@cogitator-ai/evals 0.13.2 → 0.14.0

package/dist/plugin-evals/types.d.ts

@@ -0,0 +1,378 @@
+import type { CogitatorConfig, LocalWorkspaceExecConfig } from '@cogitator-ai/core';
+/**
+ * Whether each case also runs without the plugin: `none` runs the with-arm only, `with-without`
+ * adds the no-plugin baseline to every case. Unset, it is decided per case
+ */
+export type PluginEvalAblation = 'none' | 'with-without';
+/** The runs of a case with the plugin loaded (`with`) or with no plugin (`without`) */
+export type PluginEvalArm = 'with' | 'without';
+/**
+ * What a `regex` grader matches or an `llm` grader judges: the final reply, the session as JSON
+ * lines, the paths created during the run, the contents of one workspace file, or the calls made
+ * to mocked MCP tools
+ */
+export type PluginEvalTarget = 'last_message' | 'trace' | 'files' | 'mock_calls' | {
+    source: 'file';
+    path: string;
+};
+/** A tool call a `tool_order` grader looks for: a tool name, or a name with an input regex */
+export interface PluginEvalToolMatcher {
+    tool: string;
+    inputMatch?: string;
+}
+interface PluginEvalGraderBase {
+    /** The file name without `.md`, or the `name` of a `case.yaml` entry */
+    name: string;
+    /** Relative weight in the run's score, any positive number */
+    weight: number;
+    /**
+     * `with-only` keeps the grader out of the score of a two-arm run, `both` scores it in both arms
+     * even when it would be excluded
+     */
+    arm?: 'with-only' | 'both';
+}
+export interface PluginEvalRegexGrader extends PluginEvalGraderBase {
+    type: 'regex';
+    pattern: string;
+    flags: string;
+    /** `contains`, `not_contains` or `count:N` for exactly N matches */
+    match: string;
+    target: PluginEvalTarget;
+}
+export interface PluginEvalToolUsedGrader extends PluginEvalGraderBase {
+    type: 'tool_used';
+    /** Claude Code's name of the tool (`Read`, `Skill`, `mcp__plugin_p_s__t`) or Cogitator's own */
+    tool: string;
+    inputMatch?: string;
+    min: number;
+    max?: number;
+}
+export interface PluginEvalToolOrderGrader extends PluginEvalGraderBase {
+    type: 'tool_order';
+    before: PluginEvalToolMatcher;
+    after: PluginEvalToolMatcher;
+}
+export interface PluginEvalFileExistsGrader extends PluginEvalGraderBase {
+    type: 'file_exists';
+    /** A glob over the paths created during the run, relative to the workspace */
+    path: string;
+    exists: boolean;
+}
+export interface PluginEvalLlmGrader extends PluginEvalGraderBase {
+    type: 'llm';
+    criteria: string;
+    focus: PluginEvalTarget;
+}
+export interface PluginEvalBaselineGrader extends PluginEvalGraderBase {
+    type: 'baseline';
+    /** The reference transcript, a `.jsonl` file of the case directory */
+    baselineFile: string;
+    criteria: string;
+}
+/** One check applied after a run, in the format of Claude Code's plugin eval graders */
+export type PluginEvalGrader = PluginEvalRegexGrader | PluginEvalToolUsedGrader | PluginEvalToolOrderGrader | PluginEvalFileExistsGrader | PluginEvalLlmGrader | PluginEvalBaselineGrader;
+/** The type of an `expect:` check of a mocked tool's input */
+export type PluginEvalExpectType = 'string' | 'number' | 'boolean' | 'array' | 'object' | 'null';
+/** How one input field of a mocked tool call is checked */
+export type PluginEvalExpectation = {
+    kind: 'type';
+    type: PluginEvalExpectType;
+} | {
+    kind: 'pattern';
+    source: string;
+    flags: string;
+} | {
+    kind: 'literal';
+    value: string | number | boolean | null;
+} | {
+    kind: 'oneOf';
+    values: readonly (string | number | boolean | null)[];
+};
+/** A mocked MCP tool: `mocks/<server>/<tool>.md` */
+export interface PluginEvalMockTool {
+    server: string;
+    tool: string;
+    /** The tool result, with `{{input.<field>}}` and `{{file:fixtures/...}}` substitutions */
+    body: string;
+    /** Return the body as a tool error */
+    error: boolean;
+    /** Checks of the call's input by dotted path, a violation aborts the run */
