|
21
|
+
}
|
|
22
|
+
function scenarioJobs(suite, scenario, hosts) {
|
|
23
|
+
const jobs = [];
|
|
24
|
+
for (const host of hosts) {
|
|
25
|
+
if (scenario.host && scenario.host !== host) {
|
|
26
|
+
continue;
|
|
27
|
+
}
|
|
28
|
+
if (scenario.compare?.length) {
|
|
29
|
+
for (const arm of scenario.compare) {
|
|
143
30
|
jobs.push({
|
|
144
|
-
suite: suite
|
|
31
|
+
suite: suite,
|
|
145
32
|
scenario: scenario.name,
|
|
146
33
|
host,
|
|
147
|
-
|
|
34
|
+
arm: arm.id,
|
|
35
|
+
prompt: arm.prompt ?? scenario.prompt,
|
|
148
36
|
});
|
|
149
37
|
}
|
|
38
|
+
continue;
|
|
150
39
|
}
|
|
40
|
+
jobs.push({
|
|
41
|
+
suite: suite,
|
|
42
|
+
scenario: scenario.name,
|
|
43
|
+
host,
|
|
44
|
+
prompt: scenario.prompt,
|
|
45
|
+
});
|
|
151
46
|
}
|
|
152
47
|
return jobs;
|
|
153
48
|
}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"catalog.js","sourceRoot":"","sources":["../../src/viewer/catalog.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"catalog.js","sourceRoot":"","sources":["../../src/viewer/catalog.ts"],"names":[],"mappings":"AAsEA,0EAA0E;AAC1E,MAAM,UAAU,gBAAgB,CAAC,OAAsB,EAAE,OAAyB;IACjF,MAAM,aAAa,GAAG,OAAO,CAAC,KAAK,IAAI,OAAO,CAAC,oBAAoB,CAAC;IACpE,MAAM,MAAM,GAAG,OAAO,CAAC,KAAK;QAC3B,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,KAAK,OAAO,CAAC,KAAK,CAAC;QAChE,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC;IAClB,MAAM,IAAI,GAAgB,EAAE,CAAC;IAC7B,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;QAC5B,MAAM,KAAK,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,KAAK,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC;QACzE,MAAM,SAAS,GAAG,OAAO,CAAC,QAAQ;YACjC,CAAC,CAAC,KAAK,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,KAAK,OAAO,CAAC,QAAQ,CAAC;YAC1E,CAAC,CAAC,KAAK,CAAC,SAAS,CAAC;QACnB,KAAK,MAAM,QAAQ,IAAI,SAAS,EAAE,CAAC;YAClC,IAAI,QAAQ,CAAC,IAAI,EAAE,CAAC;gBACnB,SAAS;YACV,CAAC;YACD,IAAI,CAAC,IAAI,CAAC,GAAG,YAAY,CAAC,KAAK,CAAC,IAAI,EAAE,QAAQ,EAAE,KAAK,CAAC,CAAC,CAAC;QACzD,CAAC;IACF,CAAC;IACD,OAAO,IAAI,CAAC;AACb,CAAC;AAED,SAAS,YAAY,CACpB,KAAa,EACb,QAA+B,EAC/B,KAAkB;IAElB,MAAM,IAAI,GAAgB,EAAE,CAAC;IAC7B,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QAC1B,IAAI,QAAQ,CAAC,IAAI,IAAI,QAAQ,CAAC,IAAI,KAAK,IAAI,EAAE,CAAC;YAC7C,SAAS;QACV,CAAC;QACD,IAAI,QAAQ,CAAC,OAAO,EAAE,MAAM,EAAE,CAAC;YAC9B,KAAK,MAAM,GAAG,IAAI,QAAQ,CAAC,OAAO,EAAE,CAAC;gBACpC,IAAI,CAAC,IAAI,CAAC;oBACT,KAAK,EAAE,KAAK;oBACZ,QAAQ,EAAE,QAAQ,CAAC,IAAI;oBACvB,IAAI;oBACJ,GAAG,EAAE,GAAG,CAAC,EAAE;oBACX,MAAM,EAAE,GAAG,CAAC,MAAM,IAAI,QAAQ,CAAC,MAAM;iBACrC,CAAC,CAAC;YACJ,CAAC;YACD,SAAS;QACV,CAAC;QACD,IAAI,CAAC,IAAI,CAAC;YACT,KAAK,EAAE,KAAK;YACZ,QAAQ,EAAE,QAAQ,CAAC,IAAI;YACvB,IAAI;YACJ,MAAM,EAAE,QAAQ,CAAC,MAAM;SACvB,CAAC,CAAC;IACJ,CAAC;IACD,OAAO,IAAI,CAAC;AACb,CAAC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"client.d.ts","sourceRoot":"","sources":["../../src/viewer/client.tsx"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAKX,cAAc,EAId,MAAM,aAAa,CAAC;AAOrB,OAAO,KAAK,EACX,eAAe,EAIf,MAAM,aAAa,CAAC;
|
|
1
|
+
{"version":3,"file":"client.d.ts","sourceRoot":"","sources":["../../src/viewer/client.tsx"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAKX,cAAc,EAId,MAAM,aAAa,CAAC;AAOrB,OAAO,KAAK,EACX,eAAe,EAIf,MAAM,aAAa,CAAC;AA8gCrB,wBAAgB,sBAAsB,CAAC,MAAM,EAAE,cAAc,GAAG,MAAM,CAuBrE;AAgvBD,MAAM,CAAC,OAAO,UAAU,SAAS,CAAC,EAAE,SAAS,EAAE,EAAE;IAAE,SAAS,EAAE,eAAe,CAAA;CAAE,+BAK9E"}
|