@vercel/agent-eval 0.0.15 → 0.2.0
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/cli.js +3 -5
- package/dist/cli.js.map +1 -1
- package/dist/lib/config.d.ts +1 -0
- package/dist/lib/config.d.ts.map +1 -1
- package/dist/lib/config.js +53 -6
- package/dist/lib/config.js.map +1 -1
- package/dist/lib/fixture.d.ts +2 -1
- package/dist/lib/fixture.d.ts.map +1 -1
- package/dist/lib/fixture.js +24 -11
- package/dist/lib/fixture.js.map +1 -1
- package/dist/lib/init.js +2 -2
- package/dist/lib/results.d.ts +2 -2
- package/dist/lib/results.d.ts.map +1 -1
- package/dist/lib/results.js +2 -4
- package/dist/lib/results.js.map +1 -1
- package/dist/lib/types.d.ts +3 -1
- package/dist/lib/types.d.ts.map +1 -1
- package/package.json +2 -1
package/dist/cli.js
CHANGED
|
@@ -23,7 +23,7 @@ const __dirname = dirname(fileURLToPath(import.meta.url));
|
|
|
23
23
|
const pkg = JSON.parse(readFileSync(resolve(__dirname, '../package.json'), 'utf-8'));
|
|
24
24
|
const program = new Command();
|
|
25
25
|
program
|
|
26
|
-
.name('agent-eval')
|
|
26
|
+
.name('@vercel/agent-eval')
|
|
27
27
|
.description('Framework for testing AI coding agents in isolated sandboxes')
|
|
28
28
|
.version(pkg.version);
|
|
29
29
|
/**
|
|
@@ -123,10 +123,8 @@ async function runExperimentCommand(configInput, options) {
|
|
|
123
123
|
for (const model of models) {
|
|
124
124
|
// Create a config for this specific model
|
|
125
125
|
const modelConfig = { ...config, model };
|
|
126
|
-
// Include model in experiment name
|
|
127
|
-
const experimentName =
|
|
128
|
-
? `${baseExperimentName}/${model}`
|
|
129
|
-
: baseExperimentName;
|
|
126
|
+
// Include model in experiment name for organized results
|
|
127
|
+
const experimentName = `${baseExperimentName}/${model}`;
|
|
130
128
|
if (models.length > 1) {
|
|
131
129
|
console.log(chalk.blue(`\n--- Running with model: ${model} ---`));
|
|
132
130
|
}
|
package/dist/cli.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"cli.js","sourceRoot":"","sources":["../src/cli.ts"],"names":[],"mappings":";AAEA;;GAEG;AAEH,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,MAAM,IAAI,YAAY,EAAE,MAAM,QAAQ,CAAC;AAChD,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,QAAQ,EAAE,MAAM,MAAM,CAAC;AAClD,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,IAAI,CAAC;AAC9C,OAAO,EAAE,aAAa,EAAE,MAAM,KAAK,CAAC;AACpC,OAAO,KAAK,MAAM,OAAO,CAAC;AAC1B,OAAO,EAAE,UAAU,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AAC/D,OAAO,EAAE,eAAe,EAAE,MAAM,kBAAkB,CAAC;AACnD,OAAO,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AAChD,OAAO,EAAE,WAAW,EAAE,uBAAuB,EAAE,MAAM,eAAe,CAAC;AACrE,OAAO,EAAE,QAAQ,EAAE,MAAM,uBAAuB,CAAC;AACjD,OAAO,EAAE,qBAAqB,EAAE,MAAM,kBAAkB,CAAC;AACzD,OAAO,EAAE,SAAS,EAAE,MAAM,eAAe,CAAC;AAE1C,uEAAuE;AACvE,YAAY,CAAC,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,CAAC;AACrC,YAAY,EAAE,CAAC;AAEf,iCAAiC;AACjC,MAAM,SAAS,GAAG,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAC1D,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,OAAO,CAAC,SAAS,EAAE,iBAAiB,CAAC,EAAE,OAAO,CAAC,CAAC,CAAC;AAErF,MAAM,OAAO,GAAG,IAAI,OAAO,EAAE,CAAC;AAE9B,OAAO;KACJ,IAAI,CAAC,
|
|
1
|
+
{"version":3,"file":"cli.js","sourceRoot":"","sources":["../src/cli.ts"],"names":[],"mappings":";AAEA;;GAEG;AAEH,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,MAAM,IAAI,YAAY,EAAE,MAAM,QAAQ,CAAC;AAChD,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,QAAQ,EAAE,MAAM,MAAM,CAAC;AAClD,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,IAAI,CAAC;AAC9C,OAAO,EAAE,aAAa,EAAE,MAAM,KAAK,CAAC;AACpC,OAAO,KAAK,MAAM,OAAO,CAAC;AAC1B,OAAO,EAAE,UAAU,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AAC/D,OAAO,EAAE,eAAe,EAAE,MAAM,kBAAkB,CAAC;AACnD,OAAO,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AAChD,OAAO,EAAE,WAAW,EAAE,uBAAuB,EAAE,MAAM,eAAe,CAAC;AACrE,OAAO,EAAE,QAAQ,EAAE,MAAM,uBAAuB,CAAC;AACjD,OAAO,EAAE,qBAAqB,EAAE,MAAM,kBAAkB,CAAC;AACzD,OAAO,EAAE,SAAS,EAAE,MAAM,eAAe,CAAC;AAE1C,uEAAuE;AACvE,YAAY,CAAC,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,CAAC;AACrC,YAAY,EAAE,CAAC;AAEf,iCAAiC;AACjC,MAAM,SAAS,GAAG,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAC1D,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,OAAO,CAAC,SAAS,EAAE,iBAAiB,CAAC,EAAE,OAAO,CAAC,CAAC,CAAC;AAErF,MAAM,OAAO,GAAG,IAAI,OAAO,EAAE,CAAC;AAE9B,OAAO;KACJ,IAAI,CAAC,oBAAoB,CAAC;KAC1B,WAAW,CAAC,8DAA8D,CAAC;KAC3E,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;AAExB;;;;GAIG;AACH,SAAS,iBAAiB,CAAC,KAAa;IACtC,6DAA6D;IAC7D,IAAI,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;QAClG,OAAO,KAAK,CAAC;IACf,CAAC;IACD,6DAA6D;IAC7D,OAAO,eAAe,KAAK,KAAK,CAAC;AACnC,CAAC;AAED;;GAEG;AACH,KAAK,UAAU,oBAAoB,CAAC,WAAmB,EAAE,OAA0B;IACjF,IAAI,CAAC;QACH,MAAM,UAAU,GAAG,iBAAiB,CAAC,WAAW,CAAC,CAAC;QAClD,MAAM,kBAAkB,GAAG,OAAO,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,UAAU,CAAC,CAAC;QAE9D,IAAI,CAAC,UAAU,CAAC,kBAAkB,CAAC,EAAE,CAAC;YACpC,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,0BAA0B,kBAAkB,EAAE,CAAC,CAAC,CAAC;YACzE,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;QAED,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,uBAAuB,UAAU,KAAK,CAAC,CAAC,CAAC;QAChE,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,kBAAkB,CAAC,CAAC;QAEpD,mDAAmD;QACnD,kEAAkE;QAClE,MAAM,UAAU,GAAG,OAAO,CAAC,OAAO,CAAC,kBAAkB,CAAC,CAAC,CAAC;QACxD,MAAM,QAAQ,GAAG,OAAO,CAAC,UAAU,EAAE,OAAO,CAAC,CAAC;QAC9C,IAAI,CAAC,UAAU,CAAC,QAAQ,CAAC,EAAE,CAAC;YAC1B,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,8BAA8B,QAAQ,EAAE,CAAC,CAAC,CAAC;YACnE,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,yDAAyD,CAAC,CAAC,CAAC;YACrF,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;QAED,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,wBAAwB,QAAQ,KAAK,CAAC,CAAC,CAAC;QAC/D,MAAM,EAAE,QAAQ,EAAE,MAAM,EAAE,GAAG,eAAe,CAAC,QAAQ,CAAC,CAAC;QAEvD,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACtB,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,cAAc,MAAM,CAAC,MAAM,sBAAsB,CAAC,CAAC,CAAC;YAC7E,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;gBAC3B,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,OAAO,KAAK,CAAC,WAAW,KAAK,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;YAC1E,CAAC;QACH,CAAC;QAED,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC1B,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,8BAA8B,CAAC,CAAC,CAAC;YACzD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;QAED,6BAA6B;QAC7B,MAAM,cAAc,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;QACnD,MAAM,SAAS,GAAG,gBAAgB,CAAC,MAAM,CAAC,KAAK,EAAE,cAAc,CAAC,CAAC;QAEjE,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC3B,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,6BAA6B,CAAC,CAAC,CAAC;YACxD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;QAED,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,WAAW,QAAQ,CAAC,MAAM,+BAA+B,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QACvG,KAAK,MAAM,IAAI,IAAI,SAAS,EAAE,CAAC;YAC7B,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,CAAC;QAC1C,CAAC;QAEJ,MAAM,MAAM,GAAG,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QAExE,2BAA2B;QAC3B,MAAM,iBAAiB,GAAG,SAAS,CAAC,MAAM,GAAG,MAAM,CAAC,IAAI,CAAC;QACzD,MAAM,SAAS,GAAG,iBAAiB,GAAG,MAAM,CAAC,MAAM,CAAC;QAEpD,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACtB,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,aAAa,SAAS,CAAC,MAAM,cAAc,MAAM,CAAC,IAAI,aAAa,MAAM,CAAC,MAAM,eAAe,SAAS,aAAa,CAAC,CAAC,CAAC;YAC/I,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,UAAU,MAAM,CAAC,KAAK,aAAa,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,cAAc,MAAM,CAAC,OAAO,kBAAkB,MAAM,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC;QAChJ,CAAC;aAAM,CAAC;YACN,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