+    expect: Readonly<Record<string, PluginEvalExpectation>>;
+    /** The server's mock directory, where `fixtures/` lives */
+    dir: string;
+    description?: string;
+    /** The input JSON Schema from the server's `_tools.json` */
+    inputSchema?: Readonly<Record<string, unknown>>;
+}
+/** A case as it was loaded: its prompt, its run settings and its graders */
+export interface PluginEvalCase {
+    name: string;
+    /** The case directory */
+    dir: string;
+    /** The case directory relative to the eval directory */
+    path: string;
+    description?: string;
+    expectedOutcome?: string;
+    tags: string[];
+    /** Plugin directories under test, absolute */
+    plugins: string[];
+    runs: number;
+    model?: string;
+    maxTurns: number;
+    timeoutSeconds: number;
+    allowedTools: string[];
+    appendSystemPrompt?: string;
+    env: Record<string, string>;
+    prompt: string;
+    context: {
+        /** The scaffold script, absolute */
+        scaffoldScript?: string;
+        /** The transcript to resume, absolute */
+        historyFile?: string;
+        /** Read-only fixture directories, relative to the case directory */
+        addDirs: string[];
+    };
+    graders: PluginEvalGrader[];
+    /** Mocked tools by server, the case's own over the suite's */
+    mocks: Record<string, PluginEvalMockTool[]>;
+    /**
+     * Why a suite mock server cannot be used, by server: it fails the case only when the case uses
+     * the server, that is when one of its plugins declares it
+     */
+    mockProblems: Record<string, string>;
+}
+/** A judge's vote on an `llm` or `baseline` grader */
+export interface PluginEvalVote {
+    verdict: 'PASS' | 'FAIL' | 'ERROR';
+    reasoning?: string;
+}
+export interface PluginEvalGraderResult {
+    name: string;
+    type: PluginEvalGrader['type'];
+    passed: boolean;
+    /** Whether the grader counts toward the run's score: false for plugin-fired indicators */
+    scored: boolean;
+    weight: number;
+    explanation: string;
+    /** The judge's votes, for `llm` and `baseline` graders */
+    votes?: PluginEvalVote[];
+    /** What the judge was shown */
+    evidence?: string;
+    /** The cost ceiling skipped this judge grader */
+    skipped?: boolean;
+    costUsd?: number;
+}
+/** A tool call of a run, under Claude Code's name and input shape */
+export interface PluginEvalToolCall {
+    id: string;
+    /** Claude Code's name, such as `Read`, `Skill` or `mcp__plugin_p_s__t` */
+    name: string;
+    /** The Cogitator tool that ran */
+    tool: string;
+    input: Record<string, unknown>;
+    /** The delegated agent that made the call, unset for a call of the agent under test */
+    agent?: string;
+}
+/** A call to a mocked MCP tool */
+export interface PluginEvalMockCall {
+    server: string;
+    tool: string;
+    input: Record<string, unknown>;
+    output: string;
+    error: boolean;
+}
+/** Why a mock stopped a run */
+export interface PluginEvalAbort {
+    server: string;
+    tool: string;
+    reason: string;
+}
+export interface PluginEvalRunResult {
+    arm: PluginEvalArm;
+    /** The run's number within its arm, from 1 */
+    index: number;
+    /** Weighted fraction of the scored graders that passed */
+    score: number;
+    /** `null`, or why the run ended abnormally */
+    error: string | null;
+    /** The run ended on a model provider error and scored 0 without grading */
+    apiError?: boolean;
+    /** A mock stopped the run: it scores 0 and `error` stays `null` */
+    aborted?: PluginEvalAbort;
+    /** The cost ceiling skipped this run's judge graders, so its score is not comparable */
+    skippedPaidGraders?: boolean;
+    graders: PluginEvalGraderResult[];
+    lastMessage: string;
+    toolCalls: PluginEvalToolCall[];
+    mockCalls: PluginEvalMockCall[];
+    /** Paths created during the run, relative to the workspace */
+    filesCreated: string[];
+    /** List-price estimate of the run's model calls and its judge calls */
+    costUsd: number;
+    durationSeconds: number;
+    /** The run's temporary directory, kept with `keepTemp` */
