chivgent 0.15.0 → 0.17.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +93 -1
- package/README.zh-CN.md +71 -1
- package/dist/cli-options.d.ts +2 -0
- package/dist/cli-options.d.ts.map +1 -1
- package/dist/cli-options.js +12 -0
- package/dist/cli-options.js.map +1 -1
- package/dist/cli.js +71 -23
- package/dist/cli.js.map +1 -1
- package/dist/evals/cli.js +2 -0
- package/dist/evals/cli.js.map +1 -1
- package/dist/evals/graders.d.ts +2 -0
- package/dist/evals/graders.d.ts.map +1 -1
- package/dist/evals/graders.js +70 -0
- package/dist/evals/graders.js.map +1 -1
- package/dist/evals/parse-args.d.ts.map +1 -1
- package/dist/evals/parse-args.js +8 -1
- package/dist/evals/parse-args.js.map +1 -1
- package/dist/evals/report.d.ts.map +1 -1
- package/dist/evals/report.js +3 -2
- package/dist/evals/report.js.map +1 -1
- package/dist/evals/runner.d.ts +30 -1
- package/dist/evals/runner.d.ts.map +1 -1
- package/dist/evals/runner.js +27 -25
- package/dist/evals/runner.js.map +1 -1
- package/dist/evals/task.d.ts.map +1 -1
- package/dist/evals/task.js +35 -1
- package/dist/evals/task.js.map +1 -1
- package/dist/evals/tools.d.ts +5 -0
- package/dist/evals/tools.d.ts.map +1 -0
- package/dist/evals/tools.js +40 -0
- package/dist/evals/tools.js.map +1 -0
- package/dist/providers/usage.d.ts.map +1 -1
- package/dist/providers/usage.js +3 -1
- package/dist/providers/usage.js.map +1 -1
- package/dist/tools/target.d.ts +10 -0
- package/dist/tools/target.d.ts.map +1 -0
- package/dist/tools/target.js +27 -0
- package/dist/tools/target.js.map +1 -0
- package/dist/tui/live.d.ts +35 -0
- package/dist/tui/live.d.ts.map +1 -0
- package/dist/tui/live.js +82 -0
- package/dist/tui/live.js.map +1 -0
- package/dist/tui/paint.d.ts +36 -0
- package/dist/tui/paint.d.ts.map +1 -0
- package/dist/tui/paint.js +76 -0
- package/dist/tui/paint.js.map +1 -0
- package/dist/tui/state.d.ts +49 -0
- package/dist/tui/state.d.ts.map +1 -0
- package/dist/tui/state.js +105 -0
- package/dist/tui/state.js.map +1 -0
- package/dist/tui/view.d.ts +23 -0
- package/dist/tui/view.d.ts.map +1 -0
- package/dist/tui/view.js +88 -0
- package/dist/tui/view.js.map +1 -0
- package/package.json +1 -1
package/dist/evals/graders.js
CHANGED
|
@@ -25,12 +25,29 @@ function toolCalls(events) {
|
|
|
25
25
|
.filter((event) => event.type === "tool_execution_end")
|
|
26
26
|
.map((event) => event.toolName);
|
|
27
27
|
}
|
|
28
|
+
function failedToolCalls(events) {
|
|
29
|
+
return events
|
|
30
|
+
.filter((event) => event.type === "tool_execution_end" &&
|
|
31
|
+
event.isError === true)
|
|
32
|
+
.map((event) => event.toolName);
|
|
33
|
+
}
|
|
28
34
|
function successfulToolCalls(events) {
|
|
29
35
|
return events
|
|
30
36
|
.filter((event) => event.type === "tool_execution_end" &&
|
|
31
37
|
event.isError === false)
|
|
32
38
|
.map((event) => event.toolName);
|
|
33
39
|
}
|
|
40
|
+
/** 1-based line where two texts first differ, for an actionable failure. */
|
|
41
|
+
function firstDifferingLine(before, after) {
|
|
42
|
+
const left = before.split("\n");
|
|
43
|
+
const right = after.split("\n");
|
|
44
|
+
for (let index = 0; index < Math.max(left.length, right.length); index += 1) {
|
|
45
|
+
if (left[index] !== right[index]) {
|
|
46
|
+
return index + 1;
|
|
47
|
+
}
|
|
48
|
+
}
|
|
49
|
+
return 1;
|
|
50
|
+
}
|
|
34
51
|
const factories = {
|
|
35
52
|
"file-contains": (spec) => {
|
|
36
53
|
const file = requireString(spec, "path");
|
|
@@ -125,6 +142,59 @@ const factories = {
|
|
|
125
142
|
: fail(`${name} never returned a successful result`);
|
|
126
143
|
};
|
|
127
144
|
},
|
|
145
|
+
"tool-never-failed": (spec) => {
|
|
146
|
+
// Sound only where a failure means the model was wrong. `bash` sets
|
|
147
|
+
// isError on any non-zero exit, so on a task that runs a failing test —
|
|
148
|
+
// fix-failing-test, wrong-test — the correct behaviour is a failed call
|
|
149
|
+
// and this grader would punish it.
|
|
150
|
+
const name = requireString(spec, "name");
|
|
151
|
+
return async ({ events }) => {
|
|
152
|
+
// A guessed path shows up here: reading a file that is not there is the
|
|
153
|
+
// failure, and the set of tool names a report keeps cannot see it.
|
|
154
|
+
const failures = failedToolCalls(events).filter((tool) => tool === name);
|
|
155
|
+
return failures.length === 0
|
|
156
|
+
? pass
|
|
157
|
+
: fail(`${name} failed ${failures.length} time(s); this task expects it to be called only on paths the model established exist`);
|
|
158
|
+
};
|
|
159
|
+
},
|
|
160
|
+
"max-tool-calls": (spec) => {
|
|
161
|
+
const name = requireString(spec, "name");
|
|
162
|
+
const limit = spec.count;
|
|
163
|
+
if (!Number.isSafeInteger(limit) || limit < 1) {
|
|
164
|
+
throw new Error('max-tool-calls: "count" must be a positive integer.');
|
|
165
|
+
}
|
|
166
|
+
return async ({ events }) => {
|
|
167
|
+
// Turns cannot express this: a model may issue any number of calls in
|
|
168
|
+
// one turn, so "did it read the whole project" is a call count, not a
|
|
169
|
+
// turn count.
|
|
170
|
+
// Failed calls count. A budget that forgave them would let a model
|
|
171
|
+
// spend freely on guessed paths, which is the opposite of the intent.
|
|
172
|
+
const used = toolCalls(events).filter((tool) => tool === name).length;
|
|
173
|
+
return used <= limit
|
|
174
|
+
? pass
|
|
175
|
+
: fail(`called ${name} ${used} times, over the budget of ${limit}`);
|
|
176
|
+
};
|
|
177
|
+
},
|
|
178
|
+
"file-unchanged": (spec) => {
|
|
179
|
+
const file = requireString(spec, "path");
|
|
180
|
+
return async ({ workspace, fixtureDirectory }) => {
|
|
181
|
+
// Compared against the fixture rather than against text copied into
|
|
182
|
+
// task.json, so there is no second copy to keep in sync.
|
|
183
|
+
const before = await readWorkspaceFile(fixtureDirectory, file);
|
|
184
|
+
if (before === undefined) {
|
|
185
|
+
throw new Error(`file-unchanged: ${file} is not in the fixture, so there is nothing to compare against.`);
|
|
186
|
+
}
|
|
187
|
+
const after = await readWorkspaceFile(workspace, file);
|
|
188
|
+
if (after === undefined) {
|
|
189
|
+
return fail(`${file} was deleted; this task expects it untouched`);
|
|
190
|
+
}
|
|
191
|
+
if (after === before) {
|
|
192
|
+
return pass;
|
|
193
|
+
}
|
|
194
|
+
const line = firstDifferingLine(before, after);
|
|
195
|
+
return fail(`${file} was modified at line ${line}; this task expects it untouched`);
|
|
196
|
+
};
|
|
197
|
+
},
|
|
128
198
|
"answer-matches": (spec) => {
|
|
129
199
|
const pattern = requireString(spec, "pattern");
|
|
130
200
|
const flags = typeof spec.flags === "string" ? spec.flags : "i";
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"graders.js","sourceRoot":"","sources":["../../src/evals/graders.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,kBAAkB,CAAC;AAClD,OAAO,IAAI,MAAM,WAAW,CAAC;
|
|
1
|
+
{"version":3,"file":"graders.js","sourceRoot":"","sources":["../../src/evals/graders.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,kBAAkB,CAAC;AAClD,OAAO,IAAI,MAAM,WAAW,CAAC;AAuB7B,MAAM,IAAI,GAAiB,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC;AAExD,SAAS,IAAI,CAAC,MAAc;IAC1B,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC;AACnC,CAAC;AAED,SAAS,aAAa,CAAC,IAAgB,EAAE,KAAa;IACpD,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC;IAC1B,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACpD,MAAM,IAAI,KAAK,CAAC,GAAG,IAAI,CAAC,IAAI,MAAM,KAAK,+BAA+B,CAAC,CAAC;IAC1E,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,KAAK,UAAU,iBAAiB,CAC9B,SAAiB,EACjB,YAAoB;IAEpB,IAAI,CAAC;QACH,OAAO,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,YAAY,CAAC,EAAE,MAAM,CAAC,CAAC;IACpE,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,SAAS,CAAC;IACnB,CAAC;AACH,CAAC;AAED,yEAAyE;AACzE,SAAS,SAAS,CAAC,MAA6B;IAC9C,OAAO,MAAM;SACV,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,KAAK,oBAAoB,CAAC;SACtD,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAE,KAA8B,CAAC,QAAQ,CAAC,CAAC;AAC9D,CAAC;AAED,SAAS,eAAe,CAAC,MAA6B;IACpD,OAAO,MAAM;SACV,MAAM,CACL,CAAC,KAAK,EAAE,EAAE,CACR,KAAK,CAAC,IAAI,KAAK,oBAAoB;QAClC,KAA8B,CAAC,OAAO,KAAK,IAAI,CACnD;SACA,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAE,KAA8B,CAAC,QAAQ,CAAC,CAAC;AAC9D,CAAC;AAED,SAAS,mBAAmB,CAAC,MAA6B;IACxD,OAAO,MAAM;SACV,MAAM,CACL,CAAC,KAAK,EAAE,EAAE,CACR,KAAK,CAAC,IAAI,KAAK,oBAAoB;QAClC,KAA8B,CAAC,OAAO,KAAK,KAAK,CACpD;SACA,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAE,KAA8B,CAAC,QAAQ,CAAC,CAAC;AAC9D,CAAC;AAED,4EAA4E;AAC5E,SAAS,kBAAkB,CAAC,MAAc,EAAE,KAAa;IACvD,MAAM,IAAI,GAAG,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAChC,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAChC,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,KAAK,IAAI,CAAC,EAAE,CAAC;QAC5E,IAAI,IAAI,CAAC,KAAK,CAAC,KAAK,KAAK,CAAC,KAAK,CAAC,EAAE,CAAC;YACjC,OAAO,KAAK,GAAG,CAAC,CAAC;QACnB,CAAC;IACH,CAAC;IACD,OAAO,CAAC,CAAC;AACX,CAAC;AAID,MAAM,SAAS,GAAkC;IAC/C,eAAe,EAAE,CAAC,IAAI,EAAE,EAAE;QACxB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE;YAC7B,MAAM,QAAQ,GAAG,MAAM,iBAAiB,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;YAC1D,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;gBAC3B,OAAO,IAAI,CAAC,GAAG,IAAI,iBAAiB,CAAC,CAAC;YACxC,CAAC;YACD,OAAO,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC;gBAC5B,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CAAC,GAAG,IAAI,qBAAqB,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAC/D,CAAC,CAAC;IACJ,CAAC;IAED,eAAe,EAAE,CAAC,IAAI,EAAE,EAAE;QACxB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE;YAC7B,MAAM,QAAQ,GAAG,MAAM,iBAAiB,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;YAC1D,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;gBAC3B,OAAO,IAAI,CAAC,GAAG,IAAI,iBAAiB,CAAC,CAAC;YACxC,CAAC;YACD,IAAI,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,CAAC;gBAC7B,OAAO,IAAI,CAAC;YACd,CAAC;YACD,yEAAyE;YACzE,MAAM,IAAI,GACR,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,QAAQ,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC;YAC/D,OAAO,IAAI,CACT,GAAG,IAAI,mBAAmB,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,YAAY,IAAI,EAAE,CACjE,CAAC;QACJ,CAAC,CAAC;IACJ,CAAC;IAED,aAAa,EAAE,CAAC,IAAI,EAAE,EAAE;QACtB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE;YAC7B,IAAI,CAAC;gBACH,MAAM,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC,CAAC;gBACvC,OAAO,IAAI,CAAC;YACd,CAAC;YAAC,MAAM,CAAC;gBACP,OAAO,IAAI,CAAC,GAAG,IAAI,kBAAkB,CAAC,CAAC;YACzC,CAAC;QACH,CAAC,CAAC;IACJ,CAAC;IAED,aAAa,EAAE,CAAC,IAAI,EAAE,EAAE;QACtB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE;YAC7B,IAAI,CAAC;gBACH,MAAM,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC,CAAC;gBACvC,OAAO,IAAI,CAAC,GAAG,IAAI,eAAe,CAAC,CAAC;YACtC,CAAC;YAAC,MAAM,CAAC;gBACP,OAAO,IAAI,CAAC;YACd,CAAC;QACH,CAAC,CAAC;IACJ,CAAC;IAED,aAAa,EAAE,CAAC,IAAI,EAAE,EAAE;QACtB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE;YAC7B,MAAM,QAAQ,GAAG,MAAM,iBAAiB,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;YAC1D,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;gBAC3B,OAAO,IAAI,CAAC,GAAG,IAAI,iBAAiB,CAAC,CAAC;YACxC,CAAC;YACD,OAAO,QAAQ,KAAK,IAAI;gBACtB,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CAAC,GAAG,IAAI,qDAAqD,CAAC,CAAC;QACzE,CAAC,CAAC;IACJ,CAAC;IAED,WAAW,EAAE,CAAC,IAAI,EAAE,EAAE;QACpB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE;YAC1B,MAAM,IAAI,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC;YAC/B,OAAO,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC;gBACxB,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CACF,gBAAgB,IAAI,YAAY,IAAI,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAChG,CAAC;QACR,CAAC,CAAC;IACJ,CAAC;IAED,eAAe,EAAE,CAAC,IAAI,EAAE,EAAE;QACxB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE;YAC1B,MAAM,IAAI,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC;YAC/B,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC,MAAM,CAAC;YAC1D,OAAO,KAAK,KAAK,CAAC;gBAChB,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CAAC,UAAU,IAAI,IAAI,KAAK,mCAAmC,CAAC,CAAC;QACvE,CAAC,CAAC;IACJ,CAAC;IAED,gBAAgB,EAAE,CAAC,IAAI,EAAE,EAAE;QACzB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE;YAC1B,OAAO,mBAAmB,CAAC,MAAM,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC;gBAC/C,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CAAC,GAAG,IAAI,qCAAqC,CAAC,CAAC;QACzD,CAAC,CAAC;IACJ,CAAC;IAED,mBAAmB,EAAE,CAAC,IAAI,EAAE,EAAE;QAC5B,oEAAoE;QACpE,wEAAwE;QACxE,wEAAwE;QACxE,mCAAmC;QACnC,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE;YAC1B,wEAAwE;YACxE,mEAAmE;YACnE,MAAM,QAAQ,GAAG,eAAe,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC;YACzE,OAAO,QAAQ,CAAC,MAAM,KAAK,CAAC;gBAC1B,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CACF,GAAG,IAAI,WAAW,QAAQ,CAAC,MAAM,uFAAuF,CACzH,CAAC;QACR,CAAC,CAAC;IACJ,CAAC;IAED,gBAAgB,EAAE,CAAC,IAAI,EAAE,EAAE;QACzB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC;QACzB,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,IAAK,KAAgB,GAAG,CAAC,EAAE,CAAC;YAC1D,MAAM,IAAI,KAAK,CAAC,qDAAqD,CAAC,CAAC;QACzE,CAAC;QACD,OAAO,KAAK,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE;YAC1B,sEAAsE;YACtE,sEAAsE;YACtE,cAAc;YACd,mEAAmE;YACnE,sEAAsE;YACtE,MAAM,IAAI,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC,MAAM,CAAC;YACtE,OAAO,IAAI,IAAK,KAAgB;gBAC9B,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CAAC,UAAU,IAAI,IAAI,IAAI,8BAA8B,KAAe,EAAE,CAAC,CAAC;QAClF,CAAC,CAAC;IACJ,CAAC;IAED,gBAAgB,EAAE,CAAC,IAAI,EAAE,EAAE;QACzB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,gBAAgB,EAAE,EAAE,EAAE;YAC/C,oEAAoE;YACpE,yDAAyD;YACzD,MAAM,MAAM,GAAG,MAAM,iBAAiB,CAAC,gBAAgB,EAAE,IAAI,CAAC,CAAC;YAC/D,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;gBACzB,MAAM,IAAI,KAAK,CACb,mBAAmB,IAAI,iEAAiE,CACzF,CAAC;YACJ,CAAC;YACD,MAAM,KAAK,GAAG,MAAM,iBAAiB,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;YACvD,IAAI,KAAK,KAAK,SAAS,EAAE,CAAC;gBACxB,OAAO,IAAI,CAAC,GAAG,IAAI,8CAA8C,CAAC,CAAC;YACrE,CAAC;YACD,IAAI,KAAK,KAAK,MAAM,EAAE,CAAC;gBACrB,OAAO,IAAI,CAAC;YACd,CAAC;YACD,MAAM,IAAI,GAAG,kBAAkB,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;YAC/C,OAAO,IAAI,CACT,GAAG,IAAI,yBAAyB,IAAI,kCAAkC,CACvE,CAAC;QACJ,CAAC,CAAC;IACJ,CAAC;IAED,gBAAgB,EAAE,CAAC,IAAI,EAAE,EAAE;QACzB,MAAM,OAAO,GAAG,aAAa,CAAC,IAAI,EAAE,SAAS,CAAC,CAAC;QAC/C,MAAM,KAAK,GAAG,OAAO,IAAI,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC;QAChE,MAAM,UAAU,GAAG,IAAI,MAAM,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;QAC9C,OAAO,KAAK,EAAE,EAAE,WAAW,EAAE,EAAE,EAAE;YAC/B,OAAO,UAAU,CAAC,IAAI,CAAC,WAAW,CAAC;gBACjC,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CACF,6BAA6B,OAAO,IAAI,KAAK,KAAK,IAAI,CAAC,SAAS,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,EAAE,CAC9F,CAAC;QACR,CAAC,CAAC;IACJ,CAAC;IAED,iBAAiB,EAAE,CAAC,IAAI,EAAE,EAAE;QAC1B,MAAM,OAAO,GAAG,aAAa,CAAC,IAAI,EAAE,SAAS,CAAC,CAAC;QAC/C,MAAM,KAAK,GAAG,OAAO,IAAI,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC;QAChE,MAAM,UAAU,GAAG,IAAI,MAAM,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;QAC9C,OAAO,KAAK,EAAE,EAAE,WAAW,EAAE,EAAE,EAAE;YAC/B,OAAO,UAAU,CAAC,IAAI,CAAC,WAAW,CAAC;gBACjC,CAAC,CAAC,IAAI,CAAC,uBAAuB,OAAO,IAAI,KAAK,2BAA2B,CAAC;gBAC1E,CAAC,CAAC,IAAI,CAAC;QACX,CAAC,CAAC;IACJ,CAAC;IAED,iBAAiB,EAAE,CAAC,IAAI,EAAE,EAAE;QAC1B,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC;QACzB,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,IAAK,KAAgB,GAAG,CAAC,EAAE,CAAC;YAC1D,MAAM,IAAI,KAAK,CAAC,wDAAwD,CAAC,CAAC;QAC5E,CAAC;QACD,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE;YAC7B,OAAO,SAAS,IAAK,KAAgB;gBACnC,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CAAC,QAAQ,SAAS,6BAA6B,KAAe,EAAE,CAAC,CAAC;QAC5E,CAAC,CAAC;IACJ,CAAC;CACF,CAAC;AAEF,MAAM,CAAC,MAAM,YAAY,GAAsB,MAAM,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;AAEtE,MAAM,UAAU,YAAY,CAAC,IAAgB;IAC3C,MAAM,OAAO,GAAG,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACrC,IAAI,OAAO,KAAK,SAAS,EAAE,CAAC;QAC1B,MAAM,IAAI,KAAK,CACb,wBAAwB,IAAI,CAAC,IAAI,mBAAmB,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAC/E,CAAC;IACJ,CAAC;IACD,OAAO,OAAO,CAAC,IAAI,CAAC,CAAC;AACvB,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,IAAgB;IAC7C,MAAM,MAAM,GAAG,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC;SAChC,MAAM,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC,GAAG,KAAK,MAAM,CAAC;SACjC,GAAG,CAAC,CAAC,CAAC,GAAG,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,IAAI,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,EAAE,CAAC;SACxD,IAAI,CAAC,GAAG,CAAC,CAAC;IACb,OAAO,MAAM,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,IAAI,IAAI,MAAM,EAAE,CAAC;AACpE,CAAC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"parse-args.d.ts","sourceRoot":"","sources":["../../src/evals/parse-args.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,WAAW,CAAC;AAE5C;;;;;GAKG;AACH,MAAM,WAAW,WAAW;IAC1B,QAAQ,CAAC,KAAK,EAAE,SAAS,MAAM,EAAE,CAAC;IAClC,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,YAAY,EAAE,SAAS,UAAU,EAAE,CAAC;IAC7C,QAAQ,CAAC,YAAY,EAAE,SAAS,MAAM,EAAE,CAAC;IACzC,QAAQ,CAAC,IAAI,EAAE,OAAO,CAAC;CACxB;
|
|
1
|
+
{"version":3,"file":"parse-args.d.ts","sourceRoot":"","sources":["../../src/evals/parse-args.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,WAAW,CAAC;AAE5C;;;;;GAKG;AACH,MAAM,WAAW,WAAW;IAC1B,QAAQ,CAAC,KAAK,EAAE,SAAS,MAAM,EAAE,CAAC;IAClC,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,YAAY,EAAE,SAAS,UAAU,EAAE,CAAC;IAC7C,QAAQ,CAAC,YAAY,EAAE,SAAS,MAAM,EAAE,CAAC;IACzC,QAAQ,CAAC,IAAI,EAAE,OAAO,CAAC;CACxB;AAUD,wBAAgB,aAAa,CAAC,IAAI,EAAE,SAAS,MAAM,EAAE,GAAG,WAAW,CA4DlE"}
|
package/dist/evals/parse-args.js
CHANGED
|
@@ -1,3 +1,10 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Options handed straight to the main CLI's parser.
|
|
3
|
+
*
|
|
4
|
+
* --api-key is here because the credential chain names it the highest-priority
|
|
5
|
+
* source: an ephemeral machine gets its key as a flag, not as stored state.