,aAAa,SAAS,CAAC,MAAM,cAAc,MAAM,CAAC,IAAI,aAAa,SAAS,aAAa,CAAC,CAAC,CAAC;YACnH,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,UAAU,MAAM,CAAC,KAAK,YAAY,MAAM,CAAC,CAAC,CAAC,cAAc,MAAM,CAAC,OAAO,kBAAkB,MAAM,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC;QACvI,CAAC;QAED,0CAA0C;QAC1C,MAAM,WAAW,GAAG,qBAAqB,CAAC,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC;QACvE,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,YAAY,WAAW,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC;QAE/D,IAAI,OAAO,CAAC,GAAG,EAAE,CAAC;YAChB,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,sCAAsC,CAAC,CAAC,CAAC;YAClE,OAAO;QACT,CAAC;QAED,8CAA8C;QAC9C,MAAM,KAAK,GAAG,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QACrC,MAAM,YAAY,GAAG,KAAK,CAAC,eAAe,EAAE,CAAC;QAC7C,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC;QACzC,IAAI,CAAC,MAAM,EAAE,CAAC;YACZ,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,YAAY,mCAAmC,CAAC,CAAC,CAAC;YAC7E,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,iEAAiE,CAAC,CAAC,CAAC;YAC7F,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;QAED,kDAAkD;QAClD,MAAM,gBAAgB,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC;QAE5E,uCAAuC;QACvC,MAAM,kBAAkB,GAAG,QAAQ,CAAC,UAAU,EAAE,KAAK,CAAC,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;QAC5E,MAAM,UAAU,GAAG,OAAO,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,SAAS,CAAC,CAAC;QAErD,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC,CAAC;QAEpD,iCAAiC;QACjC,IAAI,SAAS,GAAG,IAAI,CAAC;QACrB,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;YAC3B,0CAA0C;YAC1C,MAAM,WAAW,GAAG,EAAE,GAAG,MAAM,EAAE,KAAK,EAAE,CAAC;YAEzC,yDAAyD;YACzD,MAAM,cAAc,GAAG,GAAG,kBAAkB,IAAI,KAAK,EAAE,CAAC;YAExD,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBACtB,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,6BAA6B,KAAK,MAAM,CAAC,CAAC,CAAC;YACpE,CAAC;YAED,qBAAqB;YACrB,MAAM,OAAO,GAAG,MAAM,aAAa,CAAC;gBAClC,MAAM,EAAE,WAAW;gBACnB,QAAQ,EAAE,gBAAgB;gBAC1B,MAAM;gBACN,UAAU;gBACV,cAAc;gBACd,UAAU,EAAE,CAAC,GAAG,EAAE,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC;aACtC,CAAC,CAAC;YAEH,kCAAkC;YAClC,MAAM,gBAAgB,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,UAAU,KAAK,CAAC,CAAC,SAAS,CAAC,CAAC;YAClF,IAAI,CAAC,gBAAgB,EAAE,CAAC;gBACtB,SAAS,GAAG,KAAK,CAAC;YACpB,CAAC;QACH,CAAC;QAED,6BAA6B;QAC7B,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAClC,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,KAAK,YAAY,KAAK,EAAE,CAAC;YAC3B,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,UAAU,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QACtD,CAAC;aAAM,CAAC;YACN,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,2BAA2B,CAAC,CAAC,CAAC;QACxD,CAAC;QACD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;AACH,CAAC;AAED;;GAEG;AACH,OAAO;KACJ,OAAO,CAAC,MAAM,CAAC;KACf,QAAQ,CAAC,QAAQ,EAAE,+BAA+B,CAAC;KACnD,WAAW,CAAC,iDAAiD,CAAC;KAC9D,MAAM,CAAC,KAAK,EAAE,IAAY,EAAE,EAAE;IAC7B,IAAI,CAAC;QACH,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,8BAA8B,IAAI,EAAE,CAAC,CAAC,CAAC;QAE9D,MAAM,UAAU,GAAG,WAAW,CAAC;YAC7B,IAAI;YACJ,SAAS,EAAE,OAAO,CAAC,GAAG,EAAE;SACzB,CAAC,CAAC;QAEH,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,+BAA+B,CAAC,CAAC,CAAC;QAC1D,OAAO,CAAC,GAAG,CAAC,uBAAuB,CAAC,UAAU,EAAE,IAAI,CAAC,CAAC,CAAC;IACzD,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,KAAK,YAAY,KAAK,EAAE,CAAC;YAC3B,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,UAAU,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QACtD,CAAC;aAAM,CAAC;YACN,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,2BAA2B,CAAC,CAAC,CAAC;QACxD,CAAC;QACD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;AACH,CAAC,CAAC,CAAC;AAEL;;;GAGG;AACH,OAAO;KACJ,OAAO,CAAC,YAAY,CAAC;KACrB,WAAW,CAAC,iEAAiE,CAAC;KAC9E,MAAM,CAAC,eAAe,EAAE,kBAAkB,EAAE,MAAM,CAAC;KACnD,MAAM,CAAC,qBAAqB,EAAE,2BAA2B,EAAE,WAAW,CAAC;KACvE,MAAM,CAAC,mBAAmB,EAAE,yBAAyB,EAAE,SAAS,CAAC;KACjE,MAAM,CAAC,SAAS,EAAE,wDAAwD,CAAC;KAC3E,MAAM,CAAC,KAAK,EAAE,OAAgF,EAAE,EAAE;IACjG,MAAM,UAAU,GAAG,OAAO,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,OAAO,CAAC,UAAU,CAAC,CAAC;IAC9D,MAAM,QAAQ,GAAG,OAAO,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,OAAO,CAAC,QAAQ,CAAC,CAAC;IAE1D,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,mCAAmC,CAAC,CAAC,CAAC;IAE7D,oCAAoC;IACpC,MAAM,cAAc,GAAG;QACrB,eAAe,EAAE,UAAU;QAC3B,aAAa,EAAE,QAAQ;QACvB,QAAQ,EAAE,OAAO,CAAC,IAAI;KACvB,CAAC;IACF,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;QAClB,cAAc,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IACjC,CAAC;IAED,+DAA+D;IAC/D,MAAM,MAAM,GAAG,SAAS,CACtB,KAAK,EACL,CAAC,+BAA+B,EAAE,GAAG,cAAc,CAAC,EACpD,EAAE,KAAK,EAAE,SAAS,EAAE,GAAG,EAAE,OAAO,CAAC,GAAG,EAAE,EAAE,CACzC,CAAC;IAEF,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,IAAI,CAAC,CAAC,CAAC;AACnC,CAAC,CAAC,CAAC;AAEL;;;GAGG;AACH,OAAO;KACJ,QAAQ,CAAC,UAAU,EAAE,sCAAsC,CAAC;KAC5D,MAAM,CAAC,OAAO,EAAE,0CAA0C,CAAC;KAC3D,MAAM,CAAC,KAAK,EAAE,WAA+B,EAAE,OAA0B,EAAE,EAAE;IAC5E,IAAI,CAAC,WAAW,EAAE,CAAC;QACjB,OAAO,CAAC,IAAI,EAAE,CAAC;QACf,OAAO;IACT,CAAC;IACD,MAAM,oBAAoB,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;AACnD,CAAC,CAAC,CAAC;AAEL,OAAO,CAAC,KAAK,EAAE,CAAC"}
|
package/dist/lib/config.d.ts
CHANGED
|
@@ -30,6 +30,7 @@ export declare function resolveConfig(config: ExperimentConfig): ResolvedExperim
|
|
|
30
30
|
export declare function loadConfig(configPath: string): Promise<ResolvedExperimentConfig>;
|
|
31
31
|
/**
|
|
32
32
|
* Resolves the evals filter to a list of eval names.
|
|
33
|
+
* Supports glob patterns like "vercel-cli/*" for nested directories.
|
|
33
34
|
*/
|
|
34
35
|
export declare function resolveEvalNames(filter: string | string[] | EvalFilter, availableEvals: string[]): string[];
|
|
35
36
|
//# sourceMappingURL=config.d.ts.map
|
package/dist/lib/config.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"config.d.ts","sourceRoot":"","sources":["../../src/lib/config.ts"],"names":[],"mappings":"AAAA;;GAEG;
|
|
1
|
+
{"version":3,"file":"config.d.ts","sourceRoot":"","sources":["../../src/lib/config.ts"],"names":[],"mappings":"AAAA;;GAEG;AAIH,OAAO,KAAK,EACV,gBAAgB,EAChB,wBAAwB,EACxB,UAAU,EACX,MAAM,YAAY,CAAC;AAGpB;;GAEG;AACH,eAAO,MAAM,eAAe;;;;;aAKX,MAAM,EAAE;;;CAGxB,CAAC;AA0BF;;;GAGG;AACH,wBAAgB,cAAc,CAAC,MAAM,EAAE,OAAO,GAAG,gBAAgB,CAWhE;AAED;;GAEG;AACH,wBAAgB,aAAa,CAAC,MAAM,EAAE,gBAAgB,GAAG,wBAAwB,CAmBhF;AAED;;;GAGG;AACH,wBAAsB,UAAU,CAAC,UAAU,EAAE,MAAM,GAAG,OAAO,CAAC,wBAAwB,CAAC,CA8BtF;AAgBD;;;GAGG;AACH,wBAAgB,gBAAgB,CAC9B,MAAM,EAAE,MAAM,GAAG,MAAM,EAAE,GAAG,UAAU,EACtC,cAAc,EAAE,MAAM,EAAE,GACvB,MAAM,EAAE,CA0DV"}
|
package/dist/lib/config.js
CHANGED
|
@@ -2,6 +2,7 @@
|
|
|
2
2
|
* Experiment configuration validation and defaults.
|
|
3
3
|
*/
|
|
4
4
|
import { z } from 'zod';
|
|
5
|
+
import { minimatch } from 'minimatch';
|
|
5
6
|
import { getAgent } from './agents/index.js';
|
|
6
7
|
/**
|
|
7
8
|
* Default configuration values.
|
|
@@ -107,27 +108,73 @@ export async function loadConfig(configPath) {
|
|
|
107
108
|
throw error;
|
|
108
109
|
}
|
|
109
110
|
}
|
|
111
|
+
/**
|
|
112
|
+
* Check if a name matches a glob-style pattern.
|
|
113
|
+
* Supports:
|
|
114
|
+
* - "*" matches everything
|
|
115
|
+
* - "vercel-cli/*" matches all under vercel-cli/
|
|
116
|
+
* - "* /deploy" matches any deploy in any folder
|
|
117
|
+
* - "vercel-cli/deploy" exact match
|
|
118
|
+
*
|
|
119
|
+
* Uses minimatch for robust glob matching.
|
|
120
|
+
*/
|
|
121
|
+
function matchesPattern(name, pattern) {
|
|
122
|
+
return minimatch(name, pattern);
|
|
123
|
+
}
|
|
110
124
|
/**
|
|
111
125
|
* Resolves the evals filter to a list of eval names.
|
|
126
|
+
* Supports glob patterns like "vercel-cli/*" for nested directories.