+    tempDir?: string;
+}
+export interface PluginEvalCaseAggregates {
+    /** Mean with-arm run score, `null` when no with-arm run happened */
+    score: number | null;
+    /** Mean without-arm run score, absent for a single-arm case */
+    withoutScore?: number | null;
+    /** With-arm score minus without-arm score, absent when the arms are not comparable */
+    delta?: number;
+    passed: boolean;
+    runs: number;
+    costUsd: number;
+}
+export interface PluginEvalCaseResult {
+    name: string;
+    /** The case directory relative to the eval directory */
+    path: string;
+    tags: string[];
+    /** Why the case did not load: it scores 0 and does not run */
+    error?: string;
+    /** The case ran its with-arm only */
+    singleArm: boolean;
+    notices: string[];
+    prompt: string;
+    graders: PluginEvalGrader[];
+    arms: {
+        with: PluginEvalRunResult[];
+        without: PluginEvalRunResult[];
+    };
+    aggregates: PluginEvalCaseAggregates;
+}
+/** Why a run did not finish every case */
+export type PluginEvalPartialReason = 'cost_ceiling' | 'interrupted' | 'auth_failed';
+/** How the run ended, which the CLI turns into its exit code */
+export type PluginEvalOutcome = 'passed' | 'failed' | 'partial' | 'interrupted';
+/**
+ * The result document of a plugin eval run, `aggregate-result.json`: versioned, camelCase, new
+ * fields added without renaming the ones there
+ */
+export interface PluginEvalResult {
+    schemaVersion: 1;
+    /** The version of `@cogitator-ai/core` that ran the suite */
+    cogitatorVersion: string;
+    plugin: {
+        name: string;
+        root: string;
+        version?: string;
+    };
+    evalDir: string;
+    startedAt: string;
+    outcome: PluginEvalOutcome;
+    partial: boolean;
+    partialReason: PluginEvalPartialReason | null;
+    config: {
+        runs?: number;
+        concurrency: number;
+        model?: string;
+        judgeModel?: string;
+        ablation?: PluginEvalAblation;
+        threshold: number;
+        maxCostUsd?: number;
+        allowTools: string[];
+        scaffold: boolean;
+        caseFilter?: string;
+        tagFilter: string[];
+    };
+    aggregates: {
+        /** Mean case score across the suite */
+        overallScore: number;
+        casesPassed: number;
+        casesTotal: number;
+        /** Mean delta across the cases that have one */
+        meanDelta?: number;
+        /** Mean without-arm score across the cases that have one */
+        baselineScore?: number;
+        /** Share of with-arm runs where every grader passed */
+        perfectRuns: number;
+    };
+    /** Runs that ended on a model provider error, which fail the command whatever the scores */
+    apiErrors: number;
+    /** Cases that failed to load */
+    loadErrors: number;
+    warnings: string[];
+    cases: PluginEvalCaseResult[];
+    costUsd: number;
+    durationSeconds: number;
+}
+/** What a plugin eval run reports while it works */
+export type PluginEvalProgressEvent = {
+    type: 'warning';
+    message: string;
+} | {
+    type: 'case';
+    case: string;
+    notices: readonly string[];
+    arms: readonly PluginEvalArm[];
+} | {
+    type: 'run';
+    case: string;
+    run: PluginEvalRunResult;
+    of: number;
+} | {
+    type: 'skipped';
+    case: string;
+    reason: PluginEvalPartialReason;
+};
+export interface PluginEvalOptions {
+    /**
+     * A plugin directory, a `prompt.md` or `case.yaml` file, or an installed plugin as `name` or
+     * `name@marketplace` (default: the working directory)
+     */
+    target?: string;
+    /** The directory a relative target and a named plugin's results start from (default `process.cwd()`) */
+    cwd?: string;
+    /** The LLM configuration of the runs and the judge: providers, keys, backends, default model */
+    llm: CogitatorConfig['llm'];
+    /** Model of the agent under test, over each case's `model` and `llm.defaultModel` */
+    model?: string;
+    /** Model of `llm` and `baseline` graders (default: each case's agent model) */