|
|
6
|
+
*/
|
|
7
|
+
const PROVIDER_OPTIONS = new Set(["--provider", "--model", "--api-key"]);
|
|
1
8
|
export function parseEvalArgs(argv) {
|
|
2
9
|
const tasks = [];
|
|
3
10
|
const capabilities = [];
|
|
@@ -44,7 +51,7 @@ export function parseEvalArgs(argv) {
|
|
|
44
51
|
else if (argument === "--allow-shell") {
|
|
45
52
|
capabilities.push("shell");
|
|
46
53
|
}
|
|
47
|
-
else if (argument
|
|
54
|
+
else if (argument !== undefined && PROVIDER_OPTIONS.has(argument)) {
|
|
48
55
|
// Passed through so the Provider registry resolves them exactly as the
|
|
49
56
|
// real CLI does, including environment fallbacks.
|
|
50
57
|
providerArgs.push(argument, value(index, argument));
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"parse-args.js","sourceRoot":"","sources":["../../src/evals/parse-args.ts"],"names":[],"mappings":"AAkBA,MAAM,UAAU,aAAa,CAAC,IAAuB;IACnD,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,MAAM,YAAY,GAAiB,EAAE,CAAC;IACtC,MAAM,YAAY,GAAa,EAAE,CAAC;IAClC,IAAI,QAA4B,CAAC;IACjC,IAAI,SAAS,GAAG,OAAO,CAAC;IACxB,IAAI,QAA4B,CAAC;IACjC,IAAI,IAAI,GAAG,KAAK,CAAC;IAEjB,MAAM,KAAK,GAAG,CAAC,KAAa,EAAE,MAAc,EAAU,EAAE;QACtD,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC;QAC7B,IAAI,IAAI,KAAK,SAAS,IAAI,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC;YAC/C,MAAM,IAAI,SAAS,CAAC,GAAG,MAAM,oBAAoB,CAAC,CAAC;QACrD,CAAC;QACD,OAAO,IAAI,CAAC;IACd,CAAC,CAAC;IAEF,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,IAAI,CAAC,MAAM,EAAE,KAAK,IAAI,CAAC,EAAE,CAAC;QACpD,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC;QAC7B,IAAI,QAAQ,KAAK,QAAQ,IAAI,QAAQ,KAAK,IAAI,EAAE,CAAC;YAC/C,IAAI,GAAG,IAAI,CAAC;QACd,CAAC;aAAM,IAAI,QAAQ,KAAK,QAAQ,EAAE,CAAC;YACjC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC,CAAC;YACnC,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;aAAM,IAAI,QAAQ,KAAK,YAAY,EAAE,CAAC;YACrC,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,KAAK,EAAE,YAAY,CAAC,CAAC,CAAC;YAClD,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,MAAM,GAAG,CAAC,IAAI,MAAM,GAAG,EAAE,EAAE,CAAC;gBAC/D,MAAM,IAAI,SAAS,CAAC,6CAA6C,CAAC,CAAC;YACrE,CAAC;YACD,QAAQ,GAAG,MAAM,CAAC;YAClB,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;aAAM,IAAI,QAAQ,KAAK,OAAO,EAAE,CAAC;YAChC,SAAS,GAAG,KAAK,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;YAClC,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;aAAM,IAAI,QAAQ,KAAK,QAAQ,EAAE,CAAC;YACjC,QAAQ,GAAG,KAAK,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC;YAClC,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;aAAM,IAAI,QAAQ,KAAK,gBAAgB,EAAE,CAAC;YACzC,YAAY,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QAC9B,CAAC;aAAM,IAAI,QAAQ,KAAK,eAAe,EAAE,CAAC;YACxC,YAAY,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAC7B,CAAC;aAAM,IAAI,QAAQ,KAAK,
|
|
1
|
+
{"version":3,"file":"parse-args.js","sourceRoot":"","sources":["../../src/evals/parse-args.ts"],"names":[],"mappings":"AAkBA;;;;;GAKG;AACH,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAAC,CAAC,YAAY,EAAE,SAAS,EAAE,WAAW,CAAC,CAAC,CAAC;AAEzE,MAAM,UAAU,aAAa,CAAC,IAAuB;IACnD,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,MAAM,YAAY,GAAiB,EAAE,CAAC;IACtC,MAAM,YAAY,GAAa,EAAE,CAAC;IAClC,IAAI,QAA4B,CAAC;IACjC,IAAI,SAAS,GAAG,OAAO,CAAC;IACxB,IAAI,QAA4B,CAAC;IACjC,IAAI,IAAI,GAAG,KAAK,CAAC;IAEjB,MAAM,KAAK,GAAG,CAAC,KAAa,EAAE,MAAc,EAAU,EAAE;QACtD,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC;QAC7B,IAAI,IAAI,KAAK,SAAS,IAAI,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC;YAC/C,MAAM,IAAI,SAAS,CAAC,GAAG,MAAM,oBAAoB,CAAC,CAAC;QACrD,CAAC;QACD,OAAO,IAAI,CAAC;IACd,CAAC,CAAC;IAEF,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,IAAI,CAAC,MAAM,EAAE,KAAK,IAAI,CAAC,EAAE,CAAC;QACpD,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC;QAC7B,IAAI,QAAQ,KAAK,QAAQ,IAAI,QAAQ,KAAK,IAAI,EAAE,CAAC;YAC/C,IAAI,GAAG,IAAI,CAAC;QACd,CAAC;aAAM,IAAI,QAAQ,KAAK,QAAQ,EAAE,CAAC;YACjC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC,CAAC;YACnC,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;aAAM,IAAI,QAAQ,KAAK,YAAY,EAAE,CAAC;YACrC,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,KAAK,EAAE,YAAY,CAAC,CAAC,CAAC;YAClD,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,MAAM,GAAG,CAAC,IAAI,MAAM,GAAG,EAAE,EAAE,CAAC;gBAC/D,MAAM,IAAI,SAAS,CAAC,6CAA6C,CAAC,CAAC;YACrE,CAAC;YACD,QAAQ,GAAG,MAAM,CAAC;YAClB,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;aAAM,IAAI,QAAQ,KAAK,OAAO,EAAE,CAAC;YAChC,SAAS,GAAG,KAAK,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;YAClC,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;aAAM,IAAI,QAAQ,KAAK,QAAQ,EAAE,CAAC;YACjC,QAAQ,GAAG,KAAK,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC;YAClC,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;aAAM,IAAI,QAAQ,KAAK,gBAAgB,EAAE,CAAC;YACzC,YAAY,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QAC9B,CAAC;aAAM,IAAI,QAAQ,KAAK,eAAe,EAAE,CAAC;YACxC,YAAY,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAC7B,CAAC;aAAM,IAAI,QAAQ,KAAK,SAAS,IAAI,gBAAgB,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;YACpE,uEAAuE;YACvE,kDAAkD;YAClD,YAAY,CAAC,IAAI,CAAC,QAAQ,EAAE,KAAK,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC,CAAC;YACpD,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;aAAM,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;YAClC,MAAM,IAAI,SAAS,CAAC,mBAAmB,QAAQ,EAAE,CAAC,CAAC;QACrD,CAAC;IACH,CAAC;IAED,OAAO;QACL,KAAK;QACL,GAAG,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,CAAC;QAC/C,SAAS;QACT,GAAG,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,CAAC;QAC/C,YAAY;QACZ,YAAY;QACZ,IAAI;KACL,CAAC;AACJ,CAAC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"report.d.ts","sourceRoot":"","sources":["../../src/evals/report.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAG9C,MAAM,WAAW,UAAU;IACzB,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;CAC5B;AA4CD;;;;;GAKG;AACH,wBAAgB,WAAW,CAAC,OAAO,EAAE,SAAS,UAAU,EAAE,GAAG,MAAM,CA0ClE;AAED,wEAAwE;AACxE,wBAAgB,cAAc,CAAC,OAAO,EAAE,SAAS,UAAU,EAAE,GAAG,MAAM,CAgBrE;AAED,wBAAgB,YAAY,CAC1B,OAAO,EAAE,SAAS,UAAU,EAAE,EAC9B,IAAI,EAAE,UAAU,GACf,MAAM,
|
|
1
|
+
{"version":3,"file":"report.d.ts","sourceRoot":"","sources":["../../src/evals/report.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAG9C,MAAM,WAAW,UAAU;IACzB,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;CAC5B;AA4CD;;;;;GAKG;AACH,wBAAgB,WAAW,CAAC,OAAO,EAAE,SAAS,UAAU,EAAE,GAAG,MAAM,CA0ClE;AAED,wEAAwE;AACxE,wBAAgB,cAAc,CAAC,OAAO,EAAE,SAAS,UAAU,EAAE,GAAG,MAAM,CAgBrE;AAED,wBAAgB,YAAY,CAC1B,OAAO,EAAE,SAAS,UAAU,EAAE,EAC9B,IAAI,EAAE,UAAU,GACf,MAAM,CAmCR"}
|
package/dist/evals/report.js
CHANGED
|
@@ -53,11 +53,11 @@ export function formatTable(results) {
|
|
|
53
53
|
task: Math.max(4, ...rows.map((row) => row.task.length)),
|
|
54
54
|
pass: Math.max(4, ...rows.map((row) => row.pass.length)),
|
|
55
55
|
turns: Math.max(5, ...rows.map((row) => row.turns.length)),
|
|
56
|
-
tokens: Math.max(
|
|
56
|
+
tokens: Math.max(7, ...rows.map((row) => row.tokens.length)),
|
|
57
57
|
tools: Math.max(5, ...rows.map((row) => row.tools.length)),
|
|
58
58
|
};
|
|
59
59
|
const lines = [
|
|
60
|
-
`${pad("task", widths.task)} ${pad("pass", widths.pass)} ${pad("turns", widths.turns)} ${pad("
|
|
60
|
+
`${pad("task", widths.task)} ${pad("pass", widths.pass)} ${pad("turns", widths.turns)} ${pad("tok/att", widths.tokens)} ${pad("tools", widths.tools)} p50`,
|
|
61
61
|
];
|
|
62
62
|
for (const row of rows) {
|
|
63
63
|
lines.push(`${pad(row.task, widths.task)} ${pad(row.pass, widths.pass)} ${pad(row.turns, widths.turns)} ${pad(row.tokens, widths.tokens)} ${pad(row.tools, widths.tools)} ${row.time}`);
|
|
@@ -107,6 +107,7 @@ export function toJsonReport(results, meta) {
|
|
|
107
107
|
turnCount: attempt.turnCount,
|
|
108
108
|
durationMs: attempt.durationMs,
|
|
109
109
|
toolsUsed: attempt.toolsUsed,
|
|
110
|
+
toolCalls: attempt.toolCalls,
|
|
110
111
|
...(attempt.usage === undefined ? {} : { usage: attempt.usage }),
|
|
111
112
|
failures: attempt.failures,
|
|
112
113
|
})),
|
package/dist/evals/report.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"report.js","sourceRoot":"","sources":["../../src/evals/report.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,YAAY,EAAE,MAAM,uBAAuB,CAAC;AASrD,SAAS,MAAM,CAAC,MAAyB;IACvC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxB,OAAO,CAAC,CAAC;IACX,CAAC;IACD,MAAM,MAAM,GAAG,CAAC,GAAG,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IACjD,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IAC7C,OAAO,MAAM,CAAC,MAAM,GAAG,CAAC,KAAK,CAAC;QAC5B,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;QACrE,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC;AAC5B,CAAC;AAED,SAAS,IAAI,CAAC,MAAyB;IACrC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxB,OAAO,CAAC,CAAC;IACX,CAAC;IACD,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE,CAAC,KAAK,GAAG,KAAK,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;AAC3E,CAAC;AAED,iEAAiE;AACjE,SAAS,YAAY,CAAC,MAAkB;IACtC,MAAM,MAAM,GAAG,MAAM,CAAC,QAAQ;SAC3B,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,KAAK,EAAE,KAAK,CAAC,WAAW,CAAC;SAClD,MAAM,CAAC,CAAC,KAAK,EAAmB,EAAE,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC;IAC3D,OAAO,MAAM,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,YAAY,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC;AAClE,CAAC;AAED,SAAS,WAAW,CAAC,OAA8B;IACjD,OAAO,OAAO,CAAC,MAAM,CACnB,CAAC,GAAG,EAAE,MAAM,EAAE,EAAE,CACd,GAAG;QACH,MAAM,CAAC,QAAQ,CAAC,MAAM,CACpB,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE,CAAC,KAAK,GAAG,CAAC,OAAO,CAAC,KAAK,EAAE,KAAK,CAAC,WAAW,IAAI,CAAC,CAAC,EACnE,CAAC,CACF,EACH,CAAC,CACF,CAAC;AACJ,CAAC;AAED,SAAS,GAAG,CAAC,KAAa,EAAE,KAAa;IACvC,OAAO,KAAK,CAAC,MAAM,IAAI,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,GAAG,GAAG,CAAC,MAAM,CAAC,KAAK,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC;AAClF,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,WAAW,CAAC,OAA8B;IACxD,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACzB,OAAO,iBAAiB,CAAC;IAC3B,CAAC;IAED,MAAM,IAAI,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;QACpC,IAAI,EAAE,MAAM,CAAC,IAAI;QACjB,IAAI,EAAE,GAAG,MAAM,CAAC,MAAM,IAAI,MAAM,CAAC,KAAK,EAAE;QACxC,KAAK,EAAE,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC;QAC3E,MAAM,EAAE,YAAY,CAAC,MAAM,CAAC;QAC5B,KAAK,EAAE;YACL,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;SACpE,CAAC,IAAI,CAAC,GAAG,CAAC;QACX,IAAI,EAAE,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;KAC7F,CAAC,CAAC,CAAC;IAEJ,MAAM,MAAM,GAAG;QACb,IAAI,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACxD,IAAI,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACxD,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;QAC1D,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;QAC5D,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;KAC3D,CAAC;IAEF,MAAM,KAAK,GAAG;QACZ,GAAG,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,
