Press n or j to go to the next uncovered block, b, p or k for the previous block.
| 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 | 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x | import { existsSync, readFileSync, rmSync } from 'node:fs';
import { basename, dirname, join } from 'node:path';
import { z } from 'zod';
import { namedJudge } from 'vitest-evals';
import type { JudgeContext } from 'vitest-evals';
import {
normalizeContent,
normalizeMetadata,
toJsonValue,
type Harness,
} from 'vitest-evals/harness';
import { resolveSkillAsync } from '../skills/loader.js';
import { verifyFindings } from '../sdk/verify.js';
import { FindingSchema, type Finding, type UsageStats } from '../types/index.js';
import { RuntimeNameSchema, type RuntimeName } from '../sdk/runtimes/types.js';
import { discoverEvalScenarioFiles, resolveEvalSkillName } from './index.js';
import { formatEvalId } from './names.js';
import { setupEvalRepo } from './runner.js';
import type { EvalMeta } from './types.js';
import { usageToSummary } from './usage.js';
const DEFAULT_VERIFICATION_RUNTIME: RuntimeName = 'pi';
const DEFAULT_VERIFICATION_MODEL = 'anthropic/claude-sonnet-4-6';
const VerificationVerdictSchema = z.enum(['keep', 'reject']);
const VerificationExpectationSchema = z.object({
verdict: VerificationVerdictSchema,
});
const VerificationScenarioFileSchema = z.object({
name: z.string().optional(),
given: z.string(),
files: z.array(z.string()).min(1),
candidate: FindingSchema,
expect: VerificationExpectationSchema,
model: z.string().optional(),
runtime: RuntimeNameSchema.optional(),
});
export const VerificationEvalOutputSchema = z.object({
name: z.string(),
given: z.string(),
expectedVerdict: VerificationVerdictSchema,
verdict: VerificationVerdictSchema,
runtime: RuntimeNameSchema,
model: z.string(),
findings: z.array(FindingSchema),
});
export type VerificationEvalOutput = z.infer<typeof VerificationEvalOutputSchema>;
export interface VerificationEvalMeta {
name: string;
category: string;
skillName: string;
given: string;
skillPath: string;
filePaths: string[];
candidate: Finding;
expectedVerdict: z.infer<typeof VerificationVerdictSchema>;
model: string;
runtime: RuntimeName;
}
export interface VerificationScenarioSetOptions {
category: string;
skill: string;
runtime?: RuntimeName;
model?: string;
baseDir?: string;
}
export interface RunVerificationEvalOptions {
apiKey: string;
runtime?: RuntimeName;
model?: string;
verbose?: boolean;
}
interface VerificationEvalRunResult {
name: string;
meta: VerificationEvalMeta;
findings: Finding[];
verdict: z.infer<typeof VerificationVerdictSchema>;
logs: string[];
durationMs: number;
usage?: UsageStats;
}
function getEvalsDir(): string {
return join(import.meta.dirname, '..', '..', 'evals');
}
function loadVerificationScenario(filePath: string) {
const content = readFileSync(filePath, 'utf-8');
let parsed: unknown;
try {
parsed = JSON.parse(content);
} catch (error) {
const message = error instanceof Error ? error.message : String(error);
throw new Error(`Invalid verification eval ${filePath}: ${message}`);
}
const validated = VerificationScenarioFileSchema.safeParse(parsed);
if (!validated.success) {
const issues = validated.error.issues.map((issue) => `${issue.path.join('.')}: ${issue.message}`).join(', ');
throw new Error(`Invalid verification eval ${filePath}: ${issues}`);
}
return validated.data;
}
function copiedFixturePath(srcPath: string): string {
return `${basename(dirname(srcPath))}/${basename(srcPath)}`;
}
export function resolveVerificationEvalMeta(
scenarioPath: string,
options: VerificationScenarioSetOptions
): VerificationEvalMeta {
const evalsDir = options.baseDir ?? getEvalsDir();
const scenario = loadVerificationScenario(scenarioPath);
const name = scenario.name ?? basename(scenarioPath).replace(/\.json$/, '');
const skillPath = join(evalsDir, options.skill);
Iif (!existsSync(skillPath)) {
throw new Error(`Verification eval skill not found for ${options.category}/${name}: ${options.skill}`);
}
const filePaths = scenario.files.map((file) => {
const filePath = join(evalsDir, file);
if (!existsSync(filePath)) {
throw new Error(`Verification eval fixture not found for ${options.category}/${name}: ${file}`);
}
return filePath;
});
return {
name,
category: options.category,
skillName: resolveEvalSkillName(skillPath),
given: scenario.given,
skillPath,
filePaths,
candidate: scenario.candidate,
expectedVerdict: scenario.expect.verdict,
model: scenario.model ?? options.model ?? DEFAULT_VERIFICATION_MODEL,