+    judgeModel?: string;
+    /** Runs per case per arm, 1 to 50, over each case's `runs` */
+    runs?: number;
+    /** Agent runs at once, 1 to 8 (default 1) */
+    concurrency?: number;
+    ablation?: PluginEvalAblation;
+    /** A case passes when its with-arm score is at least this, 0 to 1 (default 1) */
+    threshold?: number;
+    /** Ceiling on the list-price cost estimate: once spent, no further run starts */
+    maxCostUsd?: number;
+    /** Tools granted beyond the read-only set, as permission rules: `Write`, `Bash(npm test *)` */
+    allowTools?: readonly string[];
+    /** Run each case's `context.scaffold_script` */
+    scaffold?: boolean;
+    /** A glob the case names must match */
+    caseGlob?: string;
+    /** Keep the cases with any of these tags */
+    tags?: readonly string[];
+    /** The eval directory below the plugin, relative and of plain directory names */
+    evalDir?: string;
+    /** Where `aggregate-result.json` and `report.html` go (default `<evalDir>/results/<timestamp>/`) */
+    outputDir?: string;
+    /** Keep every run's temporary directory */
+    keepTemp?: boolean;
+    /**
+     * How granted `Bash` commands are spawned in a run's workspace, such as a `launch` that wraps
+     * them in an OS sandbox. Commands otherwise run as this process, confined to the workspace
+     * directory only
+     */
+    exec?: Omit<LocalWorkspaceExecConfig, 'env'>;
+    /** Where installed plugins are kept, for a named target (default `defaultPluginsRoot()`) */
+    pluginsRoot?: string;
+    /** Stops starting runs and aborts the running ones: the result is partial, `interrupted` */
+    signal?: AbortSignal;
+    onProgress?: (event: PluginEvalProgressEvent) => void;
+}
+/** What `experimental_runPluginEvals` returns: the result document and where it was written */
+export interface PluginEvalReport {
+    result: PluginEvalResult;
+    /** The directory of `aggregate-result.json` and `report.html` */
+    outputDir: string;
+    resultFile: string;
+    reportFile: string;
+}
+export {};
+//# sourceMappingURL=types.d.ts.map

package/dist/plugin-evals/types.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/plugin-evals/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,eAAe,EAAE,wBAAwB,EAAE,MAAM,oBAAoB,CAAC;AAEpF;;;GAGG;AACH,MAAM,MAAM,kBAAkB,GAAG,MAAM,GAAG,cAAc,CAAC;AAEzD,uFAAuF;AACvF,MAAM,MAAM,aAAa,GAAG,MAAM,GAAG,SAAS,CAAC;AAE/C;;;;GAIG;AACH,MAAM,MAAM,gBAAgB,GAC1B,cAAc,GAAG,OAAO,GAAG,OAAO,GAAG,YAAY,GAAG;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,MAAM,CAAA;CAAE,CAAC;AAEvF,8FAA8F;AAC9F,MAAM,WAAW,qBAAqB;IACpC,IAAI,EAAE,MAAM,CAAC;IACb,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,UAAU,oBAAoB;IAC5B,wEAAwE;IACxE,IAAI,EAAE,MAAM,CAAC;IACb,8DAA8D;IAC9D,MAAM,EAAE,MAAM,CAAC;IACf;;;OAGG;IACH,GAAG,CAAC,EAAE,WAAW,GAAG,MAAM,CAAC;CAC5B;AAED,MAAM,WAAW,qBAAsB,SAAQ,oBAAoB;IACjE,IAAI,EAAE,OAAO,CAAC;IACd,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;IACd,oEAAoE;IACpE,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,gBAAgB,CAAC;CAC1B;AAED,MAAM,WAAW,wBAAyB,SAAQ,oBAAoB;IACpE,IAAI,EAAE,WAAW,CAAC;IAClB,gGAAgG;IAChG,IAAI,EAAE,MAAM,CAAC;IACb,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,GAAG,EAAE,MAAM,CAAC;IACZ,GAAG,CAAC,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,yBAA0B,SAAQ,oBAAoB;IACrE,IAAI,EAAE,YAAY,CAAC;IACnB,MAAM,EAAE,qBAAqB,CAAC;IAC9B,KAAK,EAAE,qBAAqB,CAAC;CAC9B;AAED,MAAM,WAAW,0BAA2B,SAAQ,oBAAoB;IACtE,IAAI,EAAE,aAAa,CAAC;IACpB,8EAA8E;IAC9E,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,OAAO,CAAC;CACjB;AAED,MAAM,WAAW,mBAAoB,SAAQ,oBAAoB;IAC/D,IAAI,EAAE,KAAK,CAAC;IACZ,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,gBAAgB,CAAC;CACzB;AAED,MAAM,WAAW,wBAAyB,SAAQ,oBAAoB;IACpE,IAAI,EAAE,UAAU,CAAC;IACjB,sEAAsE;IACtE,YAAY,EAAE,MAAM,CAAC;IACrB,QAAQ,EAAE,MAAM,CAAC;CAClB;AAED,wFAAwF;AACxF,MAAM,MAAM,gBAAgB,GACxB,qBAAqB,GACrB,wBAAwB,GACxB,yBAAyB,GACzB,0BAA0B,GAC1B,mBAAmB,GACnB,wBAAwB,CAAC;AAE7B,8DAA8D;AAC9D,MAAM,MAAM,oBAAoB,GAAG,QAAQ,GAAG,QAAQ,GAAG,SAAS,GAAG,OAAO,GAAG,QAAQ,GAAG,MAAM,CAAC;AAEjG,2DAA2D;AAC3D,MAAM,MAAM,qBAAqB,GAC7B;