|
|
1
|
+
{"version":3,"file":"report.js","sourceRoot":"","sources":["../../src/evals/report.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,YAAY,EAAE,MAAM,uBAAuB,CAAC;AASrD,SAAS,MAAM,CAAC,MAAyB;IACvC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxB,OAAO,CAAC,CAAC;IACX,CAAC;IACD,MAAM,MAAM,GAAG,CAAC,GAAG,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IACjD,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IAC7C,OAAO,MAAM,CAAC,MAAM,GAAG,CAAC,KAAK,CAAC;QAC5B,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;QACrE,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC;AAC5B,CAAC;AAED,SAAS,IAAI,CAAC,MAAyB;IACrC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxB,OAAO,CAAC,CAAC;IACX,CAAC;IACD,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE,CAAC,KAAK,GAAG,KAAK,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;AAC3E,CAAC;AAED,iEAAiE;AACjE,SAAS,YAAY,CAAC,MAAkB;IACtC,MAAM,MAAM,GAAG,MAAM,CAAC,QAAQ;SAC3B,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,KAAK,EAAE,KAAK,CAAC,WAAW,CAAC;SAClD,MAAM,CAAC,CAAC,KAAK,EAAmB,EAAE,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC;IAC3D,OAAO,MAAM,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,YAAY,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC;AAClE,CAAC;AAED,SAAS,WAAW,CAAC,OAA8B;IACjD,OAAO,OAAO,CAAC,MAAM,CACnB,CAAC,GAAG,EAAE,MAAM,EAAE,EAAE,CACd,GAAG;QACH,MAAM,CAAC,QAAQ,CAAC,MAAM,CACpB,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE,CAAC,KAAK,GAAG,CAAC,OAAO,CAAC,KAAK,EAAE,KAAK,CAAC,WAAW,IAAI,CAAC,CAAC,EACnE,CAAC,CACF,EACH,CAAC,CACF,CAAC;AACJ,CAAC;AAED,SAAS,GAAG,CAAC,KAAa,EAAE,KAAa;IACvC,OAAO,KAAK,CAAC,MAAM,IAAI,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,GAAG,GAAG,CAAC,MAAM,CAAC,KAAK,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC;AAClF,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,WAAW,CAAC,OAA8B;IACxD,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACzB,OAAO,iBAAiB,CAAC;IAC3B,CAAC;IAED,MAAM,IAAI,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;QACpC,IAAI,EAAE,MAAM,CAAC,IAAI;QACjB,IAAI,EAAE,GAAG,MAAM,CAAC,MAAM,IAAI,MAAM,CAAC,KAAK,EAAE;QACxC,KAAK,EAAE,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC;QAC3E,MAAM,EAAE,YAAY,CAAC,MAAM,CAAC;QAC5B,KAAK,EAAE;YACL,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;SACpE,CAAC,IAAI,CAAC,GAAG,CAAC;QACX,IAAI,EAAE,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;KAC7F,CAAC,CAAC,CAAC;IAEJ,MAAM,MAAM,GAAG;QACb,IAAI,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACxD,IAAI,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACxD,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;QAC1D,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;QAC5D,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;KAC3D,CAAC;IAEF,MAAM,KAAK,GAAG;QACZ,GAAG,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,SAAS,EAAE,MAAM,CAAC,MAAM,CAAC,KAAK,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO;KAChK,CAAC;IACF,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;QACvB,KAAK,CAAC,IAAI,CACR,GAAG,GAAG,CAAC,GAAG,CAAC,IAAI,EAAE,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,GAAG,CAAC,IAAI,EAAE,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,GAAG,CAAC,KAAK,EAAE,MAAM,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,CAAC,KAAK,GAAG,CAAC,GAAG,CAAC,KAAK,EAAE,MAAM,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,IAAI,EAAE,CACjL,CAAC;IACJ,CAAC;IAED,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,KAAK,GAAG,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IAC3E,MAAM,KAAK,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,MAAM,EAAE,EAAE,CAAC,GAAG,GAAG,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;IACrE,MAAM,OAAO,GAAG,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,MAAM,GAAG,KAAK,CAAC,GAAG,GAAG,CAAC,CAAC;IACrE,MAAM,KAAK,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC;IACnC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACf,KAAK,CAAC,IAAI,CACR,YAAY,MAAM,IAAI,KAAK,KAAK,OAAO,KAAK,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,KAAK,YAAY,CAAC,KAAK,CAAC,SAAS,EAAE,CACnG,CAAC;IACF,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC;AACjC,CAAC;AAED,wEAAwE;AACxE,MAAM,UAAU,cAAc,CAAC,OAA8B;IAC3D,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC7B,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAoB,CAAC;QAC7C,KAAK,MAAM,OAAO,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACtC,KAAK,MAAM,OAAO,IAAI,OAAO,CAAC,QAAQ,EAAE,CAAC;gBACvC,MAAM,QAAQ,GAAG,QAAQ,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC;gBAC7C,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;gBAC/B,QAAQ,CAAC,GAAG,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAC;YAClC,CAAC;QACH,CAAC;QACD,KAAK,MAAM,CAAC,MAAM,EAAE,QAAQ,CAAC,IAAI,QAAQ,EAAE,CAAC;YAC1C,KAAK,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,IAAI,YAAY,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,MAAM,EAAE,CAAC,CAAC;QACxE,CAAC;IACH,CAAC;IACD,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC;AAC3D,CAAC;AAED,MAAM,UAAU,YAAY,CAC1B,OAA8B,EAC9B,IAAgB;IAEhB,OAAO;QACL,aAAa,EAAE,CAAC;QAChB,GAAG,IAAI;QACP,KAAK,EAAE,OAAO,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;YAC9B,IAAI,EAAE,MAAM,CAAC,IAAI;YACjB,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,KAAK,EAAE,MAAM,CAAC,KAAK;YACnB,QAAQ,EAAE,MAAM,CAAC,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,GAAG,MAAM,CAAC,KAAK;YAC/D,SAAS,EAAE,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;YACpE,WAAW,EAAE,MAAM,CAAC,QAAQ,CAAC,MAAM,CACjC,CAAC,GAAG,EAAE,OAAO,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,OAAO,CAAC,KAAK,EAAE,KAAK,CAAC,WAAW,IAAI,CAAC,CAAC,EAC/D,CAAC,CACF;YACD,gBAAgB,EAAE,MAAM,CACtB,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,UAAU,CAAC,CACrD;YACD,QAAQ,EAAE,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;gBAC1C,OAAO,EAAE,OAAO,CAAC,OAAO;gBACxB,MAAM,EAAE,OAAO,CAAC,MAAM;gBACtB,MAAM,EAAE,OAAO,CAAC,MAAM;gBACtB,SAAS,EAAE,OAAO,CAAC,SAAS;gBAC5B,UAAU,EAAE,OAAO,CAAC,UAAU;gBAC9B,SAAS,EAAE,OAAO,CAAC,SAAS;gBAC5B,SAAS,EAAE,OAAO,CAAC,SAAS;gBAC5B,GAAG,CAAC,OAAO,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,CAAC;gBAChE,QAAQ,EAAE,OAAO,CAAC,QAAQ;aAC3B,CAAC,CAAC;SACJ,CAAC,CAAC;QACH,OAAO,EAAE;YACP,MAAM,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,KAAK,GAAG,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;YACnE,KAAK,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,MAAM,EAAE,EAAE,CAAC,GAAG,GAAG,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;YAC7D,WAAW,EAAE,WAAW,CAAC,OAAO,CAAC;SAClC;KACF,CAAC;AACJ,CAAC"}
|
package/dist/evals/runner.d.ts
CHANGED
|
@@ -2,6 +2,22 @@ import type { LLMClient } from "../llm.js";
|
|
|
2
2
|
import type { UsageTotal } from "../providers/usage.js";
|
|
3
3
|
import { type AttemptFacts } from "./graders.js";
|
|
4
4
|
import type { Capability, Task } from "./task.js";
|
|
5
|
+
/** One tool result, in call order. */
|
|
6
|
+
export interface ToolCallRecord {
|
|
7
|
+
readonly name: string;
|
|
8
|
+
readonly ok: boolean;
|
|
9
|
+
/**
|
|
10
|
+
* What the call was aimed at: the `path` argument, or a truncated `command`.
|
|
11
|
+
*
|
|
12
|
+
* Full arguments are not kept — an edit carries both halves of the change
|
|
13
|
+
* and would bury the report in file contents. But a name and a success flag
|
|
14
|
+
* alone cannot tell a model that read the four files on an import chain from
|
|
15
|
+
* one that read four unrelated files, and on a task built out of 41 files
|
|
16
|
+
* that distinction is the entire measurement. Omitted when the tool takes
|
|
17
|
+
* neither.
|
|
18
|
+
*/
|
|
19
|
+
readonly target?: string;
|
|
20
|
+
}
|
|
5
21
|
export interface AttemptResult {
|
|
6
22
|
readonly attempt: number;
|
|
7
23
|
readonly passed: boolean;
|
|
@@ -9,6 +25,14 @@ export interface AttemptResult {
|
|
|
9
25
|
readonly turnCount: number;
|
|
10
26
|
readonly durationMs: number;
|
|
11
27
|
readonly toolsUsed: readonly string[];
|
|
28
|
+
/**
|
|
29
|
+
* Every call in order, with whether it succeeded.
|
|
30
|
+
*
|
|
31
|
+
* `toolsUsed` is a deduplicated set, so a model that guessed a path, got an
|
|
32
|
+
* error and recovered looks identical there to one that never guessed. This
|
|
33
|
+
* is what makes that question answerable from the report.
|
|
34
|
+
*/
|
|
35
|
+
readonly toolCalls: readonly ToolCallRecord[];
|
|
12
36
|
/** What the attempt cost, when the Provider reported it. */
|
|
13
37
|
readonly usage?: UsageTotal;
|
|
14
38
|
/** One entry per failed grader, in task order. */
|
|
@@ -21,7 +45,12 @@ export interface TaskResult {
|
|
|
21
45
|
readonly total: number;
|
|
22
46
|
}
|
|
23
47
|
export interface RunnerOptions {
|
|
24
|
-
/**
|
|
48
|
+
/**
|
|
49
|
+
* Called once per attempt, so a test can hand out a fresh fake each time.
