@cursor/july 0.1.16 → 0.1.17
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 +9 -4
- package/dist/bin/agent-serve.js +60 -2
- package/dist/channels/slack/api.d.ts.map +1 -1
- package/dist/channels/slack/api.js +13 -3
- package/dist/channels/slack/auth.d.ts +7 -1
- package/dist/channels/slack/auth.d.ts.map +1 -1
- package/dist/channels/slack/auth.js +14 -0
- package/dist/channels/slack/blocks.d.ts +61 -0
- package/dist/channels/slack/blocks.d.ts.map +1 -0
- package/dist/channels/slack/blocks.js +90 -0
- package/dist/channels/slack/defaults.d.ts +2 -0
- package/dist/channels/slack/defaults.d.ts.map +1 -1
- package/dist/channels/slack/defaults.js +29 -3
- package/dist/channels/slack/dispatch.d.ts +3 -17
- package/dist/channels/slack/dispatch.d.ts.map +1 -1
- package/dist/channels/slack/index.d.ts +1 -0
- package/dist/channels/slack/index.d.ts.map +1 -1
- package/dist/channels/slack/index.js +1 -0
- package/dist/channels/slack/interactive.d.ts +39 -3
- package/dist/channels/slack/interactive.d.ts.map +1 -1
- package/dist/channels/slack/interactive.js +152 -7
- package/dist/channels/slack/manifest.d.ts +3 -0
- package/dist/channels/slack/manifest.d.ts.map +1 -1
- package/dist/channels/slack/manifest.js +3 -0
- package/dist/channels/slack/setup.d.ts +4 -1
- package/dist/channels/slack/setup.d.ts.map +1 -1
- package/dist/channels/slack/setup.js +16 -2
- package/dist/channels/slack/slack-channel.d.ts.map +1 -1
- package/dist/channels/slack/slack-channel.js +24 -3
- package/dist/channels/slack/types.d.ts +77 -0
- package/dist/channels/slack/types.d.ts.map +1 -1
- package/dist/connections.d.ts +4 -1
- package/dist/connections.d.ts.map +1 -1
- package/dist/connections.js +4 -1
- package/dist/docs/404.html +2 -2
- package/dist/docs/ab.html +3 -3
- package/dist/docs/assets/{app.DjHC-Ggf.js → app.BbEXr2KO.js} +1 -1
- package/dist/docs/assets/chunks/@localSearchIndexroot.DCBsrfwP.js +1 -0
- package/dist/docs/assets/chunks/{VPLocalSearchBox.DwviwmIq.js → VPLocalSearchBox.DBEi4HDp.js} +1 -1
- package/dist/docs/assets/chunks/{theme.BIpeYSon.js → theme.DGfptcHw.js} +2 -2
- package/dist/docs/assets/{guides_mcp-oauth.md.DqmEOQY1.js → guides_mcp-oauth.md.Dd8EgSem.js} +3 -2
- package/dist/docs/assets/{guides_mcp-oauth.md.DqmEOQY1.lean.js → guides_mcp-oauth.md.Dd8EgSem.lean.js} +1 -1
- package/dist/docs/assets/{guides_slack.md.FWale6Ip.js → guides_slack.md.D_xFKlnF.js} +1 -1
- package/dist/docs/assets/{reference_connections.md.CsSr3rDz.js → reference_connections.md.B9Q3TOve.js} +12 -5
- package/dist/docs/assets/{reference_connections.md.CsSr3rDz.lean.js → reference_connections.md.B9Q3TOve.lean.js} +1 -1
- package/dist/docs/building-with-agents.html +3 -3
- package/dist/docs/concepts.html +3 -3
- package/dist/docs/deployment.html +3 -3
- package/dist/docs/evals.html +3 -3
- package/dist/docs/example-agents/approval-buddy.html +3 -3
- package/dist/docs/example-agents/benny.html +3 -3
- package/dist/docs/example-agents/bugbot.html +3 -3
- package/dist/docs/example-agents/codebase-wiki.html +3 -3
- package/dist/docs/example-agents/codeowners-review.html +3 -3
- package/dist/docs/example-agents/concierge.html +3 -3
- package/dist/docs/example-agents/fsd.html +3 -3
- package/dist/docs/example-agents/index.html +3 -3
- package/dist/docs/example-agents/knowledge-base.html +3 -3
- package/dist/docs/example-agents/oncall.html +3 -3
- package/dist/docs/example-agents/security-reviewer.html +3 -3
- package/dist/docs/example-agents/slack-agent.html +3 -3
- package/dist/docs/example-agents/weather-agent.html +3 -3
- package/dist/docs/guides/agent-to-agent.html +3 -3
- package/dist/docs/guides/cloud-runtime.html +3 -3
- package/dist/docs/guides/github.html +3 -3
- package/dist/docs/guides/human-in-the-loop.html +3 -3
- package/dist/docs/guides/mcp-oauth.html +6 -5
- package/dist/docs/guides/slack.html +5 -5
- package/dist/docs/guides/webhooks.html +3 -3
- package/dist/docs/hashmap.json +1 -1
- package/dist/docs/hillclimbing.html +3 -3
- package/dist/docs/index.html +3 -3
- package/dist/docs/quickstart.html +3 -3
- package/dist/docs/reference/agent-config.html +3 -3