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,oBAAoB,CAAA;CAAE,GAC5C;IAAE,IAAI,EAAE,SAAS,CAAC;IAAC,MAAM,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,GAClD;IAAE,IAAI,EAAE,SAAS,CAAC;IAAC,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,OAAO,GAAG,IAAI,CAAA;CAAE,GAC5D;IAAE,IAAI,EAAE,OAAO,CAAC;IAAC,MAAM,EAAE,SAAS,CAAC,MAAM,GAAG,MAAM,GAAG,OAAO,GAAG,IAAI,CAAC,EAAE,CAAA;CAAE,CAAC;AAE7E,oDAAoD;AACpD,MAAM,WAAW,kBAAkB;IACjC,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,0FAA0F;IAC1F,IAAI,EAAE,MAAM,CAAC;IACb,sCAAsC;IACtC,KAAK,EAAE,OAAO,CAAC;IACf,4EAA4E;IAC5E,MAAM,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,qBAAqB,CAAC,CAAC,CAAC;IACxD,2DAA2D;IAC3D,GAAG,EAAE,MAAM,CAAC;IACZ,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,4DAA4D;IAC5D,WAAW,CAAC,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC;CACjD;AAED,4EAA4E;AAC5E,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE,MAAM,CAAC;IACb,yBAAyB;IACzB,GAAG,EAAE,MAAM,CAAC;IACZ,wDAAwD;IACxD,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,IAAI,EAAE,MAAM,EAAE,CAAC;IACf,8CAA8C;IAC9C,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,cAAc,EAAE,MAAM,CAAC;IACvB,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,GAAG,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC5B,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE;QACP,oCAAoC;QACpC,cAAc,CAAC,EAAE,MAAM,CAAC;QACxB,yCAAyC;QACzC,WAAW,CAAC,EAAE,MAAM,CAAC;QACrB,oEAAoE;QACpE,OAAO,EAAE,MAAM,EAAE,CAAC;KACnB,CAAC;IACF,OAAO,EAAE,gBAAgB,EAAE,CAAC;IAC5B,8DAA8D;IAC9D,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,kBAAkB,EAAE,CAAC,CAAC;IAC5C;;;OAGG;IACH,YAAY,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CACtC;AAED,sDAAsD;AACtD,MAAM,WAAW,cAAc;IAC7B,OAAO,EAAE,MAAM,GAAG,MAAM,GAAG,OAAO,CAAC;IACnC,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,sBAAsB;IACrC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,gBAAgB,CAAC,MAAM,CAAC,CAAC;IAC/B,MAAM,EAAE,OAAO,CAAC;IAChB,0FAA0F;IAC1F,MAAM,EAAE,OAAO,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,EAAE,MAAM,CAAC;IACpB,0DAA0D;IAC1D,KAAK,CAAC,EAAE,cAAc,EAAE,CAAC;IACzB,+BAA+B;IAC/B,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,iDAAiD;IACjD,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB;AAED,qEAAqE;AACrE,MAAM,WAAW,kBAAkB;IACjC,EAAE,EAAE,MAAM,CAAC;IACX,0EAA0E;IAC1E,IAAI,EAAE,MAAM,CAAC;IACb,kCAAkC;IAClC,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAC/B,uFAAuF;IACvF,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,kCAAkC;AAClC,MAAM,WAAW,kBAAkB;IACjC,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAC/B,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,OAAO,CAAC;CAChB;AAED,+BAA+B;AAC/B,MAAM,WAAW,eAAe;IAC9B,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,mBAAmB;IAClC,GAAG,EAAE,aAAa,CAAC;IACnB,8CAA8C;IAC9C,KAAK,EAAE,MAAM,CAAC;IACd,0DAA0D;IAC1D,KAAK,EAAE,MAAM,CAAC;IACd,8CAA8C;IAC9C,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,2EAA2E;IAC3E,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,mEAAmE;IACnE,OAAO,CAAC,EAAE,eAAe,CAAC;IAC1B,wFAAwF;IACxF,kBAAkB,CAAC,EAAE,OAAO,CAAC;IAC7B,OAAO,EAAE,sBAAsB,EAAE,CAAC;IAClC,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,EAAE,kBAAkB,EAAE,CAAC;IAChC,SAAS,EAAE,kBAAkB,EAAE,CAAC;IAChC,8DAA8D;IAC9D,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,uEAAuE;IACvE,OAAO,EAAE,MAAM,CAAC;IAChB,eAAe,EAAE,MAAM,CAAC;IACxB,0DAA0D;IAC1D,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,wBAAwB;IACvC,oEAAoE;IACpE,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,+DAA+D;IAC/D,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,sFAAsF;IACtF,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,OAAO,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,oBAAoB;IACnC,IAAI,EAAE,MAAM,CAAC;IACb,wDAAwD;IACxD,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,EAAE,CAAC;IACf,8DAA8D;IAC9D,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,qCAAqC;IACrC,SAAS,EAAE,OAAO,CAAC;IACnB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,gBAAgB,EAAE,CAAC;IAC5B,IAAI,EAAE;QAAE,IAAI,EAAE,mBAAmB,EAAE,CAAC;QAAC,OAAO,EAAE,mBAAmB,EAAE,CAAA;KAAE,CAAC;IACtE,UAAU,EAAE,wBAAwB,