|
|
112
127
|
*/
|
|
113
128
|
export function resolveEvalNames(filter, availableEvals) {
|
|
114
|
-
// Single eval name
|
|
129
|
+
// Single eval name or pattern
|
|
115
130
|
if (typeof filter === 'string') {
|
|
116
131
|
if (filter === '*') {
|
|
117
132
|
return availableEvals;
|
|
118
133
|
}
|
|
134
|
+
// Check if it's a glob pattern
|
|
135
|
+
if (filter.includes('*')) {
|
|
136
|
+
const matched = availableEvals.filter((name) => matchesPattern(name, filter));
|
|
137
|
+
if (matched.length === 0) {
|
|
138
|
+
throw new Error(`No evals matched pattern "${filter}". Available evals: ${availableEvals.join(', ')}`);
|
|
139
|
+
}
|
|
140
|
+
return matched;
|
|
141
|
+
}
|
|
142
|
+
// Exact match
|
|
119
143
|
if (!availableEvals.includes(filter)) {
|
|
120
144
|
throw new Error(`Eval "${filter}" not found. Available evals: ${availableEvals.join(', ')}`);
|
|
121
145
|
}
|
|
122
146
|
return [filter];
|
|
123
147
|
}
|
|
124
|
-
// Array of eval names
|
|
148
|
+
// Array of eval names or patterns
|
|
125
149
|
if (Array.isArray(filter)) {
|
|
126
|
-
const
|
|
127
|
-
|
|
128
|
-
|
|
150
|
+
const result = [];
|
|
151
|
+
const seen = new Set();
|
|
152
|
+
for (const item of filter) {
|
|
153
|
+
// Handle glob patterns in arrays
|
|
154
|
+
if (item.includes('*')) {
|
|
155
|
+
const matched = availableEvals.filter((name) => matchesPattern(name, item));
|
|
156
|
+
if (matched.length === 0) {
|
|
157
|
+
throw new Error(`No evals matched pattern "${item}". Available evals: ${availableEvals.join(', ')}`);
|
|
158
|
+
}
|
|
159
|
+
for (const name of matched) {
|
|
160
|
+
if (!seen.has(name)) {
|
|
161
|
+
result.push(name);
|
|
162
|
+
seen.add(name);
|
|
163
|
+
}
|
|
164
|
+
}
|
|
165
|
+
}
|
|
166
|
+
else {
|
|
167
|
+
// Exact match
|
|
168
|
+
if (!availableEvals.includes(item)) {
|
|
169
|
+
throw new Error(`Eval "${item}" not found. Available evals: ${availableEvals.join(', ')}`);
|
|
170
|
+
}
|
|
171
|
+
if (!seen.has(item)) {
|
|
172
|
+
result.push(item);
|
|
173
|
+
seen.add(item);
|
|
174
|
+
}
|
|
175
|
+
}
|
|
129
176
|
}
|
|
130
|
-
return
|
|
177
|
+
return result;
|
|
131
178
|
}
|
|
132
179
|
// Filter function
|
|
133
180
|
return availableEvals.filter(filter);
|
package/dist/lib/config.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"config.js","sourceRoot":"","sources":["../../src/lib/config.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;
|
|
1
|
+
{"version":3,"file":"config.js","sourceRoot":"","sources":["../../src/lib/config.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,SAAS,EAAE,MAAM,WAAW,CAAC;AAMtC,OAAO,EAAE,QAAQ,EAAE,MAAM,mBAAmB,CAAC;AAE7C;;GAEG;AACH,MAAM,CAAC,MAAM,eAAe,GAAG;IAC7B,KAAK,EAAE,MAAe;IACtB,KAAK,EAAE,GAAY;IACnB,IAAI,EAAE,CAAC;IACP,SAAS,EAAE,IAAI;IACf,OAAO,EAAE,EAAc;IACvB,OAAO,EAAE,GAAG,EAAE,aAAa;IAC3B,OAAO,EAAE,MAAe;CACzB,CAAC;AAEF;;GAEG;AACH,MAAM,sBAAsB,GAAG,CAAC,CAAC,MAAM,CAAC;IACtC,KAAK,EAAE,CAAC,CAAC,IAAI,CAAC;QACZ,+BAA+B;QAC/B,aAAa;QACb,yBAAyB;QACzB,OAAO;QACP,4BAA4B;KAC7B,CAAC;IACF,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;IAC5D,KAAK,EAAE,CAAC;SACL,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;SAC5F,QAAQ,EAAE;IACb,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC5C,SAAS,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE;IACjC,OAAO,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE;IACvC,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACzC,KAAK,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC9B,OAAO,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,QAAQ,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC,CAAC,QAAQ,EAAE;IACxD,UAAU,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE;CACzE,CAAC,CAAC;AAEH;;;GAGG;AACH,MAAM,UAAU,cAAc,CAAC,MAAe;IAC5C,MAAM,MAAM,GAAG,sBAAsB,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;IAExD,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;QACpB,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,MAAM;aAC/B,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,KAAK,CAAC,OAAO,EAAE,CAAC;aAC/D,IAAI,CAAC,IAAI,CAAC,CAAC;QACd,MAAM,IAAI,KAAK,CAAC,sCAAsC,MAAM,EAAE,CAAC,CAAC;IAClE,CAAC;IAED,OAAO,MAAM,CAAC,IAAwB,CAAC;AACzC,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,aAAa,CAAC,MAAwB;IACpD,wBAAwB;IACxB,MAAM,KAAK,GAAG,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IAErC,gDAAgD;IAChD,MAAM,YAAY,GAAG,MAAM,CAAC,KAAK,IAAI,KAAK,CAAC,eAAe,EAAE,CAAC;IAE7D,OAAO;QACL,KAAK,EAAE,MAAM,CAAC,KAAK;QACnB,KAAK,EAAE,YAAY;QACnB,KAAK,EAAE,MAAM,CAAC,KAAK,IAAI,GAAG;QAC1B,IAAI,EAAE,MAAM,CAAC,IAAI,IAAI,eAAe,CAAC,IAAI;QACzC,SAAS,EAAE,MAAM,CAAC,SAAS,IAAI,eAAe,CAAC,SAAS;QACxD,OAAO,EAAE,MAAM,CAAC,OAAO,IAAI,eAAe,CAAC,OAAO;QAClD,OAAO,EAAE,MAAM,CAAC,OAAO,IAAI,eAAe,CAAC,OAAO;QAClD,KAAK,EAAE,MAAM,CAAC,KAAK;QACnB,OAAO,EAAE,MAAM,CAAC,OAAO,IAAI,eAAe,CAAC,OAAO;QAClD,UAAU,EAAE,MAAM,CAAC,UAAU;KAC9B,CAAC;AACJ,CAAC;AAED;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,UAAU,CAAC,UAAkB;IACjD,IAAI,CAAC;QACH,IAAI,SAAkB,CAAC;QAEvB,gCAAgC;QAChC,IAAI,UAAU,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;YAC/B,MAAM,EAAE,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC,MAAM,CAAC,CAAC;YAC5C,MAAM,IAAI,GAAG,UAAU,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,EAAE;gBACvC,cAAc,EAAE,IAAI;gBACpB,WAAW,EAAE,KAAK;aACnB,CAAC,CAAC;YACH,SAAS,GAAG,MAAM,IAAI,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC;QAC5C,CAAC;aAAM,CAAC;YACN,sCAAsC;YACtC,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,UAAU,CAAC,CAAC;YACxC,SAAS,GAAG,MAAM,CAAC,OAAO,CAAC;QAC7B,CAAC;QAED,IAAI,CAAC,SAAS,EAAE,CAAC;YACf,MAAM,IAAI,KAAK,CAAC,wCAAwC,CAAC,CAAC;QAC5D,CAAC;QAED,MAAM,MAAM,GAAG,cAAc,CAAC,SAAS,CAAC,CAAC;QACzC,OAAO,aAAa,CAAC,MAAM,CAAC,CAAC;IAC/B,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,KAAK,YAAY,KAAK,EAAE,CAAC;YAC3B,MAAM,IAAI,KAAK,CAAC,8BAA8B,UAAU,KAAK,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC;QAChF,CAAC;QACD,MAAM,KAAK,CAAC;IACd,CAAC;AACH,CAAC;AAED;;;;;;;;;GASG;AACH,SAAS,cAAc,CAAC,IAAY,EAAE,OAAe;IACnD,OAAO,SAAS,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;AAClC,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,gBAAgB,CAC9B,MAAsC,EACtC,cAAwB;IAExB,8BAA8B;IAC9B,IAAI,OAAO,MAAM,KAAK,QAAQ,EAAE,CAAC;QAC/B,IAAI,MAAM,KAAK,GAAG,EAAE,CAAC;YACnB,OAAO,cAAc,CAAC;QACxB,CAAC;QAED,+BAA+B;QAC/B,IAAI,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC;YACzB,MAAM,OAAO,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,cAAc,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC,CAAC;YAC9E,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBACzB,MAAM,IAAI,KAAK,CAAC,6BAA6B,MAAM,uBAAuB,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YACzG,CAAC;YACD,OAAO,OAAO,CAAC;QACjB,CAAC;QAED,cAAc;QACd,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,MAAM,CAAC,EAAE,CAAC;YACrC,MAAM,IAAI,KAAK,CAAC,SAAS,MAAM,iCAAiC,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAC/F,CAAC;QACD,OAAO,CAAC,MAAM,CAAC,CAAC;IAClB,CAAC;IAED,kCAAkC;IAClC,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;QAC1B,MAAM,MAAM,GAAa,EAAE,CAAC;QAC5B,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAC;QAE/B,KAAK,MAAM,IAAI,IAAI,MAAM,EAAE,CAAC;YAC1B,iCAAiC;YACjC,IAAI,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC;gBACvB,MAAM,OAAO,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,cAAc,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC,CAAC;gBAC5E,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;oBACzB,MAAM,IAAI,KAAK,CAAC,6BAA6B,IAAI,uBAAuB,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;gBACvG,CAAC;gBACD,KAAK,MAAM,IAAI,IAAI,OAAO,EAAE,CAAC;oBAC3B,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC;wBACpB,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;wBAClB,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;oBACjB,CAAC;gBACH,CAAC;YACH,CAAC;iBAAM,CAAC;gBACN,cAAc;gBACd,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,CAAC;oBACnC,MAAM,IAAI,KAAK,CAAC,SAAS,IAAI,iCAAiC,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;gBAC7F,CAAC;gBACD,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC;oBACpB,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;oBAClB,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;gBACjB,CAAC;YACH,CAAC;QACH,CAAC;QAED,OAAO,MAAM,CAAC;IAChB,CAAC;IAED,kBAAkB;IAClB,OAAO,cAAc,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;AACvC,CAAC"}
|
package/dist/lib/fixture.d.ts
CHANGED
|
@@ -11,7 +11,8 @@ export declare class FixtureValidationError extends Error {
|
|
|
11
11
|
}
|
|
12
12
|
/**
|
|
13
13
|
* Discovers all eval fixtures in a directory.
|
|
14
|
-
*
|
|
14
|
+
* Recursively searches for directories containing PROMPT.md.
|
|
15
|
+
* Supports nested organization like "vercel-cli/deploy" or "flags/create-flag".