|
|
50
|
+
* The CLI returns the same client every time on purpose: it holds only
|
|
51
|
+
* readonly config and takes the history per request, so attempts cannot
|
|
52
|
+
* leak into one another through it.
|
|
53
|
+
*/
|
|
25
54
|
readonly createClient: () => LLMClient;
|
|
26
55
|
readonly systemPrompt: string;
|
|
27
56
|
readonly capabilities: readonly Capability[];
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../src/evals/runner.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,WAAW,CAAC;AAC3C,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,uBAAuB,CAAC;
|
|
1
|
+
{"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../src/evals/runner.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,WAAW,CAAC;AAC3C,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,uBAAuB,CAAC;AAKxD,OAAO,EAAgC,KAAK,YAAY,EAAE,MAAM,cAAc,CAAC;AAC/E,OAAO,KAAK,EAAE,UAAU,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAElD,sCAAsC;AACtC,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,EAAE,EAAE,OAAO,CAAC;IACrB;;;;;;;;;OASG;IACH,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM,CAAC;CAC1B;AAED,MAAM,WAAW,aAAa;IAC5B,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,MAAM,EAAE,OAAO,CAAC;IACzB,QAAQ,CAAC,MAAM,EAAE,YAAY,CAAC,QAAQ,CAAC,CAAC;IACxC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,SAAS,EAAE,SAAS,MAAM,EAAE,CAAC;IACtC;;;;;;OAMG;IACH,QAAQ,CAAC,SAAS,EAAE,SAAS,cAAc,EAAE,CAAC;IAC9C,4DAA4D;IAC5D,QAAQ,CAAC,KAAK,CAAC,EAAE,UAAU,CAAC;IAC5B,kDAAkD;IAClD,QAAQ,CAAC,QAAQ,EAAE,SAAS,MAAM,EAAE,CAAC;CACtC;AAED,MAAM,WAAW,UAAU;IACzB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,QAAQ,EAAE,SAAS,aAAa,EAAE,CAAC;IAC5C,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,aAAa;IAC5B;;;;;OAKG;IACH,QAAQ,CAAC,YAAY,EAAE,MAAM,SAAS,CAAC;IACvC,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,YAAY,EAAE,SAAS,UAAU,EAAE,CAAC;IAC7C,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC,MAAM,EAAE,aAAa,EAAE,IAAI,EAAE,IAAI,KAAK,IAAI,CAAC;IACjE,QAAQ,CAAC,MAAM,CAAC,EAAE,WAAW,CAAC;CAC/B;AAED,qBAAa,sBAAuB,SAAQ,KAAK;IACnC,QAAQ,CAAC,IAAI,EAAE,MAAM;IAAE,QAAQ,CAAC,OAAO,EAAE,SAAS,UAAU,EAAE;IAA1E,YAAqB,IAAI,EAAE,MAAM,EAAW,OAAO,EAAE,SAAS,UAAU,EAAE,EAOzE;CACF;AAwHD;;;;;;GAMG;AACH,wBAAsB,OAAO,CAC3B,IAAI,EAAE,IAAI,EACV,OAAO,EAAE,aAAa,GACrB,OAAO,CAAC,UAAU,CAAC,CAoBrB;AAED,wBAAsB,QAAQ,CAC5B,KAAK,EAAE,SAAS,IAAI,EAAE,EACtB,OAAO,EAAE,aAAa,GACrB,OAAO,CAAC,SAAS,UAAU,EAAE,CAAC,CAMhC"}
|
package/dist/evals/runner.js
CHANGED
|
@@ -1,11 +1,7 @@
|
|
|
1
1
|
import { Agent as AgentClass } from "../agent.js";
|
|
2
2
|
import { LocalWorkspace } from "../workspace.js";
|
|
3
|
-
import {
|
|
4
|
-
import {
|
|
5
|
-
import { SearchTextTool } from "../tools/search-text.js";
|
|
6
|
-
import { WriteFileTool } from "../tools/write-file.js";
|
|
7
|
-
import { EditFileTool } from "../tools/edit-file.js";
|
|
8
|
-
import { BashTool } from "../tools/bash.js";
|
|
3
|
+
import { callTarget } from "../tools/target.js";
|
|
4
|
+
import { toolsFor } from "./tools.js";
|
|
9
5
|
import { createAttemptWorkspace } from "./fixture.js";
|
|
10
6
|
import { createGrader, describeGrader } from "./graders.js";
|
|
11
7
|
export class MissingCapabilityError extends Error {
|
|
@@ -20,20 +16,6 @@ export class MissingCapabilityError extends Error {
|
|
|
20
16
|
this.name = "MissingCapabilityError";
|
|
21
17
|
}
|
|
22
18
|
}
|
|
23
|
-
function toolsFor(capabilities, cwd) {
|
|
24
|
-
const tools = [
|
|
25
|
-
new ListFilesTool(),
|
|
26
|
-
new SearchTextTool(),
|
|
27
|
-
new ReadFileTool(),
|
|
28
|
-
];
|
|
29
|
-
if (capabilities.includes("writes")) {
|
|
30
|
-
tools.push(new WriteFileTool(), new EditFileTool());
|
|
31
|
-
}
|
|
32
|
-
if (capabilities.includes("shell")) {
|
|
33
|
-
tools.push(new BashTool({ cwd }));
|
|
34
|
-
}
|
|
35
|
-
return tools;
|
|
36
|
-
}
|
|
37
19
|
function missingCapabilities(task, granted) {
|
|
38
20
|
return task.capabilities.filter((capability) => !granted.includes(capability));
|
|
39
21
|
}
|
|
@@ -75,6 +57,7 @@ async function runAttempt(task, attempt, options) {
|
|
|
75
57
|
}
|
|
76
58
|
const facts = {
|
|
77
59
|
workspace: workspace.path,
|
|
60
|
+
fixtureDirectory: task.fixtureDirectory,
|
|
78
61
|
finalAnswer,
|
|
79
62
|
turnCount,
|
|
80
63
|
status,
|
|
@@ -92,6 +75,28 @@ async function runAttempt(task, attempt, options) {
|
|
|
92
75
|
failures.push(`${describeGrader(spec)} — ${result.reason}`);
|
|
93
76
|
}
|
|
94
77
|
}
|
|
78
|
+
// The arguments live on the start event, the outcome on the end event;
|
|
79
|
+
// the call id is what joins them.
|
|
80
|
+
const targets = new Map();
|
|
81
|
+
for (const event of events) {
|
|
82
|
+
if (event.type === "tool_execution_start") {
|
|
83
|
+
const target = callTarget(event.arguments);
|
|
84
|
+
if (target !== undefined) {
|
|
85
|
+
targets.set(event.toolCallId, target);
|
|
86
|
+
}
|
|
87
|
+
}
|
|
88
|
+
}
|
|
89
|
+
const calls = events
|
|
90
|
+
.filter((event) => event.type === "tool_execution_end")
|
|
91
|
+
.map((event) => {
|
|
92
|
+
const end = event;
|
|
93
|
+
const target = targets.get(end.toolCallId);
|
|
94
|
+
return {
|
|
95
|
+
name: end.toolName,
|
|
96
|
+
ok: end.isError === false,
|
|
97
|
+
...(target === undefined ? {} : { target }),
|
|
98
|
+
};
|
|
99
|
+
});
|
|
95
100
|
return {
|
|
96
101
|
attempt,
|
|
97
102
|
passed: failures.length === 0,
|
|
@@ -99,11 +104,8 @@ async function runAttempt(task, attempt, options) {
|
|
|
99
104
|
turnCount,
|
|
100
105
|
durationMs: Date.now() - startedAt,
|
|
101
106
|
...(usage === undefined ? {} : { usage }),
|
|
102
|
-
toolsUsed: [
|
|
103
|
-
|
|
104
|
-
.filter((event) => event.type === "tool_execution_end")
|
|
105
|
-
.map((event) => event.toolName)),
|
|
106
|
-
],
|
|
107
|
+
toolsUsed: [...new Set(calls.map((call) => call.name))],
|
|
108
|
+
toolCalls: calls,
|
|
107
109
|
failures,
|
|
108
110
|
};
|
|
109
111
|
}
|
package/dist/evals/runner.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"runner.js","sourceRoot":"","sources":["../../src/evals/runner.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,KAAK,IAAI,UAAU,EAAE,MAAM,aAAa,CAAC;
|
|
1
|
+
{"version":3,"file":"runner.js","sourceRoot":"","sources":["../../src/evals/runner.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,KAAK,IAAI,UAAU,EAAE,MAAM,aAAa,CAAC;AAIlD,OAAO,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AACjD,OAAO,EAAE,UAAU,EAAE,MAAM,oBAAoB,CAAC;AAChD,OAAO,EAAE,QAAQ,EAAE,MAAM,YAAY,CAAC;AACtC,OAAO,EAAE,sBAAsB,EAAE,MAAM,cAAc,CAAC;AACtD,OAAO,EAAE,YAAY,EAAE,cAAc,EAAqB,MAAM,cAAc,CAAC;AA+D/E,MAAM,OAAO,sBAAuB,SAAQ,KAAK;IAC1B,IAAI;IAAmB,OAAO;IAAnD,YAAqB,IAAY,EAAW,OAA8B;QACxE,KAAK,CACH,SAAS,IAAI,WAAW,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,gBAAgB,OAAO;aACjE,GAAG,CAAC,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC,UAAU,KAAK,QAAQ,CAAC,CAAC,CAAC,gBAAgB,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC;aACnF,IAAI,CAAC,GAAG,CAAC,GAAG,CAChB,CAAC;oBALiB,IAAI;uBAAmB,OAAO;QAMjD,IAAI,CAAC,IAAI,GAAG,wBAAwB,CAAC;IACvC,CAAC;CACF;AAED,SAAS,mBAAmB,CAC1B,IAAU,EACV,OAA8B;IAE9B,OAAO,IAAI,CAAC,YAAY,CAAC,MAAM,CAC7B,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAC,CAC9C,CAAC;AACJ,CAAC;AAED,KAAK,UAAU,UAAU,CACvB,IAAU,EACV,OAAe,EACf,OAAsB;IAEtB,MAAM,SAAS,GAAG,MAAM,sBAAsB,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC;IACtE,MAAM,MAAM,GAAiB,EAAE,CAAC;IAChC,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IAE7B,IAAI,CAAC;QACH,MAAM,KAAK,GAAU,IAAI,UAAU,CAAC;YAClC,YAAY,EAAE,OAAO,CAAC,YAAY;YAClC,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,GAAG,EAAE,OAAO,CAAC,YAAY,EAAE;YAC3B,KAAK,EAAE,QAAQ,CAAC,IAAI,CAAC,YAAY,EAAE,SAAS,CAAC,IAAI,CAAC;YAClD,SAAS,EAAE,IAAI,cAAc,CAAC,SAAS,CAAC,IAAI,EAAE;gBAC5C,WAAW,EAAE,IAAI,CAAC,YAAY,CAAC,QAAQ,CAAC,QAAQ,CAAC;aAClD,CAAC;YACF,SAAS,EAAE,KAAK;YAChB,OAAO,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC;SACvC,CAAC,CAAC;QAEH,IAAI,MAAM,GAA2B,OAAO,CAAC;QAC7C,IAAI,WAAW,GAAG,EAAE,CAAC;QACrB,IAAI,SAAS,GAAG,CAAC,CAAC;QAClB,IAAI,KAA6B,CAAC;QAClC,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,EAAE;gBAC1C,GAAG,CAAC,OAAO,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC;aACpE,CAAC,CAAC;YACH,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC;YACvB,SAAS,GAAG,MAAM,CAAC,SAAS,CAAC;YAC7B,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC;YACrB,WAAW;gBACT,MAAM,CAAC,MAAM,KAAK,WAAW,CAAC,CAAC,CAAC,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC;QACrE,CAAC;QAAC,OAAO,KAAc,EAAE,CAAC;YACxB,yEAAyE;YACzE,mEAAmE;YACnE,MAAM,GAAG,OAAO,CAAC;YACjB,WAAW,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QACvE,CAAC;QAED,MAAM,KAAK,GAAiB;YAC1B,SAAS,EAAE,SAAS,CAAC,IAAI;YACzB,gBAAgB,EAAE,IAAI,CAAC,gBAAgB;YACvC,WAAW;YACX,SAAS;YACT,MAAM;YACN,MAAM;SACP,CAAC;QAEF,MAAM,QAAQ,GAAa,EAAE,CAAC;QAC9B,IAAI,MAAM,KAAK,WAAW,EAAE,CAAC;YAC3B,QAAQ,CAAC,IAAI,CACX,MAAM,KAAK,WAAW;gBACpB,CAAC,CAAC,WAAW,IAAI,CAAC,QAAQ,oCAAoC;gBAC9D,CAAC,CAAC,gBAAgB,MAAM,KAAK,WAAW,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,EAAE,CACrE,CAAC;QACJ,CAAC;QACD,KAAK,MAAM,IAAI,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;YAChC,MAAM,MAAM,GAAG,MAAM,YAAY,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC;YAC/C,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC;gBACnB,QAAQ,CAAC,IAAI,CAAC,GAAG,cAAc,CAAC,IAAI,CAAC,MAAM,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;YAC9D,CAAC;QACH,CAAC;QAED,uEAAuE;QACvE,kCAAkC;QAClC,MAAM,OAAO,GAAG,IAAI,GAAG,EAAkB,CAAC;QAC1C,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;YAC3B,IAAI,KAAK,CAAC,IAAI,KAAK,sBAAsB,EAAE,CAAC;gBAC1C,MAAM,MAAM,GAAG,UAAU,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;gBAC3C,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;oBACzB,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,UAAU,EAAE,MAAM,CAAC,CAAC;gBACxC,CAAC;YACH,CAAC;QACH,CAAC;QACD,MAAM,KAAK,GAA8B,MAAM;aAC5C,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,KAAK,oBAAoB,CAAC;aACtD,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE;YACb,MAAM,GAAG,GAAG,KAIX,CAAC;YACF,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC;YAC3C,OAAO;gBACL,IAAI,EAAE,GAAG,CAAC,QAAQ;gBAClB,EAAE,EAAE,GAAG,CAAC,OAAO,KAAK,KAAK;gBACzB,GAAG,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC;aAC5C,CAAC;QACJ,CAAC,CAAC,CAAC;QAEL,OAAO;YACL,OAAO;YACP,MAAM,EAAE,QAAQ,CAAC,MAAM,KAAK,CAAC;YAC7B,MAAM;YACN,SAAS;YACT,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;YAClC,GAAG,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC;YACzC,SAAS,EAAE,CAAC,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;YACvD,SAAS,EAAE,KAAK;YAChB,QAAQ;SACT,CAAC;IACJ,CAAC;YAAS,CAAC;QACT,MAAM,SAAS,CAAC,OAAO,EAAE,CAAC;IAC5B,CAAC;AACH,CAAC;AAED;;;;;;GAMG;AACH,MAAM,CAAC,KAAK,UAAU,OAAO,CAC3B,IAAU,EACV,OAAsB;IAEtB,MAAM,OAAO,GAAG,mBAAmB,CAAC,IAAI,EAAE,OAAO,CAAC,YAAY,CAAC,CAAC;IAChE,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACvB,MAAM,IAAI,sBAAsB,CAAC,IAAI,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;IACvD,CAAC;IAED,MAAM,KAAK,GAAG,OAAO,CAAC,QAAQ,IAAI,IAAI,CAAC,QAAQ,CAAC;IAChD,MAAM,QAAQ,GAAoB,EAAE,CAAC;IACrC,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,IAAI,KAAK,EAAE,KAAK,IAAI,CAAC,EAAE,CAAC;QAC/C,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,IAAI,EAAE,KAAK,EAAE,OAAO,CAAC,CAAC;QACtD,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACtB,OAAO,CAAC,SAAS,EAAE,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC;IACpC,CAAC;IAED,OAAO;QACL,IAAI,EAAE,IAAI,CAAC,IAAI;QACf,QAAQ;QACR,MAAM,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,MAAM;QAC3D,KAAK;KACN,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,QAAQ,CAC5B,KAAsB,EACtB,OAAsB;IAEtB,MAAM,OAAO,GAAiB,EAAE,CAAC;IACjC,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,OAAO,CAAC,IAAI,CAAC,MAAM,OAAO,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,CAAC;IAC7C,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC"}
|
package/dist/evals/task.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"task.d.ts","sourceRoot":"","sources":["../../src/evals/task.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"task.d.ts","sourceRoot":"","sources":["../../src/evals/task.ts"],"names":[],"mappings":"AAIA,qBAAa,SAAU,SAAQ,KAAK;IAClC,YAAY,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,OAAO,CAAA;KAAE,EAMzD;CACF;AAED,yEAAyE;AACzE,MAAM,MAAM,UAAU,GAAG,QAAQ,GAAG,OAAO,CAAC;AAE5C,MAAM,WAAW,UAAU;IACzB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,EAAE,GAAG,EAAE,MAAM,GAAG,OAAO,CAAC;CACjC;AAED,MAAM,WAAW,IAAI;IACnB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,YAAY,EAAE,SAAS,UAAU,EAAE,CAAC;IAC7C,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,OAAO,EAAE,SAAS,UAAU,EAAE,CAAC;IACxC,iEAAiE;IACjE,QAAQ,CAAC,gBAAgB,EAAE,MAAM,CAAC;IAClC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;CAC5B;AAoFD,wBAAgB,SAAS,CACvB,KAAK,EAAE,OAAO,EACd,SAAS,EAAE,MAAM,EACjB,gBAAgB,EAAE,MAAM,GACvB,IAAI,CA8CN;AAED,wBAAsB,QAAQ,CAAC,SAAS,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAgB/D;AAED,6DAA6D;AAC7D,wBAAsB,SAAS,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,SAAS,IAAI,EAAE,CAAC,CAwBtE"}
|
package/dist/evals/task.js
CHANGED
|
@@ -1,5 +1,6 @@
|
|
|
1
1
|
import { readFile, readdir, stat } from "node:fs/promises";
|
|
2
2
|
import path from "node:path";
|
|
3
|
+
import { toolNamesFor } from "./tools.js";
|
|
3
4
|
export class TaskError extends Error {
|
|
4
5
|
constructor(message, options) {
|
|
5
6
|
super(message);
|
|
@@ -38,6 +39,37 @@ function readPositiveInteger(value, fallback, maximum, field, name) {
|
|
|
38
39
|
}
|
|
39
40
|
return value;
|
|
40
41
|
}
|
|
42
|
+
/** Grader types whose "name" must be a tool the task actually grants. */
|
|
43
|
+
const TOOL_NAME_GRADERS = new Set([
|
|
44
|
+
"used-tool",
|
|
45
|
+
"not-used-tool",
|
|
46
|
+
"tool-succeeded",
|
|
47
|
+
"tool-never-failed",
|
|
48
|
+
]);
|
|
49
|
+
/**
|
|
50
|
+
* Refuses a grader that names a tool the task never hands to the model.
|
|
51
|
+
*
|
|
52
|
+
* `not-used-tool: write_file` on a read-only task can never fail — it asserts
|
|
53
|
+
* the model did not call something it was never given, so it reads like a
|
|
54
|
+
* check and is worth nothing. The mirror image, `used-tool` on a tool the task
|
|
55
|
+
* withholds, can never pass. Both are task-definition mistakes, and both are
|
|
56
|
+
* caught here rather than at run time so they cost nothing to find.
|
|
57
|
+
*/
|
|
58
|
+
function assertGradersCanFire(name, capabilities, graders) {
|
|
59
|
+
const available = toolNamesFor(capabilities);
|
|
60
|
+
for (const [index, grader] of graders.entries()) {
|
|
61
|
+
if (!TOOL_NAME_GRADERS.has(grader.type)) {
|
|
62
|
+
continue;
|
|
63
|
+
}
|
|
64
|
+
const tool = grader.name;
|
|
65
|
+
if (typeof tool !== "string" || available.includes(tool)) {
|
|
66
|
+
continue;
|
|
67
|
+
}
|
|
68
|
+
throw new TaskError(`${name}: graders[${index}] (${grader.type}) names "${tool}", which this task never grants. ` +
|
|
69
|
+
`Available with capabilities [${capabilities.join(", ")}]: ${available.join(", ")}. ` +
|
|
70
|
+
`Grant the capability or drop the grader.`);
|
|
71
|
+
}
|
|
72
|
+
}
|
|
41
73
|
export function parseTask(value, directory, fixtureDirectory) {
|
|
42
74
|
if (!isRecord(value)) {
|
|
43
75
|
throw new TaskError(`${directory}: task.json must contain a JSON object.`);
|
|
@@ -57,10 +89,12 @@ export function parseTask(value, directory, fixtureDirectory) {
|
|
|
57
89
|
}
|
|
58
90
|
return entry;
|
|
59
91
|
});
|
|
92
|
+
const capabilities = readCapabilities(value.capabilities, name);
|
|
93
|
+
assertGradersCanFire(name, capabilities, graders);
|
|
60
94
|
return {
|
|
61
95
|
name,
|
|
62
96
|
prompt: value.prompt,
|
|
63
|
-
capabilities
|
|
97
|
+
capabilities,
|
|
64
98
|
attempts: readPositiveInteger(value.attempts, DEFAULT_ATTEMPTS, MAX_ATTEMPTS, "attempts", name),
|
|
65
99
|
maxTurns: readPositiveInteger(value.maxTurns, DEFAULT_MAX_TURNS, 100, "maxTurns", name),
|
|
66
100
|
graders,
|
package/dist/evals/task.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"task.js","sourceRoot":"","sources":["../../src/evals/task.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,kBAAkB,CAAC;AAC3D,OAAO,IAAI,MAAM,WAAW,CAAC;
|
|
1
|
+
{"version":3,"file":"task.js","sourceRoot":"","sources":["../../src/evals/task.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,kBAAkB,CAAC;AAC3D,OAAO,IAAI,MAAM,WAAW,CAAC;AAC7B,OAAO,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAE1C,MAAM,OAAO,SAAU,SAAQ,KAAK;IAClC,YAAY,OAAe,EAAE,OAA6B;QACxD,KAAK,CAAC,OAAO,CAAC,CAAC;QACf,IAAI,CAAC,IAAI,GAAG,WAAW,CAAC;QACxB,IAAI,OAAO,EAAE,KAAK,KAAK,SAAS,EAAE,CAAC;YACjC,IAAI,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC;QAC7B,CAAC;IACH,CAAC;CACF;AAsBD,MAAM,gBAAgB,GAAG,CAAC,CAAC;AAC3B,MAAM,iBAAiB,GAAG,EAAE,CAAC;AAC7B,MAAM,YAAY,GAAG,EAAE,CAAC;AAExB,SAAS,QAAQ,CAAC,KAAc;IAC9B,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;AAC9E,CAAC;AAED,SAAS,gBAAgB,CAAC,KAAc,EAAE,IAAY;IACpD,IAAI,KAAK,KAAK,SAAS,EAAE,CAAC;QACxB,OAAO,EAAE,CAAC;IACZ,CAAC;IACD,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC;QAC1B,MAAM,IAAI,SAAS,CAAC,GAAG,IAAI,kCAAkC,CAAC,CAAC;IACjE,CAAC;IACD,OAAO,KAAK,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE;QACzB,IAAI,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,OAAO,EAAE,CAAC;YAC5C,MAAM,IAAI,SAAS,CACjB,GAAG,IAAI,wBAAwB,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,iCAAiC,CACtF,CAAC;QACJ,CAAC;QACD,OAAO,KAAK,CAAC;IACf,CAAC,CAAC,CAAC;AACL,CAAC;AAED,SAAS,mBAAmB,CAC1B,KAAc,EACd,QAAgB,EAChB,OAAe,EACf,KAAa,EACb,IAAY;IAEZ,IAAI,KAAK,KAAK,SAAS,EAAE,CAAC;QACxB,OAAO,QAAQ,CAAC;IAClB,CAAC;IACD,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,IAAK,KAAgB,GAAG,CAAC,IAAK,KAAgB,GAAG,OAAO,EAAE,CAAC;QACzF,MAAM,IAAI,SAAS,CACjB,GAAG,IAAI,KAAK,KAAK,iCAAiC,OAAO,GAAG,CAC7D,CAAC;IACJ,CAAC;IACD,OAAO,KAAe,CAAC;AACzB,CAAC;AAED,yEAAyE;AACzE,MAAM,iBAAiB,GAAG,IAAI,GAAG,CAAC;IAChC,WAAW;IACX,eAAe;IACf,gBAAgB;IAChB,mBAAmB;CACpB,CAAC,CAAC;AAEH;;;;;;;;GAQG;AACH,SAAS,oBAAoB,CAC3B,IAAY,EACZ,YAAmC,EACnC,OAA8B;IAE9B,MAAM,SAAS,GAAG,YAAY,CAAC,YAAY,CAAC,CAAC;IAC7C,KAAK,MAAM,CAAC,KAAK,EAAE,MAAM,CAAC,IAAI,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC;QAChD,IAAI,CAAC,iBAAiB,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC;YACxC,SAAS;QACX,CAAC;QACD,MAAM,IAAI,GAAG,MAAM,CAAC,IAAI,CAAC;QACzB,IAAI,OAAO,IAAI,KAAK,QAAQ,IAAI,SAAS,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,CAAC;YACzD,SAAS;QACX,CAAC;QACD,MAAM,IAAI,SAAS,CACjB,GAAG,IAAI,aAAa,KAAK,MAAM,MAAM,CAAC,IAAI,YAAY,IAAI,mCAAmC;YAC3F,gCAAgC,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI;YACrF,0CAA0C,CAC7C,CAAC;IACJ,CAAC;AACH,CAAC;AAED,MAAM,UAAU,SAAS,CACvB,KAAc,EACd,SAAiB,EACjB,gBAAwB;IAExB,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;QACrB,MAAM,IAAI,SAAS,CAAC,GAAG,SAAS,yCAAyC,CAAC,CAAC;IAC7E,CAAC;IACD,MAAM,IAAI,GACR,OAAO,KAAK,CAAC,IAAI,KAAK,QAAQ,IAAI,KAAK,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC;QACrD,CAAC,CAAC,KAAK,CAAC,IAAI;QACZ,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;IAC/B,IAAI,OAAO,KAAK,CAAC,MAAM,KAAK,QAAQ,IAAI,KAAK,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACzE,MAAM,IAAI,SAAS,CAAC,GAAG,IAAI,sCAAsC,CAAC,CAAC;IACrE,CAAC;IACD,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAChE,MAAM,IAAI,SAAS,CAAC,GAAG,IAAI,oCAAoC,CAAC,CAAC;IACnE,CAAC;IACD,MAAM,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE;QACjD,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,OAAO,KAAK,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;YACvD,MAAM,IAAI,SAAS,CAAC,GAAG,IAAI,aAAa,KAAK,iBAAiB,CAAC,CAAC;QAClE,CAAC;QACD,OAAO,KAAmB,CAAC;IAC7B,CAAC,CAAC,CAAC;IAEH,MAAM,YAAY,GAAG,gBAAgB,CAAC,KAAK,CAAC,YAAY,EAAE,IAAI,CAAC,CAAC;IAChE,oBAAoB,CAAC,IAAI,EAAE,YAAY,EAAE,OAAO,CAAC,CAAC;IAElD,OAAO;QACL,IAAI;QACJ,MAAM,EAAE,KAAK,CAAC,MAAM;QACpB,YAAY;QACZ,QAAQ,EAAE,mBAAmB,CAC3B,KAAK,CAAC,QAAQ,EACd,gBAAgB,EAChB,YAAY,EACZ,UAAU,EACV,IAAI,CACL;QACD,QAAQ,EAAE,mBAAmB,CAC3B,KAAK,CAAC,QAAQ,EACd,iBAAiB,EACjB,GAAG,EACH,UAAU,EACV,IAAI,CACL;QACD,OAAO;QACP,gBAAgB;QAChB,SAAS;KACV,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,QAAQ,CAAC,SAAiB;IAC9C,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,WAAW,CAAC,CAAC;IAC/C,IAAI,QAAgB,CAAC;IACrB,IAAI,CAAC;QACH,QAAQ,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;IAC1C,CAAC;IAAC,OAAO,KAAc,EAAE,CAAC;QACxB,MAAM,IAAI,SAAS,CAAC,kBAAkB,IAAI,GAAG,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,CAAC;IACnE,CAAC;IAED,IAAI,MAAe,CAAC;IACpB,IAAI,CAAC;QACH,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;IAChC,CAAC;IAAC,OAAO,KAAc,EAAE,CAAC;QACxB,MAAM,IAAI,SAAS,CAAC,GAAG,IAAI,qBAAqB,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,CAAC;IACtE,CAAC;IACD,OAAO,SAAS,CAAC,MAAM,EAAE,SAAS,EAAE,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,SAAS,CAAC,CAAC,CAAC;AACvE,CAAC;AAED,6DAA6D;AAC7D,MAAM,CAAC,KAAK,UAAU,SAAS,CAAC,IAAY;IAC1C,IAAI,OAAO,CAAC;IACZ,IAAI,CAAC;QACH,OAAO,GAAG,MAAM,OAAO,CAAC,IAAI,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;IACzD,CAAC;IAAC,OAAO,KAAc,EAAE,CAAC;QACxB,MAAM,IAAI,SAAS,CAAC,qCAAqC,IAAI,GAAG,EAAE;YAChE,KAAK,EAAE,KAAK;SACb,CAAC,CAAC;IACL,CAAC;IAED,MAAM,KAAK,GAAW,EAAE,CAAC;IACzB,KAAK,MAAM,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC;QACzE,IAAI,CAAC,KAAK,CAAC,WAAW,EAAE,EAAE,CAAC;YACzB,SAAS;QACX,CAAC;QACD,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;QAC9C,IAAI,CAAC;YACH,MAAM,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,WAAW,CAAC,CAAC,CAAC;QAChD,CAAC;QAAC,MAAM,CAAC;YACP,SAAS;QACX,CAAC;QACD,KAAK,CAAC,IAAI,CAAC,MAAM,QAAQ,CAAC,SAAS,CAAC,CAAC,CAAC;IACxC,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC"}
|
|
@@ -0,0 +1,5 @@
|
|
|
1
|
+
import type { Tool } from "../tools/tool.js";
|
|
2
|
+
import type { Capability } from "./task.js";
|
|
3
|
+
export declare function toolNamesFor(capabilities: readonly Capability[]): readonly string[];
|
|
4
|
+
export declare function toolsFor(capabilities: readonly Capability[], cwd: string): readonly Tool[];
|
|
5
|
+
//# sourceMappingURL=tools.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"tools.d.ts","sourceRoot":"","sources":["../../src/evals/tools.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,kBAAkB,CAAC;AAO7C,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,WAAW,CAAC;AAgB5C,wBAAgB,YAAY,CAC1B,YAAY,EAAE,SAAS,UAAU,EAAE,GAClC,SAAS,MAAM,EAAE,CAKnB;AAED,wBAAgB,QAAQ,CACtB,YAAY,EAAE,SAAS,UAAU,EAAE,EACnC,GAAG,EAAE,MAAM,GACV,SAAS,IAAI,EAAE,CAajB"}
|
|
@@ -0,0 +1,40 @@
|
|
|
1
|
+
import { ListFilesTool } from "../tools/list-files.js";
|
|
2
|
+
import { ReadFileTool } from "../tools/read-file.js";
|
|
3
|
+
import { SearchTextTool } from "../tools/search-text.js";
|
|
4
|
+
import { WriteFileTool } from "../tools/write-file.js";
|
|
5
|
+
import { EditFileTool } from "../tools/edit-file.js";
|
|
6
|
+
import { BashTool } from "../tools/bash.js";
|
|
7
|
+
/**
|
|
8
|
+
* What each capability puts in the model's hands.
|
|
9
|
+
*
|
|
10
|
+
* This lives apart from the runner because task loading needs the same answer:
|
|
11
|
+
* a grader that names a tool the task never grants is either an assertion that
|
|
12
|
+
* cannot fail or one that cannot pass, and both are worth refusing before the
|
|
13
|
+
* run starts. Keeping one list means the two cannot drift.
|
|
14
|
+
*/
|
|
15
|
+
const ALWAYS = ["list_files", "search_text", "read_file"];
|
|
16
|
+
const BY_CAPABILITY = {
|
|
17
|
+
writes: ["write_file", "edit_file"],
|
|
18
|
+
shell: ["bash"],
|
|
19
|
+
};
|
|
20
|
+
export function toolNamesFor(capabilities) {
|
|
21
|
+
return [
|
|
22
|
+
...ALWAYS,
|
|
23
|
+
...capabilities.flatMap((capability) => BY_CAPABILITY[capability]),
|
|
24
|
+
];
|
|
25
|
+
}
|
|
26
|
+
export function toolsFor(capabilities, cwd) {
|
|
27
|
+
const tools = [
|
|
28
|
+
new ListFilesTool(),
|
|
29
|
+
new SearchTextTool(),
|
|
30
|
+
new ReadFileTool(),
|
|
31
|
+
];
|
|
32
|
+
if (capabilities.includes("writes")) {
|
|
33
|
+
tools.push(new WriteFileTool(), new EditFileTool());
|
|
34
|
+
}
|
|
35
|
+
if (capabilities.includes("shell")) {
|
|
36
|
+
tools.push(new BashTool({ cwd }));
|
|
37
|
+
}
|
|
38
|
+
return tools;
|
|
39
|
+
}
|
|
40
|
+
//# sourceMappingURL=tools.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"tools.js","sourceRoot":"","sources":["../../src/evals/tools.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,aAAa,EAAE,MAAM,wBAAwB,CAAC;AACvD,OAAO,EAAE,YAAY,EAAE,MAAM,uBAAuB,CAAC;AACrD,OAAO,EAAE,cAAc,EAAE,MAAM,yBAAyB,CAAC;AACzD,OAAO,EAAE,aAAa,EAAE,MAAM,wBAAwB,CAAC;AACvD,OAAO,EAAE,YAAY,EAAE,MAAM,uBAAuB,CAAC;AACrD,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAG5C;;;;;;;GAOG;AACH,MAAM,MAAM,GAAG,CAAC,YAAY,EAAE,aAAa,EAAE,WAAW,CAAU,CAAC;AACnE,MAAM,aAAa,GAA0C;IAC3D,MAAM,EAAE,CAAC,YAAY,EAAE,WAAW,CAAC;IACnC,KAAK,EAAE,CAAC,MAAM,CAAC;CAChB,CAAC;AAEF,MAAM,UAAU,YAAY,CAC1B,YAAmC;IAEnC,OAAO;QACL,GAAG,MAAM;QACT,GAAG,YAAY,CAAC,OAAO,CAAC,CAAC,UAAU,EAAE,EAAE,CAAC,aAAa,CAAC,UAAU,CAAC,CAAC;KACnE,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,QAAQ,CACtB,YAAmC,EACnC,GAAW;IAEX,MAAM,KAAK,GAAW;QACpB,IAAI,aAAa,EAAE;QACnB,IAAI,cAAc,EAAE;QACpB,IAAI,YAAY,EAAE;KACnB,CAAC;IACF,IAAI,YAAY,CAAC,QAAQ,CAAC,QAAQ,CAAC,EAAE,CAAC;QACpC,KAAK,CAAC,IAAI,CAAC,IAAI,aAAa,EAAE,EAAE,IAAI,YAAY,EAAE,CAAC,CAAC;IACtD,CAAC;IACD,IAAI,YAAY,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QACnC,KAAK,CAAC,IAAI,CAAC,IAAI,QAAQ,CAAC,EAAE,GAAG,EAAE,CAAC,CAAC,CAAC;IACpC,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"usage.d.ts","sourceRoot":"","sources":["../../src/providers/usage.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,KAAK,EAAE,MAAM,WAAW,CAAC;AAQvC;;;;;GAKG;AACH,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,OAAO,GAAG,KAAK,GAAG,SAAS,
|
|
1
|
+
{"version":3,"file":"usage.d.ts","sourceRoot":"","sources":["../../src/providers/usage.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,KAAK,EAAE,MAAM,WAAW,CAAC;AAQvC;;;;;GAKG;AACH,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,OAAO,GAAG,KAAK,GAAG,SAAS,CAiC1E;AAED,gFAAgF;AAChF,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,OAAO,GAAG,KAAK,GAAG,SAAS,CA4BpE;AAED,MAAM,WAAW,UAAU;IACzB,QAAQ,CAAC,KAAK,EAAE,KAAK,CAAC;IACtB;;;OAGG;IACH,QAAQ,CAAC,QAAQ,EAAE,OAAO,CAAC;CAC5B;AAED,wBAAgB,QAAQ,CACtB,KAAK,EAAE,UAAU,GAAG,SAAS,EAC7B,IAAI,EAAE,KAAK,GAAG,SAAS,GACtB,UAAU,CAuBZ;AAED,0CAA0C;AAC1C,wBAAgB,YAAY,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,CAQnD"}
|
package/dist/providers/usage.js
CHANGED
|
@@ -23,7 +23,9 @@ export function fromChatCompletionsUsage(value) {
|
|
|
23
23
|
const promptDetails = record.prompt_tokens_details;
|
|
24
24
|
const reasoningTokens = positive(details?.reasoning_tokens);
|
|
25
25
|
const cachedInputTokens = positive(promptDetails?.cached_tokens) ??