CAAC;CACtC;AAED,0CAA0C;AAC1C,MAAM,MAAM,uBAAuB,GAAG,cAAc,GAAG,aAAa,GAAG,aAAa,CAAC;AAErF,gEAAgE;AAChE,MAAM,MAAM,iBAAiB,GAAG,QAAQ,GAAG,QAAQ,GAAG,SAAS,GAAG,aAAa,CAAC;AAEhF;;;GAGG;AACH,MAAM,WAAW,gBAAgB;IAC/B,aAAa,EAAE,CAAC,CAAC;IACjB,6DAA6D;IAC7D,gBAAgB,EAAE,MAAM,CAAC;IACzB,MAAM,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;IACzD,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,iBAAiB,CAAC;IAC3B,OAAO,EAAE,OAAO,CAAC;IACjB,aAAa,EAAE,uBAAuB,GAAG,IAAI,CAAC;IAC9C,MAAM,EAAE;QACN,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,WAAW,EAAE,MAAM,CAAC;QACpB,KAAK,CAAC,EAAE,MAAM,CAAC;QACf,UAAU,CAAC,EAAE,MAAM,CAAC;QACpB,QAAQ,CAAC,EAAE,kBAAkB,CAAC;QAC9B,SAAS,EAAE,MAAM,CAAC;QAClB,UAAU,CAAC,EAAE,MAAM,CAAC;QACpB,UAAU,EAAE,MAAM,EAAE,CAAC;QACrB,QAAQ,EAAE,OAAO,CAAC;QAClB,UAAU,CAAC,EAAE,MAAM,CAAC;QACpB,SAAS,EAAE,MAAM,EAAE,CAAC;KACrB,CAAC;IACF,UAAU,EAAE;QACV,uCAAuC;QACvC,YAAY,EAAE,MAAM,CAAC;QACrB,WAAW,EAAE,MAAM,CAAC;QACpB,UAAU,EAAE,MAAM,CAAC;QACnB,gDAAgD;QAChD,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,4DAA4D;QAC5D,aAAa,CAAC,EAAE,MAAM,CAAC;QACvB,uDAAuD;QACvD,WAAW,EAAE,MAAM,CAAC;KACrB,CAAC;IACF,4FAA4F;IAC5F,SAAS,EAAE,MAAM,CAAC;IAClB,gCAAgC;IAChC,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,KAAK,EAAE,oBAAoB,EAAE,CAAC;IAC9B,OAAO,EAAE,MAAM,CAAC;IAChB,eAAe,EAAE,MAAM,CAAC;CACzB;AAED,oDAAoD;AACpD,MAAM,MAAM,uBAAuB,GAC/B;IAAE,IAAI,EAAE,SAAS,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,GACpC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,SAAS,MAAM,EAAE,CAAC;IAAC,IAAI,EAAE,SAAS,aAAa,EAAE,CAAA;CAAE,GAC1F;IAAE,IAAI,EAAE,KAAK,CAAC;IAAC,IAAI,EAAE,MAAM,CAAC;IAAC,GAAG,EAAE,mBAAmB,CAAC;IAAC,EAAE,EAAE,MAAM,CAAA;CAAE,GACnE;IAAE,IAAI,EAAE,SAAS,CAAC;IAAC,IAAI,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,uBAAuB,CAAA;CAAE,CAAC;AAEvE,MAAM,WAAW,iBAAiB;IAChC;;;OAGG;IACH,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,wGAAwG;IACxG,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,gGAAgG;IAChG,GAAG,EAAE,eAAe,CAAC,KAAK,CAAC,CAAC;IAC5B,qFAAqF;IACrF,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,+EAA+E;IAC/E,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,8DAA8D;IAC9D,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,6CAA6C;IAC7C,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,QAAQ,CAAC,EAAE,kBAAkB,CAAC;IAC9B,iFAAiF;IACjF,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,iFAAiF;IACjF,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,+FAA+F;IAC/F,UAAU,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;IAC/B,gDAAgD;IAChD,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,uCAAuC;IACvC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,4CAA4C;IAC5C,IAAI,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;IACzB,iFAAiF;IACjF,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,oGAAoG;IACpG,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,2CAA2C;IAC3C,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB;;;;OAIG;IACH,IAAI,CAAC,EAAE,IAAI,CAAC,wBAAwB,EAAE,KAAK,CAAC,CAAC;IAC7C,4FAA4F;IAC5F,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,4FAA4F;IAC5F,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,UAAU,CAAC,EAAE,CAAC,KAAK,EAAE,uBAAuB,KAAK,IAAI,CAAC;CACvD;AAED,+FAA+F;AAC/F,MAAM,WAAW,gBAAgB;IAC/B,MAAM,EAAE,gBAAgB,CAAC;IACzB,iEAAiE;IACjE,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;CACpB"}

package/dist/plugin-evals/types.js

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=types.js.map

package/dist/plugin-evals/types.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../../src/plugin-evals/types.ts"],"names":[],"mappings":""}

package/dist/reporters/html-format.d.ts

@@ -2,8 +2,11 @@
 export declare function escapeHtml(value: string): string;
 /** Text that may be long or span lines: as it is when short, folded in `<details>` otherwise */
 export declare function foldable(value: string, max?: number): string;
-/** A self-contained HTML page: inline CSS, no scripts and no external assets */
-export declare function htmlDocument(title: string, body: string): string;
+/**
+ * A self-contained HTML page: inline CSS, no scripts and no external assets. `style` adds rules