|
|
15
16
|
*/
|
|
16
17
|
export declare function discoverFixtures(evalsDir: string): string[];
|
|
17
18
|
/**
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"fixture.d.ts","sourceRoot":"","sources":["../../src/lib/fixture.ts"],"names":[],"mappings":"AAAA;;GAEG;AAIH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,YAAY,CAAC;AAG9C;;GAEG;AACH,qBAAa,sBAAuB,SAAQ,KAAK;IAEtC,WAAW,EAAE,MAAM;gBAAnB,WAAW,EAAE,MAAM,EAC1B,OAAO,EAAE,MAAM;CAKlB;AAeD
|
|
1
|
+
{"version":3,"file":"fixture.d.ts","sourceRoot":"","sources":["../../src/lib/fixture.ts"],"names":[],"mappings":"AAAA;;GAEG;AAIH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,YAAY,CAAC;AAG9C;;GAEG;AACH,qBAAa,sBAAuB,SAAQ,KAAK;IAEtC,WAAW,EAAE,MAAM;gBAAnB,WAAW,EAAE,MAAM,EAC1B,OAAO,EAAE,MAAM;CAKlB;AAeD;;;;GAIG;AACH,wBAAgB,gBAAgB,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,EAAE,CAuC3D;AAED;;;;;GAKG;AACH,wBAAgB,oBAAoB,CAAC,WAAW,EAAE,MAAM,GAAG,MAAM,EAAE,CAmBlE;AAED;;;GAGG;AACH,wBAAgB,mBAAmB,CAAC,WAAW,EAAE,MAAM,GAAG;IAAE,QAAQ,EAAE,OAAO,CAAC;IAAC,KAAK,CAAC,EAAE,MAAM,CAAA;CAAE,CAwB9F;AAED;;GAEG;AACH,wBAAgB,WAAW,CAAC,QAAQ,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,WAAW,CAmCvE;AAED;;;GAGG;AACH,wBAAgB,eAAe,CAAC,QAAQ,EAAE,MAAM,GAAG;IACjD,QAAQ,EAAE,WAAW,EAAE,CAAC;IACxB,MAAM,EAAE,sBAAsB,EAAE,CAAC;CAClC,CAmBA;AAED;;;GAGG;AACH,wBAAgB,eAAe,CAC7B,WAAW,EAAE,MAAM,EACnB,eAAe,GAAE,SAAS,MAAM,EAAmB,GAClD,MAAM,EAAE,CA2BV;AAED;;;GAGG;AACH,wBAAgB,gBAAgB,CAC9B,WAAW,EAAE,MAAM,EACnB,eAAe,CAAC,EAAE,SAAS,MAAM,EAAE,GAClC,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAUrB"}
|
package/dist/lib/fixture.js
CHANGED
|
@@ -30,26 +30,39 @@ function existsWithExactCase(dirPath, fileName) {
|
|
|
30
30
|
}
|
|
31
31
|
/**
|
|
32
32
|
* Discovers all eval fixtures in a directory.
|
|
33
|
-
*
|
|
33
|
+
* Recursively searches for directories containing PROMPT.md.
|
|
34
|
+
* Supports nested organization like "vercel-cli/deploy" or "flags/create-flag".
|
|
34
35
|
*/
|
|
35
36
|
export function discoverFixtures(evalsDir) {
|
|
36
37
|
const absolutePath = resolve(evalsDir);
|
|
37
38
|
if (!existsSync(absolutePath)) {
|
|
38
39
|
throw new Error(`Evals directory not found: ${absolutePath}`);
|
|
39
40
|
}
|
|
40
|
-
const entries = readdirSync(absolutePath);
|
|
41
41
|
const fixtures = [];
|
|
42
|
-
|
|
43
|
-
const
|
|
44
|
-
|
|
45
|
-
|
|
46
|
-
|
|
47
|
-
|
|
48
|
-
|
|
49
|
-
|
|
50
|
-
|
|
42
|
+
function walk(dir, basePath = '') {
|
|
43
|
+
const entries = readdirSync(dir);
|
|
44
|
+
for (const entry of entries) {
|
|
45
|
+
// Skip hidden directories and files
|
|
46
|
+
if (entry.startsWith('.')) {
|
|
47
|
+
continue;
|
|
48
|
+
}
|
|
49
|
+
const entryPath = join(dir, entry);
|
|
50
|
+
// Only consider directories
|
|
51
|
+
if (!statSync(entryPath).isDirectory()) {
|
|
52
|
+
continue;
|
|
53
|
+
}
|
|
54
|
+
const relativePath = basePath ? `${basePath}/${entry}` : entry;
|
|
55
|
+
// Check if this directory contains PROMPT.md (potential fixture)
|
|
56
|
+
if (existsWithExactCase(entryPath, 'PROMPT.md')) {
|
|
57
|
+
fixtures.push(relativePath);
|
|
58
|
+
}
|
|
59
|
+
else {
|
|
60
|
+
// Not a fixture directory, recurse into it
|
|
61
|
+
walk(entryPath, relativePath);
|
|
62
|
+
}
|
|
51
63
|
}
|
|
52
64
|
}
|
|
65
|
+
walk(absolutePath);
|
|
53
66
|
return fixtures.sort();
|
|
54
67
|
}
|
|
55
68
|
/**
|
package/dist/lib/fixture.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"fixture.js","sourceRoot":"","sources":["../../src/lib/fixture.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,YAAY,EAAE,WAAW,EAAE,QAAQ,EAAE,UAAU,EAAE,MAAM,IAAI,CAAC;AACrE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,MAAM,CAAC;AAErC,OAAO,EAAE,mBAAmB,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AAEjE;;GAEG;AACH,MAAM,OAAO,sBAAuB,SAAQ,KAAK;IAEtC;IADT,YACS,WAAmB,EAC1B,OAAe;QAEf,KAAK,CAAC,SAAS,WAAW,MAAM,OAAO,EAAE,CAAC,CAAC;QAHpC,gBAAW,GAAX,WAAW,CAAQ;QAI1B,IAAI,CAAC,IAAI,GAAG,wBAAwB,CAAC;IACvC,CAAC;CACF;AAED;;;GAGG;AACH,SAAS,mBAAmB,CAAC,OAAe,EAAE,QAAgB;IAC5D,IAAI,CAAC;QACH,MAAM,KAAK,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC;QACnC,OAAO,KAAK,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IAClC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,KAAK,CAAC;IACf,CAAC;AACH,CAAC;AAED
|
|
1
|
+
{"version":3,"file":"fixture.js","sourceRoot":"","sources":["../../src/lib/fixture.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,YAAY,EAAE,WAAW,EAAE,QAAQ,EAAE,UAAU,EAAE,MAAM,IAAI,CAAC;AACrE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,MAAM,CAAC;AAErC,OAAO,EAAE,mBAAmB,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AAEjE;;GAEG;AACH,MAAM,OAAO,sBAAuB,SAAQ,KAAK;IAEtC;IADT,YACS,WAAmB,EAC1B,OAAe;QAEf,KAAK,CAAC,SAAS,WAAW,MAAM,OAAO,EAAE,CAAC,CAAC;QAHpC,gBAAW,GAAX,WAAW,CAAQ;QAI1B,IAAI,CAAC,IAAI,GAAG,wBAAwB,CAAC;IACvC,CAAC;CACF;AAED;;;GAGG;AACH,SAAS,mBAAmB,CAAC,OAAe,EAAE,QAAgB;IAC5D,IAAI,CAAC;QACH,MAAM,KAAK,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC;QACnC,OAAO,KAAK,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IAClC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,KAAK,CAAC;IACf,CAAC;AACH,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,gBAAgB,CAAC,QAAgB;IAC/C,MAAM,YAAY,GAAG,OAAO,CAAC,QAAQ,CAAC,CAAC;IAEvC,IAAI,CAAC,UAAU,CAAC,YAAY,CAAC,EAAE,CAAC;QAC9B,MAAM,IAAI,KAAK,CAAC,8BAA8B,YAAY,EAAE,CAAC,CAAC;IAChE,CAAC;IAED,MAAM,QAAQ,GAAa,EAAE,CAAC;IAE9B,SAAS,IAAI,CAAC,GAAW,EAAE,WAAmB,EAAE;QAC9C,MAAM,OAAO,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC;QAEjC,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;YAC5B,oCAAoC;YACpC,IAAI,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC;gBAC1B,SAAS;YACX,CAAC;YAED,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;YAEnC,4BAA4B;YAC5B,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC,WAAW,EAAE,EAAE,CAAC;gBACvC,SAAS;YACX,CAAC;YAED,MAAM,YAAY,GAAG,QAAQ,CAAC,CAAC,CAAC,GAAG,QAAQ,IAAI,KAAK,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC;YAE/D,iEAAiE;YACjE,IAAI,mBAAmB,CAAC,SAAS,EAAE,WAAW,CAAC,EAAE,CAAC;gBAChD,QAAQ,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;YAC9B,CAAC;iBAAM,CAAC;gBACN,2CAA2C;gBAC3C,IAAI,CAAC,SAAS,EAAE,YAAY,CAAC,CAAC;YAChC,CAAC;QACH,CAAC;IACH,CAAC;IAED,IAAI,CAAC,YAAY,CAAC,CAAC;IACnB,OAAO,QAAQ,CAAC,IAAI,EAAE,CAAC;AACzB,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,oBAAoB,CAAC,WAAmB;IACtD,MAAM,OAAO,GAAa,EAAE,CAAC;IAE7B,KAAK,MAAM,IAAI,IAAI,mBAAmB,EAAE,CAAC;QACvC,wEAAwE;QACxE,IAAI,IAAI,KAAK,SAAS,EAAE,CAAC;YACvB,MAAM,SAAS,GAAG,mBAAmB,CAAC,WAAW,EAAE,SAAS,CAAC,CAAC;YAC9D,MAAM,UAAU,GAAG,mBAAmB,CAAC,WAAW,EAAE,UAAU,CAAC,CAAC;YAChE,IAAI,CAAC,SAAS,IAAI,CAAC,UAAU,EAAE,CAAC;gBAC9B,OAAO,CAAC,IAAI,CAAC,qBAAqB,CAAC,CAAC;YACtC,CAAC;QACH,CAAC;aAAM,CAAC;YACN,IAAI,CAAC,mBAAmB,CAAC,WAAW,EAAE,IAAI,CAAC,EAAE,CAAC;gBAC5C,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACrB,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,OAAO,CAAC;AACjB,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,mBAAmB,CAAC,WAAmB;IACrD,MAAM,eAAe,GAAG,IAAI,CAAC,WAAW,EAAE,cAAc,CAAC,CAAC;IAE1D,IAAI,CAAC;QACH,MAAM,OAAO,GAAG,YAAY,CAAC,eAAe,EAAE,OAAO,CAAC,CAAC;QACvD,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;QAEhC,IAAI,GAAG,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;YAC1B,OAAO;gBACL,QAAQ,EAAE,KAAK;gBACf,KAAK,EAAE,yCAAyC;aACjD,CAAC;QACJ,CAAC;QAED,OAAO,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC;IAC5B,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,KAAK,YAAY,WAAW,EAAE,CAAC;YACjC,OAAO;gBACL,QAAQ,EAAE,KAAK;gBACf,KAAK,EAAE,gCAAgC;aACxC,CAAC;QACJ,CAAC;QACD,MAAM,KAAK,CAAC;IACd,CAAC;AACH,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,WAAW,CAAC,QAAgB,EAAE,IAAY;IACxD,MAAM,WAAW,GAAG,OAAO,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;IAE5C,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC,EAAE,CAAC;QAC7B,MAAM,IAAI,sBAAsB,CAAC,IAAI,EAAE,wBAAwB,WAAW,EAAE,CAAC,CAAC;IAChF,CAAC;IAED,0BAA0B;IAC1B,MAAM,YAAY,GAAG,oBAAoB,CAAC,WAAW,CAAC,CAAC;IACvD,IAAI,YAAY,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC5B,MAAM,IAAI,sBAAsB,CAC9B,IAAI,EACJ,2BAA2B,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CACrD,CAAC;IACJ,CAAC;IAED,wBAAwB;IACxB,MAAM,aAAa,GAAG,mBAAmB,CAAC,WAAW,CAAC,CAAC;IACvD,IAAI,aAAa,CAAC,KAAK,EAAE,CAAC;QACxB,MAAM,IAAI,sBAAsB,CAAC,IAAI,EAAE,aAAa,CAAC,KAAK,CAAC,CAAC;IAC9D,CAAC;IAED,qCAAqC;IACrC,IAAI,CAAC,mBAAmB,CAAC,WAAW,EAAE,WAAW,CAAC,EAAE,CAAC;QACnD,MAAM,IAAI,sBAAsB,CAAC,IAAI,EAAE,yDAAyD,CAAC,CAAC;IACpG,CAAC;IACD,MAAM,UAAU,GAAG,IAAI,CAAC,WAAW,EAAE,WAAW,CAAC,CAAC;IAClD,MAAM,MAAM,GAAG,YAAY,CAAC,UAAU,EAAE,OAAO,CAAC,CAAC;IAEjD,OAAO;QACL,IAAI;QACJ,IAAI,EAAE,WAAW;QACjB,MAAM;QACN,QAAQ,EAAE,aAAa,CAAC,QAAQ;KACjC,CAAC;AACJ,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,eAAe,CAAC,QAAgB;IAI9C,MAAM,YAAY,GAAG,gBAAgB,CAAC,QAAQ,CAAC,CAAC;IAChD,MAAM,QAAQ,GAAkB,EAAE,CAAC;IACnC,MAAM,MAAM,GAA6B,EAAE,CAAC;IAE5C,KAAK,MAAM,IAAI,IAAI,YAAY,EAAE,CAAC;QAChC,IAAI,CAAC;YACH,MAAM,OAAO,GAAG,WAAW,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;YAC5C,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QACzB,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,IAAI,KAAK,YAAY,sBAAsB,EAAE,CAAC;gBAC5C,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;YACrB,CAAC;iBAAM,CAAC;gBACN,MAAM,KAAK,CAAC;YACd,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,EAAE,QAAQ,EAAE,MAAM,EAAE,CAAC;AAC9B,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,eAAe,CAC7B,WAAmB,EACnB,kBAAqC,cAAc;IAEnD,MAAM,KAAK,GAAa,EAAE,CAAC;IAE3B,SAAS,IAAI,CAAC,GAAW,EAAE,WAAmB,EAAE;QAC9C,MAAM,OAAO,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC;QAEjC,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;YAC5B,MAAM,YAAY,GAAG,QAAQ,CAAC,CAAC,CAAC,GAAG,QAAQ,IAAI,KAAK,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC;YAE/D,8BAA8B;YAC9B,IAAI,eAAe,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,YAAY,KAAK,OAAO,IAAI,KAAK,KAAK,OAAO,CAAC,EAAE,CAAC;gBACrF,SAAS;YACX,CAAC;YAED,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;YAClC,MAAM,IAAI,GAAG,QAAQ,CAAC,QAAQ,CAAC,CAAC;YAEhC,IAAI,IAAI,CAAC,WAAW,EAAE,EAAE,CAAC;gBACvB,IAAI,CAAC,QAAQ,EAAE,YAAY,CAAC,CAAC;YAC/B,CAAC;iBAAM,CAAC;gBACN,KAAK,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;YAC3B,CAAC;QACH,CAAC;IACH,CAAC;IAED,IAAI,CAAC,WAAW,CAAC,CAAC;IAClB,OAAO,KAAK,CAAC,IAAI,EAAE,CAAC;AACtB,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,gBAAgB,CAC9B,WAAmB,EACnB,eAAmC;IAEnC,MAAM,KAAK,GAAG,eAAe,CAAC,WAAW,EAAE,eAAe,CAAC,CAAC;IAC5D,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAkB,CAAC;IAE3C,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,MAAM,QAAQ,GAAG,IAAI,CAAC,WAAW,EAAE,IAAI,CAAC,CAAC;QACzC,QAAQ,CAAC,GAAG,CAAC,IAAI,EAAE,YAAY,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC,CAAC;IACtD,CAAC;IAED,OAAO,QAAQ,CAAC;AAClB,CAAC"}
|
package/dist/lib/init.js
CHANGED
|
@@ -118,7 +118,7 @@ Open [http://localhost:3000](http://localhost:3000) to browse results.
|
|
|
118
118
|
* Get the default experiment configuration template (Claude Code).
|
|
119
119
|
*/
|
|
120
120
|
function getCCExperiment() {
|
|
121
|
-
return `import type { ExperimentConfig } from 'agent-eval';
|
|
121
|
+
return `import type { ExperimentConfig } from '@vercel/agent-eval';
|
|
122
122
|
|
|
123
123
|
const config: ExperimentConfig = {
|
|
124
124
|
agent: 'vercel-ai-gateway/claude-code',
|
|
@@ -135,7 +135,7 @@ export default config;
|
|
|
135
135
|
* Get the Codex experiment configuration template.
|
|
136
136
|
*/
|
|
137
137
|
function getCodexExperiment() {
|
|
138
|
-
return `import type { ExperimentConfig } from 'agent-eval';
|
|
138
|
+
return `import type { ExperimentConfig } from '@vercel/agent-eval';
|
|
139
139
|
|
|
140
140
|
const config: ExperimentConfig = {
|
|
141
141
|
agent: 'vercel-ai-gateway/codex',
|
package/dist/lib/results.d.ts
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
/**
|
|
2
2
|
* Results storage and reporting for eval experiments.
|
|
3
3
|
*/
|
|
4
|
-
import type { EvalRunResult, EvalRunData, EvalSummary, ExperimentResults,
|
|
4
|
+
import type { EvalRunResult, EvalRunData, EvalSummary, ExperimentResults, RunnableExperimentConfig } from './types.js';
|
|
5
5
|
import type { AgentRunResult } from './agents/types.js';
|
|
6
6
|
/**
|
|
7
7
|
* Convert AgentRunResult to EvalRunData (result + transcript).
|
|
@@ -14,7 +14,7 @@ export declare function createEvalSummary(name: string, runData: EvalRunData[]):
|
|
|
14
14
|
/**
|
|
15
15
|
* Create experiment results from eval summaries.
|
|
16
16
|
*/
|
|
17
|
-
export declare function createExperimentResults(config:
|
|
17
|
+
export declare function createExperimentResults(config: RunnableExperimentConfig, evals: EvalSummary[], startedAt: Date, completedAt: Date): ExperimentResults;
|
|
18
18
|
/**
|
|
19
19
|
* Options for saving results.
|
|
20
20
|
*/
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"results.d.ts","sourceRoot":"","sources":["../../src/lib/results.ts"],"names":[],"mappings":"AAAA;;GAEG;AAKH,OAAO,KAAK,EACV,aAAa,EACb,WAAW,EACX,WAAW,EACX,iBAAiB,EACjB,wBAAwB,EACzB,MAAM,YAAY,CAAC;AACpB,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAC;AAGxD;;GAEG;AACH,wBAAgB,wBAAwB,CAAC,WAAW,EAAE,cAAc,GAAG,WAAW,CA4BjF;AAED;;GAEG;AACH,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,WAAW,EAAE,GAAG,WAAW,CAanF;AAED;;GAEG;AACH,wBAAgB,uBAAuB,CACrC,MAAM,EAAE,wBAAwB,EAChC,KAAK,EAAE,WAAW,EAAE,EACpB,SAAS,EAAE,IAAI,EACf,WAAW,EAAE,IAAI,GAChB,iBAAiB,CAOnB;AAED;;GAEG;AACH,MAAM,WAAW,kBAAkB;IACjC,iCAAiC;IACjC,UAAU,EAAE,MAAM,CAAC;IACnB,8CAA8C;IAC9C,cAAc,EAAE,MAAM,CAAC;CACxB;AAED;;;;;;;;;;;;;;GAcG;AACH,wBAAgB,WAAW,CACzB,OAAO,EAAE,iBAAiB,EAC1B,OAAO,EAAE,kBAAkB,GAC1B,MAAM,
|
|
1
|
+
{"version":3,"file":"results.d.ts","sourceRoot":"","sources":["../../src/lib/results.ts"],"names":[],"mappings":"AAAA;;GAEG;AAKH,OAAO,KAAK,EACV,aAAa,EACb,WAAW,EACX,WAAW,EACX,iBAAiB,EACjB,wBAAwB,EACzB,MAAM,YAAY,CAAC;AACpB,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAC;AAGxD;;GAEG;AACH,wBAAgB,wBAAwB,CAAC,WAAW,EAAE,cAAc,GAAG,WAAW,CA4BjF;AAED;;GAEG;AACH,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,WAAW,EAAE,GAAG,WAAW,CAanF;AAED;;GAEG;AACH,wBAAgB,uBAAuB,CACrC,MAAM,EAAE,wBAAwB,EAChC,KAAK,EAAE,WAAW,EAAE,EACpB,SAAS,EAAE,IAAI,EACf,WAAW,EAAE,IAAI,GAChB,iBAAiB,CAOnB;AAED;;GAEG;AACH,MAAM,WAAW,kBAAkB;IACjC,iCAAiC;IACjC,UAAU,EAAE,MAAM,CAAC;IACnB,8CAA8C;IAC9C,cAAc,EAAE,MAAM,CAAC;CACxB;AAED;;;;;;;;;;;;;;GAcG;AACH,wBAAgB,WAAW,CACzB,OAAO,EAAE,iBAAiB,EAC1B,OAAO,EAAE,kBAAkB,GAC1B,MAAM,CAoGR;AAED;;GAEG;AACH,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,iBAAiB,GAAG,MAAM,CAsCrE;AAED;;GAEG;AACH,wBAAgB,eAAe,CAC7B,QAAQ,EAAE,MAAM,EAChB,SAAS,EAAE,MAAM,EACjB,SAAS,EAAE,MAAM,EACjB,MAAM,EAAE,aAAa,GACpB,MAAM,CAYR;AAED;;GAEG;AACH,wBAAgB,qBAAqB,CACnC,QAAQ,EAAE,MAAM,EAChB,SAAS,EAAE,MAAM,EACjB,SAAS,EAAE,MAAM,GAChB,MAAM,CAER"}
|
package/dist/lib/results.js
CHANGED
|
@@ -99,16 +99,14 @@ export function saveResults(results, options) {
|
|
|
99
99
|
const runDir = join(evalDir, `run-${i + 1}`);
|
|
100
100
|
mkdirSync(runDir, { recursive: true });
|
|
101
101
|
// Build the result with paths and o11y summary