runtime: scenario.runtime ?? options.runtime ?? DEFAULT_VERIFICATION_RUNTIME,
};
}
export function discoverVerificationEvalScenarios(options: VerificationScenarioSetOptions): VerificationEvalMeta[] {
return discoverEvalScenarioFiles(options.category, options.baseDir)
.map((file) => resolveVerificationEvalMeta(file, options));
}
export async function runVerificationEval(
meta: VerificationEvalMeta,
options: RunVerificationEvalOptions
): Promise<VerificationEvalRunResult> {
const startTime = Date.now();
const name = formatEvalId(meta);
const logs: string[] = [];
const log = (msg: string): void => {
logs.push(`[${Date.now() - startTime}ms] ${msg}`);
if (options.verbose) {
console.log(` [eval:${name}] ${msg}`);
}
};
const repoMeta: EvalMeta = {
name: meta.name,
category: meta.category,
skillName: meta.skillName,
given: meta.given,
skillPath: meta.skillPath,
filePaths: meta.filePaths,
model: meta.model,
runtime: meta.runtime,
should_find: [{ finding: meta.given, required: true }],
should_not_find: [],
};
let repoDir: string | undefined;
try {
repoDir = setupEvalRepo(repoMeta, log);
const skillSrcDir = dirname(meta.skillPath);
const skillPath = existsSync(join(skillSrcDir, 'SKILL.md'))
? join(repoDir, '.warden', 'skills', basename(skillSrcDir))
: join(repoDir, '.warden', 'skills', basename(meta.skillPath));
const skill = await resolveSkillAsync(skillPath);
const runtime = options.runtime ?? meta.runtime;
const model = options.model ?? meta.model;
log(`Verifying candidate with model: ${model} [${runtime}]`);
const result = await verifyFindings([meta.candidate], {
repoPath: repoDir,
skill,
apiKey: options.apiKey,
runtime,
model,
prContext: {
title: meta.given,
changedFiles: meta.filePaths.map(copiedFixturePath),
},
});
const verdict = result.findings.length > 0 ? 'keep' : 'reject';
log(`Verifier returned ${verdict} (${result.findings.length} finding(s))`);
return {
name,
meta,
findings: result.findings,
verdict,
logs,
durationMs: Date.now() - startTime,
usage: result.usage,
};
} finally {
if (repoDir) {
rmSync(repoDir, { recursive: true, force: true });
}
}
}
export function createVerificationEvalHarness(options: RunVerificationEvalOptions): Harness<VerificationEvalMeta> {
return {
name: 'warden-verification',
prompt: async () => {
throw new Error('Warden verification evals use runVerificationEval directly.');
},
run: async (meta, context) => {
const modelOverride = typeof context.metadata['model'] === 'string'
? context.metadata['model']
: undefined;
const runtimeOverride = context.metadata['runtime'] === 'claude'
|| context.metadata['runtime'] === 'pi'
? context.metadata['runtime']
: undefined;
const model = modelOverride ?? options.model ?? meta.model;
const runtime = runtimeOverride ?? options.runtime ?? meta.runtime;
const result = await runVerificationEval(meta, {
...options,
model,
runtime,
});
const output: VerificationEvalOutput = {
name: result.name,
given: meta.given,
expectedVerdict: meta.expectedVerdict,
verdict: result.verdict,
runtime,
model,
findings: result.findings,
};
return {
output: toJsonValue(output),
session: {
messages: [
{
role: 'user',
content: normalizeContent({
given: meta.given,
candidate: meta.candidate,
expectedVerdict: meta.expectedVerdict,
}),
},
{
role: 'assistant',
content: normalizeContent(output),
},
],
outputText: result.verdict,
provider: runtime,
model,
metadata: normalizeMetadata({
category: meta.category,
scenario: meta.name,
skill: meta.skillName,
skillPath: meta.skillPath,
}),
},
usage: usageToSummary({ provider: runtime, model, usage: result.usage }),
timings: { totalMs: result.durationMs },
artifacts: {
logs: toJsonValue(result.logs) ?? [],
},
errors: [],
};
},
};
}
/** Creates the deterministic judge for verifier-only eval verdicts. */
export function createVerificationEvalJudge() {
return namedJudge<JudgeContext<VerificationEvalMeta>>('WardenVerificationEvalJudge', async ({ inputValue, run }) => {
const output = VerificationEvalOutputSchema.safeParse(run.output);
if (!output.success) {
return {
score: 0,
metadata: {
rationale: `Invalid Warden verification output: ${output.error.message}`,
},
};
}
const passed = output.data.verdict === inputValue.expectedVerdict;
return {
score: passed ? 1 : 0,
metadata: {
rationale: passed
? 'Verifier verdict matched expected result.'
: `Expected ${inputValue.expectedVerdict}, got ${output.data.verdict}.`,
expectedVerdict: inputValue.expectedVerdict,
verdict: output.data.verdict,
findings: output.data.findings.length,
},
};
});
}
|