+ * of the page's own after the shared ones
+ */
+export declare function htmlDocument(title: string, body: string, style?: string): string;
 /** A table, its cells already HTML; `numeric` marks the columns aligned right */
 export declare function htmlTable(headers: readonly string[], rows: readonly (readonly string[])[], numeric?: ReadonlySet<number>): string;
 /** A row of figures, each value already HTML */

package/dist/reporters/html-format.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"html-format.d.ts","sourceRoot":"","sources":["../../src/reporters/html-format.ts"],"names":[],"mappings":"AAAA,yDAAyD;AACzD,wBAAgB,UAAU,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,CAOhD;AAED,gGAAgG;AAChG,wBAAgB,QAAQ,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,GAAE,MAAY,GAAG,MAAM,CAQjE;AAyDD,gFAAgF;AAChF,wBAAgB,YAAY,CAAC,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,MAAM,CAkBhE;AAED,iFAAiF;AACjF,wBAAgB,SAAS,CACvB,OAAO,EAAE,SAAS,MAAM,EAAE,EAC1B,IAAI,EAAE,SAAS,CAAC,SAAS,MAAM,EAAE,CAAC,EAAE,EACpC,OAAO,GAAE,WAAW,CAAC,MAAM,CAAa,GACvC,MAAM,CAWR;AAED,gDAAgD;AAChD,wBAAgB,SAAS,CAAC,KAAK,EAAE,aAAa,CAAC,SAAS,CAAC,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,CAAC,CAAC,GAAG,MAAM,CAM/F"}
+{"version":3,"file":"html-format.d.ts","sourceRoot":"","sources":["../../src/reporters/html-format.ts"],"names":[],"mappings":"AAAA,yDAAyD;AACzD,wBAAgB,UAAU,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,CAOhD;AAED,gGAAgG;AAChG,wBAAgB,QAAQ,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,GAAE,MAAY,GAAG,MAAM,CAQjE;AAyDD;;;GAGG;AACH,wBAAgB,YAAY,CAAC,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,GAAE,MAAW,GAAG,MAAM,CAkBpF;AAED,iFAAiF;AACjF,wBAAgB,SAAS,CACvB,OAAO,EAAE,SAAS,MAAM,EAAE,EAC1B,IAAI,EAAE,SAAS,CAAC,SAAS,MAAM,EAAE,CAAC,EAAE,EACpC,OAAO,GAAE,WAAW,CAAC,MAAM,CAAa,GACvC,MAAM,CAWR;AAED,gDAAgD;AAChD,wBAAgB,SAAS,CAAC,KAAK,EAAE,aAAa,CAAC,SAAS,CAAC,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,CAAC,CAAC,GAAG,MAAM,CAM/F"}

package/dist/reporters/html-format.js

@@ -72,8 +72,11 @@ pre { white-space: pre-wrap; word-break: break-word; margin: 6px 0 0; font: 12px
 details summary { cursor: pointer; }
 ul { padding-left: 20px; }
 `;
-/** A self-contained HTML page: inline CSS, no scripts and no external assets */
-export function htmlDocument(title, body) {
+/**
+ * A self-contained HTML page: inline CSS, no scripts and no external assets. `style` adds rules
+ * of the page's own after the shared ones
+ */
+export function htmlDocument(title, body, style = '') {
     return [
         '<!doctype html>',
         '<html lang="en">',
@@ -81,7 +84,7 @@ export function htmlDocument(title, body) {
         '<meta charset="utf-8">',
         '<meta name="viewport" content="width=device-width, initial-scale=1">',
         `<title>${escapeHtml(title)}</title>`,
-        `<style>${STYLE}</style>`,
+        `<style>${STYLE}${style}</style>`,
         '</head>',
         '<body>',
         '<main>',

package/dist/reporters/html-format.js.map

@@ -1 +1 @@
-{"version":3,"file":"html-format.js","sourceRoot":"","sources":["../../src/reporters/html-format.ts"],"names":[],"mappings":"AAAA,yDAAyD;AACzD,MAAM,UAAU,UAAU,CAAC,KAAa;IACtC,OAAO,KAAK;SACT,OAAO,CAAC,IAAI,EAAE,OAAO,CAAC;SACtB,OAAO,CAAC,IAAI,EAAE,MAAM,CAAC;SACrB,OAAO,CAAC,IAAI,EAAE,MAAM,CAAC;SACrB,OAAO,CAAC,IAAI,EAAE,QAAQ,CAAC;SACvB,OAAO,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;AAC5B,CAAC;AAED,gGAAgG;AAChG,MAAM,UAAU,QAAQ,CAAC,KAAa,EAAE,MAAc,GAAG;IACvD,MAAM,IAAI,GAAG,KAAK,CAAC,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC;IAC/C,IAAI,IAAI,CAAC,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC;QAAE,OAAO,UAAU,CAAC,KAAK,CAAC,CAAC;IAC1E,MAAM,OAAO,GAAG,CAAC,GAAG,IAAI,CAAC;SACtB,KAAK,CAAC,CAAC,EAAE,GAAG,GAAG,CAAC,CAAC;SACjB,IAAI,CAAC,EAAE,CAAC;SACR,OAAO,EAAE,CAAC;IACb,OAAO,qBAAqB,UAAU,CAAC,OAAO,CAAC,qBAAqB,UAAU,CAAC,KAAK,CAAC,kBAAkB,CAAC;AAC1G,CAAC;AAED,MAAM,KAAK,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAqDb,CAAC;AAEF,gFAAgF;AAChF,MAAM,UAAU,YAAY,CAAC,KAAa,EAAE,IAAY;IACtD,OAAO;QACL,iBAAiB;QACjB,kBAAkB;QAClB,QAAQ;QACR,wBAAwB;QACxB,sEAAsE;QACtE,UAAU,UAAU,CAAC,KAAK,CAAC,UAAU;QACrC,UAAU,KAAK,UAAU;QACzB,SAAS;QACT,QAAQ;QACR,QAAQ;QACR,IAAI;QACJ,SAAS;QACT,SAAS;QACT,SAAS;QACT,EAAE;KACH,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACf,CAAC;AAED,iFAAiF;AACjF,MAAM,UAAU,SAAS,CACvB,OAA0B,EAC1B,IAAoC,EACpC,UAA+B,IAAI,GAAG,EAAE;IAExC,MAAM,IAAI,GAAG,OAAO;SACjB,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,MAAM,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,IAAI,UAAU,CAAC,CAAC,CAAC,OAAO,CAAC;SAC/E,IAAI,CAAC,EAAE,CAAC,CAAC;IACZ,MAAM,IAAI,GAAG,IAAI;SACd,GAAG,CACF,CAAC,GAAG,EAAE,EAAE,CACN,OAAO,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,MAAM,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,IAAI,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,OAAO,CACjG;SACA,IAAI,CAAC,IAAI,CAAC,CAAC;IACd,OAAO,yCAAyC,IAAI,yBAAyB,IAAI,0BAA0B,CAAC;AAC9G,CAAC;AAED,gDAAgD;AAChD,MAAM,UAAU,SAAS,CAAC,KAA6D;IACrF,OAAO,sBAAsB,KAAK;SAC/B,GAAG,CACF,CAAC,CAAC,KAAK,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,wBAAwB,KAAK,aAAa,UAAU,CAAC,KAAK,CAAC,eAAe,CAC/F;SACA,IAAI,CAAC,EAAE,CAAC,QAAQ,CAAC;AACtB,CAAC"}
+{"version":3,"file":"html-format.js","sourceRoot":"","sources":["../../src/reporters/html-format.ts"],"names":[],"mappings":"AAAA,yDAAyD;AACzD,MAAM,UAAU,UAAU,CAAC,KAAa;IACtC,OAAO,KAAK;SACT,OAAO,CAAC,IAAI,EAAE,OAAO,CAAC;SACtB,OAAO,CAAC,IAAI,EAAE,MAAM,CAAC;SACrB,OAAO,CAAC,IAAI,EAAE,MAAM,CAAC;SACrB,OAAO,CAAC,IAAI,EAAE,QAAQ,CAAC;SACvB,OAAO,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;AAC5B,CAAC;AAED,gGAAgG;AAChG,MAAM,UAAU,QAAQ,CAAC,KAAa,EAAE,MAAc,GAAG;IACvD,MAAM,IAAI,GAAG,KAAK,CAAC,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC;IAC/C,IAAI,IAAI,CAAC,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC;QAAE,OAAO,UAAU,CAAC,KAAK,CAAC,CAAC;IAC1E,MAAM,OAAO,GAAG,CAAC,GAAG,IAAI,CAAC;SACtB,KAAK,CAAC,CAAC,EAAE,GAAG,GAAG,CAAC,CAAC;SACjB,IAAI,CAAC,EAAE,CAAC;SACR,OAAO,EAAE,CAAC;IACb,OAAO,qBAAqB,UAAU,CAAC,OAAO,CAAC,qBAAqB,UAAU,CAAC,KAAK,CAAC,kBAAkB,CAAC;AAC1G,CAAC;AAED,MAAM,KAAK,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAqDb,CAAC;AAEF;;;GAGG;AACH,MAAM,UAAU,YAAY,CAAC,KAAa,EAAE,IAAY,EAAE,QAAgB,EAAE;IAC1E,OAAO;QACL,iBAAiB;QACjB,kBAAkB;QAClB,QAAQ;QACR,wBAAwB;QACxB,sEAAsE;QACtE,UAAU,UAAU,CAAC,KAAK,CAAC,UAAU;QACrC,UAAU,KAAK,GAAG,KAAK,UAAU;QACjC,SAAS;QACT,QAAQ;QACR,QAAQ;QACR,IAAI;QACJ,SAAS;QACT,SAAS;QACT,SAAS;QACT,EAAE;KACH,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACf,CAAC;AAED,iFAAiF;AACjF,MAAM,UAAU,SAAS,CACvB,OAA0B,EAC1B,IAAoC,EACpC,UAA+B,IAAI,GAAG,EAAE;IAExC,MAAM,IAAI,GAAG,OAAO;SACjB,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,MAAM,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,IAAI,UAAU,CAAC,CAAC,CAAC,OAAO,CAAC;SAC/E,IAAI,CAAC,EAAE,CAAC,CAAC;IACZ,MAAM,IAAI,GAAG,IAAI;SACd,GAAG,CACF,CAAC,GAAG,EAAE,EAAE,CACN,OAAO,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,MAAM,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,IAAI,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,OAAO,CACjG;SACA,IAAI,CAAC,IAAI,CAAC,CAAC;IACd,OAAO,yCAAyC,IAAI,yBAAyB,IAAI,0BAA0B,CAAC;AAC9G,CAAC;AAED,gDAAgD;AAChD,MAAM,UAAU,SAAS,CAAC,KAA6D;IACrF,OAAO,sBAAsB,KAAK;SAC/B,GAAG,CACF,CAAC,CAAC,KAAK,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,wBAAwB,KAAK,aAAa,UAAU,CAAC,KAAK,CAAC,eAAe,CAC/F;SACA,IAAI,CAAC,EAAE,CAAC,QAAQ,CAAC;AACtB,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@cogitator-ai/evals",
-  "version": "0.13.2",
+  "version": "0.14.0",
   "description": "Evaluation framework for Cogitator AI agents",
   "keywords": [
     "cogitator",
@@ -41,10 +41,11 @@
     "node": ">=22.12.0"
   },
   "dependencies": {
+    "yaml": "^2.9.1",
     "zod": "^4.6.5"
   },
   "peerDependencies": {
-    "@cogitator-ai/core": ">=0.70.0",
+    "@cogitator-ai/core": ">=0.72.0",
     "vitest": ">=3.0.0"
   },
   "peerDependenciesMeta": {
@@ -59,7 +60,8 @@
     "papaparse": "^5.7.0"
   },
   "devDependencies": {
-    "@cogitator-ai/core": "0.70.0",
+    "@cogitator-ai/core": "0.72.0",
+    "@cogitator-ai/test-utils": "0.14.1",
     "@types/papaparse": "^5.3.15",
     "typescript": "^6.0.3",
     "vitest": "^5.0.3"