|
|
102
|
+
const model = results.config.model;
|
|
102
103
|
const resultWithPaths = {
|
|
103
104
|
...runData.result,
|
|
105
|
+
model,
|
|
104
106
|
};
|
|
105
107
|
// Save transcripts if available
|
|
106
108
|
if (runData.transcript) {
|
|
107
109
|
// Parse the raw transcript
|
|
108
|
-
// Model can be string or array - use first if array
|
|
109
|
-
const model = Array.isArray(results.config.model)
|
|
110
|
-
? results.config.model[0]
|
|
111
|
-
: results.config.model;
|
|
112
110
|
const transcript = parseTranscript(runData.transcript, results.config.agent, model);
|
|
113
111
|
// Save parsed transcript as primary format (transcript.json)
|
|
114
112
|
writeFileSync(join(runDir, 'transcript.json'), JSON.stringify(transcript, null, 2));
|
package/dist/lib/results.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"results.js","sourceRoot":"","sources":["../../src/lib/results.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,SAAS,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AAC9C,OAAO,EAAE,IAAI,EAAE,MAAM,MAAM,CAAC;AAC5B,OAAO,KAAK,MAAM,OAAO,CAAC;AAS1B,OAAO,EAAE,eAAe,EAAmB,MAAM,iBAAiB,CAAC;AAEnE;;GAEG;AACH,MAAM,UAAU,wBAAwB,CAAC,WAA2B;IAClE,gDAAgD;IAChD,MAAM,aAAa,GAAiC,EAAE,CAAC;IAEvD,0BAA0B;IAC1B,IAAI,WAAW,CAAC,UAAU,EAAE,MAAM,EAAE,CAAC;QACnC,aAAa,CAAC,IAAI,GAAG,WAAW,CAAC,UAAU,CAAC,MAAM,CAAC;IACrD,CAAC;IAED,qEAAqE;IACrE,IAAI,WAAW,CAAC,cAAc,IAAI,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,cAAc,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACrF,aAAa,CAAC,OAAO,GAAG,EAAE,CAAC;QAC3B,KAAK,MAAM,CAAC,IAAI,EAAE,MAAM,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,WAAW,CAAC,cAAc,CAAC,EAAE,CAAC;YACxE,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC;gBAClB,aAAa,CAAC,OAAO,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;YAC9C,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO;QACL,MAAM,EAAE;YACN,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ;YACjD,KAAK,EAAE,WAAW,CAAC,KAAK;YACxB,QAAQ,EAAE,WAAW,CAAC,QAAQ,GAAG,IAAI,EAAE,qBAAqB;SAC7D;QACD,UAAU,EAAE,WAAW,CAAC,UAAU;QAClC,aAAa,EAAE,MAAM,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,SAAS;KACjF,CAAC;AACJ,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,iBAAiB,CAAC,IAAY,EAAE,OAAsB;IACpE,MAAM,IAAI,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;IAC1C,MAAM,UAAU,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,MAAM,CAAC;IACpE,MAAM,aAAa,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC;IAEnE,OAAO;QACL,IAAI;QACJ,SAAS,EAAE,IAAI,CAAC,MAAM;QACtB,UAAU;QACV,QAAQ,EAAE,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,UAAU,GAAG,IAAI,CAAC,MAAM,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC;QAChE,YAAY,EAAE,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,aAAa,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;QAC/D,IAAI,EAAE,OAAO;KACd,CAAC;AACJ,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,uBAAuB,CACrC,MAAgC,EAChC,KAAoB,EACpB,SAAe,EACf,WAAiB;IAEjB,OAAO;QACL,SAAS,EAAE,SAAS,CAAC,WAAW,EAAE;QAClC,WAAW,EAAE,WAAW,CAAC,WAAW,EAAE;QACtC,MAAM;QACN,KAAK;KACN,CAAC;AACJ,CAAC;AAYD;;;;;;;;;;;;;;GAcG;AACH,MAAM,UAAU,WAAW,CACzB,OAA0B,EAC1B,OAA2B;IAE3B,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC,OAAO,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC;IACvD,MAAM,aAAa,GAAG,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,OAAO,CAAC,cAAc,EAAE,SAAS,CAAC,CAAC;IAElF,8BAA8B;IAC9B,SAAS,CAAC,aAAa,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAE9C,wBAAwB;IACxB,KAAK,MAAM,WAAW,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;QACxC,MAAM,OAAO,GAAG,IAAI,CAAC,aAAa,EAAE,WAAW,CAAC,IAAI,CAAC,CAAC;QACtD,SAAS,CAAC,OAAO,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAExC,8CAA8C;QAC9C,MAAM,cAAc,GAAG;YACrB,SAAS,EAAE,WAAW,CAAC,SAAS;YAChC,UAAU,EAAE,WAAW,CAAC,UAAU;YAClC,QAAQ,EAAE,GAAG,WAAW,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;YAC/C,YAAY,EAAE,WAAW,CAAC,YAAY;SACvC,CAAC;QACF,aAAa,CACX,IAAI,CAAC,OAAO,EAAE,cAAc,CAAC,EAC7B,IAAI,CAAC,SAAS,CAAC,cAAc,EAAE,IAAI,EAAE,CAAC,CAAC,CACxC,CAAC;QAEF,8BAA8B;QAC9B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,WAAW,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACjD,MAAM,OAAO,GAAG,WAAW,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;YACpC,MAAM,MAAM,GAAG,IAAI,CAAC,OAAO,EAAE,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;YAC7C,SAAS,CAAC,MAAM,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;YAEvC,+CAA+C;YAC/C,MAAM,
|
|
1
|
+
{"version":3,"file":"results.js","sourceRoot":"","sources":["../../src/lib/results.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,SAAS,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AAC9C,OAAO,EAAE,IAAI,EAAE,MAAM,MAAM,CAAC;AAC5B,OAAO,KAAK,MAAM,OAAO,CAAC;AAS1B,OAAO,EAAE,eAAe,EAAmB,MAAM,iBAAiB,CAAC;AAEnE;;GAEG;AACH,MAAM,UAAU,wBAAwB,CAAC,WAA2B;IAClE,gDAAgD;IAChD,MAAM,aAAa,GAAiC,EAAE,CAAC;IAEvD,0BAA0B;IAC1B,IAAI,WAAW,CAAC,UAAU,EAAE,MAAM,EAAE,CAAC;QACnC,aAAa,CAAC,IAAI,GAAG,WAAW,CAAC,UAAU,CAAC,MAAM,CAAC;IACrD,CAAC;IAED,qEAAqE;IACrE,IAAI,WAAW,CAAC,cAAc,IAAI,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,cAAc,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACrF,aAAa,CAAC,OAAO,GAAG,EAAE,CAAC;QAC3B,KAAK,MAAM,CAAC,IAAI,EAAE,MAAM,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,WAAW,CAAC,cAAc,CAAC,EAAE,CAAC;YACxE,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC;gBAClB,aAAa,CAAC,OAAO,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;YAC9C,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO;QACL,MAAM,EAAE;YACN,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ;YACjD,KAAK,EAAE,WAAW,CAAC,KAAK;YACxB,QAAQ,EAAE,WAAW,CAAC,QAAQ,GAAG,IAAI,EAAE,qBAAqB;SAC7D;QACD,UAAU,EAAE,WAAW,CAAC,UAAU;QAClC,aAAa,EAAE,MAAM,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,SAAS;KACjF,CAAC;AACJ,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,iBAAiB,CAAC,IAAY,EAAE,OAAsB;IACpE,MAAM,IAAI,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;IAC1C,MAAM,UAAU,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,MAAM,CAAC;IACpE,MAAM,aAAa,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC;IAEnE,OAAO;QACL,IAAI;QACJ,SAAS,EAAE,IAAI,CAAC,MAAM;QACtB,UAAU;QACV,QAAQ,EAAE,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,UAAU,GAAG,IAAI,CAAC,MAAM,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC;QAChE,YAAY,EAAE,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,aAAa,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;QAC/D,IAAI,EAAE,OAAO;KACd,CAAC;AACJ,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,uBAAuB,CACrC,MAAgC,EAChC,KAAoB,EACpB,SAAe,EACf,WAAiB;IAEjB,OAAO;QACL,SAAS,EAAE,SAAS,CAAC,WAAW,EAAE;QAClC,WAAW,EAAE,WAAW,CAAC,WAAW,EAAE;QACtC,MAAM;QACN,KAAK;KACN,CAAC;AACJ,CAAC;AAYD;;;;;;;;;;;;;;GAcG;AACH,MAAM,UAAU,WAAW,CACzB,OAA0B,EAC1B,OAA2B;IAE3B,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC,OAAO,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC;IACvD,MAAM,aAAa,GAAG,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,OAAO,CAAC,cAAc,EAAE,SAAS,CAAC,CAAC;IAElF,8BAA8B;IAC9B,SAAS,CAAC,aAAa,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAE9C,wBAAwB;IACxB,KAAK,MAAM,WAAW,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;QACxC,MAAM,OAAO,GAAG,IAAI,CAAC,aAAa,EAAE,WAAW,CAAC,IAAI,CAAC,CAAC;QACtD,SAAS,CAAC,OAAO,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAExC,8CAA8C;QAC9C,MAAM,cAAc,GAAG;YACrB,SAAS,EAAE,WAAW,CAAC,SAAS;YAChC,UAAU,EAAE,WAAW,CAAC,UAAU;YAClC,QAAQ,EAAE,GAAG,WAAW,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;YAC/C,YAAY,EAAE,WAAW,CAAC,YAAY;SACvC,CAAC;QACF,aAAa,CACX,IAAI,CAAC,OAAO,EAAE,cAAc,CAAC,EAC7B,IAAI,CAAC,SAAS,CAAC,cAAc,EAAE,IAAI,EAAE,CAAC,CAAC,CACxC,CAAC;QAEF,8BAA8B;QAC9B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,WAAW,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACjD,MAAM,OAAO,GAAG,WAAW,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;YACpC,MAAM,MAAM,GAAG,IAAI,CAAC,OAAO,EAAE,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;YAC7C,SAAS,CAAC,MAAM,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;YAEvC,+CAA+C;YAC/C,MAAM,KAAK,GAAG,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC;YACnC,MAAM,eAAe,GAAqD;gBACxE,GAAG,OAAO,CAAC,MAAM;gBACjB,KAAK;aACN,CAAC;YAEF,gCAAgC;YAChC,IAAI,OAAO,CAAC,UAAU,EAAE,CAAC;gBACvB,2BAA2B;gBAC3B,MAAM,UAAU,GAAG,eAAe,CAChC,OAAO,CAAC,UAAU,EAClB,OAAO,CAAC,MAAM,CAAC,KAAK,EACpB,KAAK,CACN,CAAC;gBAEF,6DAA6D;gBAC7D,aAAa,CACX,IAAI,CAAC,MAAM,EAAE,iBAAiB,CAAC,EAC/B,IAAI,CAAC,SAAS,CAAC,UAAU,EAAE,IAAI,EAAE,CAAC,CAAC,CACpC,CAAC;gBACF,eAAe,CAAC,cAAc,GAAG,mBAAmB,CAAC;gBAErD,2DAA2D;gBAC3D,aAAa,CAAC,IAAI,CAAC,MAAM,EAAE,sBAAsB,CAAC,EAAE,OAAO,CAAC,UAAU,CAAC,CAAC;gBACxE,eAAe,CAAC,iBAAiB,GAAG,wBAAwB,CAAC;gBAE7D,kDAAkD;gBAClD,eAAe,CAAC,IAAI,GAAG,UAAU,CAAC,OAAO,CAAC;YAC5C,CAAC;YAED,uCAAuC;YACvC,MAAM,UAAU,GAAG,IAAI,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;YAC3C,SAAS,CAAC,UAAU,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;YAE3C,IAAI,OAAO,CAAC,aAAa,EAAE,CAAC;gBAC1B,MAAM,WAAW,GAAiC,EAAE,CAAC;gBAErD,2BAA2B;gBAC3B,IAAI,OAAO,CAAC,aAAa,CAAC,IAAI,EAAE,CAAC;oBAC/B,aAAa,CAAC,IAAI,CAAC,UAAU,EAAE,UAAU,CAAC,EAAE,OAAO,CAAC,aAAa,CAAC,IAAI,CAAC,CAAC;oBACxE,WAAW,CAAC,IAAI,GAAG,oBAAoB,CAAC;gBAC1C,CAAC;gBAED,sDAAsD;gBACtD,IAAI,OAAO,CAAC,aAAa,CAAC,OAAO,EAAE,CAAC;oBAClC,WAAW,CAAC,OAAO,GAAG,EAAE,CAAC;oBACzB,KAAK,MAAM,CAAC,IAAI,EAAE,OAAO,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,aAAa,CAAC,OAAO,CAAC,EAAE,CAAC;wBAC5E,IAAI,OAAO,EAAE,CAAC;4BACZ,MAAM,QAAQ,GAAG,GAAG,IAAI,MAAM,CAAC;4BAC/B,aAAa,CAAC,IAAI,CAAC,UAAU,EAAE,QAAQ,CAAC,EAAE,OAAO,CAAC,CAAC;4BACnD,WAAW,CAAC,OAAO,CAAC,IAAI,CAAC,GAAG,aAAa,QAAQ,EAAE,CAAC;wBACtD,CAAC;oBACH,CAAC;gBACH,CAAC;gBAED,IAAI,WAAW,CAAC,IAAI,IAAI,CAAC,WAAW,CAAC,OAAO,IAAI,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,EAAE,CAAC;oBAC7F,eAAe,CAAC,WAAW,GAAG,WAAW,CAAC;gBAC5C,CAAC;YACH,CAAC;YAED,+CAA+C;YAC/C,aAAa,CACX,IAAI,CAAC,MAAM,EAAE,aAAa,CAAC,EAC3B,IAAI,CAAC,SAAS,CAAC,eAAe,EAAE,IAAI,EAAE,CAAC,CAAC,CACzC,CAAC;QACJ,CAAC;IACH,CAAC;IAED,OAAO,aAAa,CAAC;AACvB,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,kBAAkB,CAAC,OAA0B;IAC3D,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,MAAM,SAAS,GAAG,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC;IAEjC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACf,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC,CAAC;IAC7C,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC;IAClC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAEf,0BAA0B;IAC1B,MAAM,SAAS,GAAG,OAAO,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC;IACzE,MAAM,WAAW,GAAG,OAAO,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC;IAC5E,MAAM,eAAe,GAAG,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,WAAW,GAAG,SAAS,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;IAE5E,KAAK,MAAM,WAAW,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;QACxC,MAAM,QAAQ,GAAG,WAAW,CAAC,UAAU,KAAK,WAAW,CAAC,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;QAC9E,MAAM,SAAS,GAAG,WAAW,CAAC,UAAU,KAAK,WAAW,CAAC,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC;QAE7F,KAAK,CAAC,IAAI,CACR,SAAS,CACP,GAAG,QAAQ,IAAI,WAAW,CAAC,IAAI,KAAK,WAAW,CAAC,UAAU,IAAI,WAAW,CAAC,SAAS,YAAY,WAAW,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CACnI,CACF,CAAC;QACF,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,oBAAoB,WAAW,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;QACnF,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACjB,CAAC;IAED,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC;IAClC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAEf,MAAM,YAAY,GAAG,eAAe,KAAK,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,eAAe,IAAI,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC;IAC9G,KAAK,CAAC,IAAI,CAAC,YAAY,CAAC,YAAY,WAAW,IAAI,SAAS,YAAY,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC;IAEzG,MAAM,QAAQ,GAAG,CAAC,IAAI,IAAI,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,OAAO,EAAE,GAAG,IAAI,IAAI,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,OAAO,EAAE,CAAC,GAAG,IAAI,CAAC;IAC1G,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,eAAe,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAC9D,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAEf,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,eAAe,CAC7B,QAAgB,EAChB,SAAiB,EACjB,SAAiB,EACjB,MAAqB;IAErB,MAAM,IAAI,GAAG,MAAM,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;IACpD,MAAM,KAAK,GAAG,MAAM,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC;IAEnE,IAAI,IAAI,GAAG,KAAK,CAAC,GAAG,IAAI,IAAI,QAAQ,KAAK,SAAS,IAAI,SAAS,GAAG,CAAC,CAAC;IACpE,IAAI,IAAI,KAAK,CAAC,IAAI,CAAC,KAAK,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;IAExD,IAAI,MAAM,CAAC,KAAK,EAAE,CAAC;QACjB,IAAI,IAAI,KAAK,CAAC,GAAG,CAAC,MAAM,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC,MAAM,GAAG,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IAC/F,CAAC;IAED,OAAO,IAAI,CAAC;AACd,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,qBAAqB,CACnC,QAAgB,EAChB,SAAiB,EACjB,SAAiB;IAEjB,OAAO,KAAK,CAAC,IAAI,CAAC,WAAW,QAAQ,KAAK,SAAS,IAAI,SAAS,MAAM,CAAC,CAAC;AAC1E,CAAC"}
|
package/dist/lib/types.d.ts
CHANGED
|
@@ -135,6 +135,8 @@ export interface EvalRunResult {
|
|
|
135
135
|
error?: string;
|
|
136
136
|
/** Duration in seconds */
|
|
137
137
|
duration: number;
|
|
138
|
+
/** Model used for this run */
|
|
139
|
+
model?: string;
|
|
138
140
|
/** Path to parsed transcript file (relative to run directory) */
|
|
139
141
|
transcriptPath?: string;
|
|
140
142
|
/** Path to raw transcript file (relative to run directory) */
|
|
@@ -189,7 +191,7 @@ export interface ExperimentResults {
|
|
|
189
191
|
/** Timestamp when experiment completed */
|
|
190
192
|
completedAt: string;
|
|
191
193
|
/** Experiment configuration used */
|
|
192
|
-
config:
|
|
194
|
+
config: RunnableExperimentConfig;
|
|
193
195
|
/** Results for each eval */
|
|
194
196
|
evals: EvalSummary[];
|
|
195
197
|
}
|
package/dist/lib/types.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/lib/types.