|
|
26
|
-
// DeepSeek reports cache hits at the top level
|
|
26
|
+
// DeepSeek also reports cache hits at the top level. It populates both,
|
|
27
|
+
// so this branch is a fallback for a Provider that reports only the flat
|
|
28
|
+
// form, not the path DeepSeek actually takes.
|
|
27
29
|
positive(record.prompt_cache_hit_tokens);
|
|
28
30
|
return {
|
|
29
31
|
inputTokens: inputTokens ?? 0,
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"usage.js","sourceRoot":"","sources":["../../src/providers/usage.ts"],"names":[],"mappings":"AAEA,SAAS,QAAQ,CAAC,KAAc;IAC9B,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,KAAK,IAAI,CAAC;QACtE,CAAC,CAAC,KAAK;QACP,CAAC,CAAC,SAAS,CAAC;AAChB,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,wBAAwB,CAAC,KAAc;IACrD,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,EAAE,CAAC;QAChD,OAAO,SAAS,CAAC;IACnB,CAAC;IACD,MAAM,MAAM,GAAG,KAAgC,CAAC;IAChD,MAAM,WAAW,GAAG,QAAQ,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC;IACnD,MAAM,YAAY,GAAG,QAAQ,CAAC,MAAM,CAAC,iBAAiB,CAAC,CAAC;IACxD,IAAI,WAAW,KAAK,SAAS,IAAI,YAAY,KAAK,SAAS,EAAE,CAAC;QAC5D,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,MAAM,OAAO,GAAG,MAAM,CAAC,yBAEV,CAAC;IACd,MAAM,aAAa,GAAG,MAAM,CAAC,qBAEhB,CAAC;IACd,MAAM,eAAe,GAAG,QAAQ,CAAC,OAAO,EAAE,gBAAgB,CAAC,CAAC;IAC5D,MAAM,iBAAiB,GACrB,QAAQ,CAAC,aAAa,EAAE,aAAa,CAAC;QACtC,
|
|
1
|
+
{"version":3,"file":"usage.js","sourceRoot":"","sources":["../../src/providers/usage.ts"],"names":[],"mappings":"AAEA,SAAS,QAAQ,CAAC,KAAc;IAC9B,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,KAAK,IAAI,CAAC;QACtE,CAAC,CAAC,KAAK;QACP,CAAC,CAAC,SAAS,CAAC;AAChB,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,wBAAwB,CAAC,KAAc;IACrD,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,EAAE,CAAC;QAChD,OAAO,SAAS,CAAC;IACnB,CAAC;IACD,MAAM,MAAM,GAAG,KAAgC,CAAC;IAChD,MAAM,WAAW,GAAG,QAAQ,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC;IACnD,MAAM,YAAY,GAAG,QAAQ,CAAC,MAAM,CAAC,iBAAiB,CAAC,CAAC;IACxD,IAAI,WAAW,KAAK,SAAS,IAAI,YAAY,KAAK,SAAS,EAAE,CAAC;QAC5D,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,MAAM,OAAO,GAAG,MAAM,CAAC,yBAEV,CAAC;IACd,MAAM,aAAa,GAAG,MAAM,CAAC,qBAEhB,CAAC;IACd,MAAM,eAAe,GAAG,QAAQ,CAAC,OAAO,EAAE,gBAAgB,CAAC,CAAC;IAC5D,MAAM,iBAAiB,GACrB,QAAQ,CAAC,aAAa,EAAE,aAAa,CAAC;QACtC,wEAAwE;QACxE,yEAAyE;QACzE,8CAA8C;QAC9C,QAAQ,CAAC,MAAM,CAAC,uBAAuB,CAAC,CAAC;IAE3C,OAAO;QACL,WAAW,EAAE,WAAW,IAAI,CAAC;QAC7B,YAAY,EAAE,YAAY,IAAI,CAAC;QAC/B,WAAW,EACT,QAAQ,CAAC,MAAM,CAAC,YAAY,CAAC,IAAI,CAAC,WAAW,IAAI,CAAC,CAAC,GAAG,CAAC,YAAY,IAAI,CAAC,CAAC;QAC3E,GAAG,CAAC,iBAAiB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,CAAC;QACjE,GAAG,CAAC,eAAe,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,CAAC;KAC9D,CAAC;AACJ,CAAC;AAED,gFAAgF;AAChF,MAAM,UAAU,kBAAkB,CAAC,KAAc;IAC/C,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,EAAE,CAAC;QAChD,OAAO,SAAS,CAAC;IACnB,CAAC;IACD,MAAM,MAAM,GAAG,KAAgC,CAAC;IAChD,MAAM,WAAW,GAAG,QAAQ,CAAC,MAAM,CAAC,YAAY,CAAC,CAAC;IAClD,MAAM,YAAY,GAAG,QAAQ,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC;IACpD,IAAI,WAAW,KAAK,SAAS,IAAI,YAAY,KAAK,SAAS,EAAE,CAAC;QAC5D,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,MAAM,aAAa,GAAG,MAAM,CAAC,qBAEhB,CAAC;IACd,MAAM,YAAY,GAAG,MAAM,CAAC,oBAEf,CAAC;IACd,MAAM,eAAe,GAAG,QAAQ,CAAC,aAAa,EAAE,gBAAgB,CAAC,CAAC;IAClE,MAAM,iBAAiB,GAAG,QAAQ,CAAC,YAAY,EAAE,aAAa,CAAC,CAAC;IAEhE,OAAO;QACL,WAAW,EAAE,WAAW,IAAI,CAAC;QAC7B,YAAY,EAAE,YAAY,IAAI,CAAC;QAC/B,WAAW,EACT,QAAQ,CAAC,MAAM,CAAC,YAAY,CAAC,IAAI,CAAC,WAAW,IAAI,CAAC,CAAC,GAAG,CAAC,YAAY,IAAI,CAAC,CAAC;QAC3E,GAAG,CAAC,iBAAiB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,CAAC;QACjE,GAAG,CAAC,eAAe,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,CAAC;KAC9D,CAAC;AACJ,CAAC;AAWD,MAAM,UAAU,QAAQ,CACtB,KAA6B,EAC7B,IAAuB;IAEvB,MAAM,IAAI,GAAG,KAAK,IAAI;QACpB,KAAK,EAAE,EAAE,WAAW,EAAE,CAAC,EAAE,YAAY,EAAE,CAAC,EAAE,WAAW,EAAE,CAAC,EAAE;QAC1D,QAAQ,EAAE,IAAI;KACf,CAAC;IACF,IAAI,IAAI,KAAK,SAAS,EAAE,CAAC;QACvB,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,QAAQ,EAAE,KAAK,EAAE,CAAC;IAChD,CAAC;IAED,MAAM,MAAM,GACV,CAAC,IAAI,CAAC,KAAK,CAAC,iBAAiB,IAAI,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,iBAAiB,IAAI,CAAC,CAAC,CAAC;IACtE,MAAM,SAAS,GACb,CAAC,IAAI,CAAC,KAAK,CAAC,eAAe,IAAI,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,eAAe,IAAI,CAAC,CAAC,CAAC;IAClE,OAAO;QACL,KAAK,EAAE;YACL,WAAW,EAAE,IAAI,CAAC,KAAK,CAAC,WAAW,GAAG,IAAI,CAAC,WAAW;YACtD,YAAY,EAAE,IAAI,CAAC,KAAK,CAAC,YAAY,GAAG,IAAI,CAAC,YAAY;YACzD,WAAW,EAAE,IAAI,CAAC,KAAK,CAAC,WAAW,GAAG,IAAI,CAAC,WAAW;YACtD,GAAG,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,MAAM,EAAE,CAAC;YACtD,GAAG,CAAC,SAAS,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,SAAS,EAAE,CAAC;SAC3D;QACD,QAAQ,EAAE,IAAI,CAAC,QAAQ;KACxB,CAAC;AACJ,CAAC;AAED,0CAA0C;AAC1C,MAAM,UAAU,YAAY,CAAC,MAAc;IACzC,IAAI,MAAM,GAAG,IAAI,EAAE,CAAC;QAClB,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC;IACxB,CAAC;IACD,IAAI,MAAM,GAAG,SAAS,EAAE,CAAC;QACvB,OAAO,GAAG,CAAC,MAAM,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;IAC1C,CAAC;IACD,OAAO,GAAG,CAAC,MAAM,GAAG,SAAS,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;AAC/C,CAAC"}
|
|
@@ -0,0 +1,10 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* The one argument worth showing beside a tool call.
|
|
3
|
+
*
|
|
4
|
+
* Full arguments are too much for a status line and for an eval report alike:
|
|
5
|
+
* an edit carries both halves of the change. But a bare tool name cannot tell
|
|
6
|
+
* a read of the right file from a read of the wrong one, so the path — or, for
|
|
7
|
+
* a shell command, its head — is kept and everything else dropped.
|
|
8
|
+
*/
|
|
9
|
+
export declare function callTarget(argumentsValue: unknown): string | undefined;
|
|
10
|
+
//# sourceMappingURL=target.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"target.d.ts","sourceRoot":"","sources":["../../src/tools/target.ts"],"names":[],"mappings":"AAEA;;;;;;;GAOG;AACH,wBAAgB,UAAU,CAAC,cAAc,EAAE,OAAO,GAAG,MAAM,GAAG,SAAS,CAiBtE"}
|