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH;;GAEG;AACH,MAAM,MAAM,SAAS,GACjB,+BAA+B,GAC/B,aAAa,GACb,yBAAyB,GACzB,OAAO,GACP,4BAA4B,CAAC;AAEjC;;;GAGG;AACH,MAAM,MAAM,SAAS,GAAG,MAAM,CAAC;AAE/B;;GAEG;AACH,MAAM,MAAM,UAAU,GAAG,CAAC,IAAI,EAAE,MAAM,KAAK,OAAO,CAAC;AAEnD;;;GAGG;AACH,MAAM,WAAW,OAAO;IACtB,mCAAmC;IACnC,UAAU,CACR,OAAO,EAAE,MAAM,EACf,IAAI,CAAC,EAAE,MAAM,EAAE,EACf,OAAO,CAAC,EAAE;QAAE,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;KAAE,GACzC,OAAO,CAAC;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACjE,mCAAmC;IACnC,QAAQ,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;IACxC,iCAAiC;IACjC,UAAU,CAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IACzD,wCAAwC;IACxC,mBAAmB,IAAI,MAAM,CAAC;CAC/B;AAED;;;GAGG;AACH,MAAM,MAAM,aAAa,GAAG,CAAC,OAAO,EAAE,OAAO,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;AAEhE;;GAEG;AACH,MAAM,MAAM,cAAc,GAAG,QAAQ,GAAG,QAAQ,CAAC;AAEjD;;;GAGG;AACH,MAAM,WAAW,gBAAgB;IAC/B,4BAA4B;IAC5B,KAAK,EAAE,SAAS,CAAC;IAEjB;;6FAEyF;IACzF,KAAK,CAAC,EAAE,SAAS,GAAG,SAAS,EAAE,CAAC;IAEhC,+FAA+F;IAC/F,KAAK,CAAC,EAAE,MAAM,GAAG,MAAM,EAAE,GAAG,UAAU,CAAC;IAEvC,kDAAkD;IAClD,IAAI,CAAC,EAAE,MAAM,CAAC;IAEd,qDAAqD;IACrD,SAAS,CAAC,EAAE,OAAO,CAAC;IAEpB,mEAAmE;IACnE,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;IAEnB,8EAA8E;IAC9E,OAAO,CAAC,EAAE,MAAM,CAAC;IAEjB,uEAAuE;IACvE,KAAK,CAAC,EAAE,aAAa,CAAC;IAEtB,qFAAqF;IACrF,OAAO,CAAC,EAAE,cAAc,GAAG,MAAM,CAAC;IAElC,+FAA+F;IAC/F,UAAU,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,KAAK,MAAM,CAAC;CACzC;AAED;;GAEG;AACH,MAAM,WAAW,wBAAwB;IACvC,KAAK,EAAE,SAAS,CAAC;IACjB,KAAK,EAAE,SAAS,GAAG,SAAS,EAAE,CAAC;IAC/B,KAAK,EAAE,MAAM,GAAG,MAAM,EAAE,GAAG,UAAU,CAAC;IACtC,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,OAAO,CAAC;IACnB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,CAAC,EAAE,aAAa,CAAC;IACtB,OAAO,EAAE,cAAc,GAAG,MAAM,CAAC;IACjC,UAAU,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,KAAK,MAAM,CAAC;CACzC;AAED;;GAEG;AACH,MAAM,WAAW,wBAAwB;IACvC,KAAK,EAAE,SAAS,CAAC;IACjB,KAAK,EAAE,SAAS,CAAC;IACjB,KAAK,EAAE,MAAM,GAAG,MAAM,EAAE,GAAG,UAAU,CAAC;IACtC,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,OAAO,CAAC;IACnB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,CAAC,EAAE,aAAa,CAAC;IACtB,OAAO,EAAE,cAAc,GAAG,MAAM,CAAC;IACjC,UAAU,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,KAAK,MAAM,CAAC;CACzC;AAED;;;GAGG;AACH,eAAO,MAAM,mBAAmB,mDAAoD,CAAC;AAErF;;;;GAIG;AACH,eAAO,MAAM,cAAc,uEAAwE,CAAC;AAEpG;;GAEG;AACH,MAAM,WAAW,WAAW;IAC1B,qCAAqC;IACrC,IAAI,EAAE,MAAM,CAAC;IACb,uCAAuC;IACvC,IAAI,EAAE,MAAM,CAAC;IACb,4BAA4B;IAC5B,MAAM,EAAE,MAAM,CAAC;IACf,gDAAgD;IAChD,QAAQ,EAAE,OAAO,CAAC;CACnB;AAED;;GAEG;AACH,MAAM,WAAW,aAAa;IAC5B,0BAA0B;IAC1B,MAAM,EAAE,QAAQ,GAAG,QAAQ,CAAC;IAC5B,8BAA8B;IAC9B,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,0BAA0B;IAC1B,QAAQ,EAAE,MAAM,CAAC;IACjB,iEAAiE;IACjE,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,8DAA8D;IAC9D,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,wDAAwD;IACxD,WAAW,CAAC,EAAE;QACZ,kCAAkC;QAClC,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,8DAA8D;QAC9D,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;KAClC,CAAC;CACH;AAED;;GAEG;AACH,MAAM,WAAW,WAAW;IAC1B,0DAA0D;IAC1D,MAAM,EAAE,aAAa,CAAC;IACtB,yEAAyE;IACzE,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,qDAAqD;IACrD,aAAa,CAAC,EAAE;QACd,0BAA0B;QAC1B,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,qDAAqD;QACrD,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;KAClC,CAAC;CACH;AAED;;GAEG;AACH,MAAM,WAAW,WAAW;IAC1B,uBAAuB;IACvB,IAAI,EAAE,MAAM,CAAC;IACb,2BAA2B;IAC3B,SAAS,EAAE,MAAM,CAAC;IAClB,4BAA4B;IAC5B,UAAU,EAAE,MAAM,CAAC;IACnB,gCAAgC;IAChC,QAAQ,EAAE,MAAM,CAAC;IACjB,oCAAoC;IACpC,YAAY,EAAE,MAAM,CAAC;IACrB,2EAA2E;IAC3E,IAAI,EAAE,WAAW,EAAE,CAAC;CACrB;AAED;;GAEG;AACH,MAAM,WAAW,iBAAiB;IAChC,wCAAwC;IACxC,SAAS,EAAE,MAAM,CAAC;IAClB,0CAA0C;IAC1C,WAAW,EAAE,MAAM,CAAC;IACpB,oCAAoC;IACpC,MAAM,EAAE,wBAAwB,CAAC;IACjC,4BAA4B;IAC5B,KAAK,EAAE,WAAW,EAAE,CAAC;CACtB"}
|
|
1
|
+
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/lib/types.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH;;GAEG;AACH,MAAM,MAAM,SAAS,GACjB,+BAA+B,GAC/B,aAAa,GACb,yBAAyB,GACzB,OAAO,GACP,4BAA4B,CAAC;AAEjC;;;GAGG;AACH,MAAM,MAAM,SAAS,GAAG,MAAM,CAAC;AAE/B;;GAEG;AACH,MAAM,MAAM,UAAU,GAAG,CAAC,IAAI,EAAE,MAAM,KAAK,OAAO,CAAC;AAEnD;;;GAGG;AACH,MAAM,WAAW,OAAO;IACtB,mCAAmC;IACnC,UAAU,CACR,OAAO,EAAE,MAAM,EACf,IAAI,CAAC,EAAE,MAAM,EAAE,EACf,OAAO,CAAC,EAAE;QAAE,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;KAAE,GACzC,OAAO,CAAC;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACjE,mCAAmC;IACnC,QAAQ,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;IACxC,iCAAiC;IACjC,UAAU,CAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IACzD,wCAAwC;IACxC,mBAAmB,IAAI,MAAM,CAAC;CAC/B;AAED;;;GAGG;AACH,MAAM,MAAM,aAAa,GAAG,CAAC,OAAO,EAAE,OAAO,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;AAEhE;;GAEG;AACH,MAAM,MAAM,cAAc,GAAG,QAAQ,GAAG,QAAQ,CAAC;AAEjD;;;GAGG;AACH,MAAM,WAAW,gBAAgB;IAC/B,4BAA4B;IAC5B,KAAK,EAAE,SAAS,CAAC;IAEjB;;6FAEyF;IACzF,KAAK,CAAC,EAAE,SAAS,GAAG,SAAS,EAAE,CAAC;IAEhC,+FAA+F;IAC/F,KAAK,CAAC,EAAE,MAAM,GAAG,MAAM,EAAE,GAAG,UAAU,CAAC;IAEvC,kDAAkD;IAClD,IAAI,CAAC,EAAE,MAAM,CAAC;IAEd,qDAAqD;IACrD,SAAS,CAAC,EAAE,OAAO,CAAC;IAEpB,mEAAmE;IACnE,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;IAEnB,8EAA8E;IAC9E,OAAO,CAAC,EAAE,MAAM,CAAC;IAEjB,uEAAuE;IACvE,KAAK,CAAC,EAAE,aAAa,CAAC;IAEtB,qFAAqF;IACrF,OAAO,CAAC,EAAE,cAAc,GAAG,MAAM,CAAC;IAElC,+FAA+F;IAC/F,UAAU,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,KAAK,MAAM,CAAC;CACzC;AAED;;GAEG;AACH,MAAM,WAAW,wBAAwB;IACvC,KAAK,EAAE,SAAS,CAAC;IACjB,KAAK,EAAE,SAAS,GAAG,SAAS,EAAE,CAAC;IAC/B,KAAK,EAAE,MAAM,GAAG,MAAM,EAAE,GAAG,UAAU,CAAC;IACtC,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,OAAO,CAAC;IACnB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,CAAC,EAAE,aAAa,CAAC;IACtB,OAAO,EAAE,cAAc,GAAG,MAAM,CAAC;IACjC,UAAU,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,KAAK,MAAM,CAAC;CACzC;AAED;;GAEG;AACH,MAAM,WAAW,wBAAwB;IACvC,KAAK,EAAE,SAAS,CAAC;IACjB,KAAK,EAAE,SAAS,CAAC;IACjB,KAAK,EAAE,MAAM,GAAG,MAAM,EAAE,GAAG,UAAU,CAAC;IACtC,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,OAAO,CAAC;IACnB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,CAAC,EAAE,aAAa,CAAC;IACtB,OAAO,EAAE,cAAc,GAAG,MAAM,CAAC;IACjC,UAAU,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,KAAK,MAAM,CAAC;CACzC;AAED;;;GAGG;AACH,eAAO,MAAM,mBAAmB,mDAAoD,CAAC;AAErF;;;;GAIG;AACH,eAAO,MAAM,cAAc,uEAAwE,CAAC;AAEpG;;GAEG;AACH,MAAM,WAAW,WAAW;IAC1B,qCAAqC;IACrC,IAAI,EAAE,MAAM,CAAC;IACb,uCAAuC;IACvC,IAAI,EAAE,MAAM,CAAC;IACb,4BAA4B;IAC5B,MAAM,EAAE,MAAM,CAAC;IACf,gDAAgD;IAChD,QAAQ,EAAE,OAAO,CAAC;CACnB;AAED;;GAEG;AACH,MAAM,WAAW,aAAa;IAC5B,0BAA0B;IAC1B,MAAM,EAAE,QAAQ,GAAG,QAAQ,CAAC;IAC5B,8BAA8B;IAC9B,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,0BAA0B;IAC1B,QAAQ,EAAE,MAAM,CAAC;IACjB,8BAA8B;IAC9B,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,iEAAiE;IACjE,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,8DAA8D;IAC9D,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,wDAAwD;IACxD,WAAW,CAAC,EAAE;QACZ,kCAAkC;QAClC,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,8DAA8D;QAC9D,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;KAClC,CAAC;CACH;AAED;;GAEG;AACH,MAAM,WAAW,WAAW;IAC1B,0DAA0D;IAC1D,MAAM,EAAE,aAAa,CAAC;IACtB,yEAAyE;IACzE,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,qDAAqD;IACrD,aAAa,CAAC,EAAE;QACd,0BAA0B;QAC1B,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,qDAAqD;QACrD,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;KAClC,CAAC;CACH;AAED;;GAEG;AACH,MAAM,WAAW,WAAW;IAC1B,uBAAuB;IACvB,IAAI,EAAE,MAAM,CAAC;IACb,2BAA2B;IAC3B,SAAS,EAAE,MAAM,CAAC;IAClB,4BAA4B;IAC5B,UAAU,EAAE,MAAM,CAAC;IACnB,gCAAgC;IAChC,QAAQ,EAAE,MAAM,CAAC;IACjB,oCAAoC;IACpC,YAAY,EAAE,MAAM,CAAC;IACrB,2EAA2E;IAC3E,IAAI,EAAE,WAAW,EAAE,CAAC;CACrB;AAED;;GAEG;AACH,MAAM,WAAW,iBAAiB;IAChC,wCAAwC;IACxC,SAAS,EAAE,MAAM,CAAC;IAClB,0CAA0C;IAC1C,WAAW,EAAE,MAAM,CAAC;IACpB,oCAAoC;IACpC,MAAM,EAAE,wBAAwB,CAAC;IACjC,4BAA4B;IAC5B,KAAK,EAAE,WAAW,EAAE,CAAC;CACtB"}
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@vercel/agent-eval",
|
|
3
|
-
"version": "0.0
|
|
3
|
+
"version": "0.2.0",
|
|
4
4
|
"description": "Framework for testing AI coding agents in isolated sandboxes",
|
|
5
5
|
"repository": {
|
|
6
6
|
"type": "git",
|
|
@@ -37,6 +37,7 @@
|
|
|
37
37
|
"dotenv": "^16.4.5",
|
|
38
38
|
"glob": "^11.0.0",
|
|
39
39
|
"jiti": "^2.6.1",
|
|
40
|
+
"minimatch": "^10.1.2",
|
|
40
41
|
"tar-stream": "^3.1.7",
|
|
41
42
|
"zod": "^3.23.8"
|
|
42
43
|
},
|