- package/dist/docs/reference/channels.html +3 -3
- package/dist/docs/reference/cli.html +3 -3
- package/dist/docs/reference/connections.html +14 -7
- package/dist/docs/reference/hooks.html +3 -3
- package/dist/docs/reference/http-api.html +3 -3
- package/dist/docs/reference/instructions.html +3 -3
- package/dist/docs/reference/playground.html +3 -3
- package/dist/docs/reference/project-layout.html +3 -3
- package/dist/docs/reference/prompt.html +3 -3
- package/dist/docs/reference/schedules.html +3 -3
- package/dist/docs/reference/sessions.html +3 -3
- package/dist/docs/reference/skills.html +3 -3
- package/dist/docs/reference/subagents.html +3 -3
- package/dist/docs/reference/tools.html +3 -3
- package/dist/docs/scaffolding-agents.html +3 -3
- package/dist/docs/storage.html +3 -3
- package/dist/docs/troubleshooting.html +3 -3
- package/dist/evals/assertions.d.ts +190 -0
- package/dist/evals/assertions.d.ts.map +1 -0
- package/dist/evals/assertions.js +461 -0
- package/dist/evals/expect.d.ts +95 -0
- package/dist/evals/expect.d.ts.map +1 -0
- package/dist/evals/expect.js +194 -0
- package/dist/evals/judge.d.ts +133 -0
- package/dist/evals/judge.d.ts.map +1 -0
- package/dist/evals/judge.js +228 -0
- package/dist/evals/loaders.d.ts +51 -0
- package/dist/evals/loaders.d.ts.map +1 -0
- package/dist/evals/loaders.js +124 -0
- package/dist/evals/matchers.d.ts +95 -0
- package/dist/evals/matchers.d.ts.map +1 -0
- package/dist/evals/matchers.js +145 -0
- package/dist/evals/reporters.d.ts +70 -0
- package/dist/evals/reporters.d.ts.map +1 -0
- package/dist/evals/reporters.js +221 -0
- package/dist/evals/results.d.ts +103 -0
- package/dist/evals/results.d.ts.map +1 -0
- package/dist/evals/results.js +28 -0
- package/dist/evals/run-facts.d.ts +77 -0
- package/dist/evals/run-facts.d.ts.map +1 -0
- package/dist/evals/run-facts.js +187 -0
- package/dist/evals.d.ts +118 -31
- package/dist/evals.d.ts.map +1 -1
- package/dist/evals.js +46 -12
- package/dist/index.d.ts +1 -1
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +1 -1
- package/dist/internal/cli-ax.d.ts +27 -0
- package/dist/internal/cli-ax.d.ts.map +1 -1
- package/dist/internal/cli-ax.js +90 -12
- package/dist/internal/cli-mcp.d.ts +8 -0
- package/dist/internal/cli-mcp.d.ts.map +1 -1
- package/dist/internal/cli-mcp.js +98 -0
- package/dist/internal/cli-slack.d.ts +29 -1
- package/dist/internal/cli-slack.d.ts.map +1 -1
- package/dist/internal/cli-slack.js +777 -12
- package/dist/internal/deploy-client.d.ts +7 -0
- package/dist/internal/deploy-client.d.ts.map +1 -1
- package/dist/internal/deploy-client.js +9 -0
- package/dist/internal/discovery.d.ts.map +1 -1
- package/dist/internal/discovery.js +29 -9
- package/dist/internal/distribution.d.ts.map +1 -1
- package/dist/internal/distribution.js +2 -0
- package/dist/internal/eval-judge-model.d.ts +59 -0
- package/dist/internal/eval-judge-model.d.ts.map +1 -0
- package/dist/internal/eval-judge-model.js +131 -0
- package/dist/internal/eval-run-store.d.ts.map +1 -1
- package/dist/internal/eval-run-store.js +12 -20
- package/dist/internal/eval-runner.d.ts +31 -25
- package/dist/internal/eval-runner.d.ts.map +1 -1
- package/dist/internal/eval-runner.js +329 -126
- package/dist/internal/evals-client.d.ts.map +1 -1
- package/dist/internal/evals-client.js +4 -1
- package/dist/internal/local-env.d.ts +8 -0
- package/dist/internal/local-env.d.ts.map +1 -0
- package/dist/internal/local-env.js +47 -0
- package/dist/internal/playground/static.d.ts +3 -0
- package/dist/internal/playground/static.d.ts.map +1 -1
- package/dist/internal/playground/static.js +3 -0
- package/dist/internal/run-client.d.ts +15 -0
- package/dist/internal/run-client.d.ts.map +1 -1
- package/dist/internal/run-client.js +3 -2
- package/dist/internal/sdk-runner.d.ts +36 -1
- package/dist/internal/sdk-runner.d.ts.map +1 -1
- package/dist/internal/sdk-runner.js +65 -25
- package/dist/internal/server.d.ts.map +1 -1
- package/dist/internal/server.js +10 -1
- package/dist/internal/session-engine.d.ts +18 -5
- package/dist/internal/session-engine.d.ts.map +1 -1
- package/dist/internal/session-engine.js +72 -19
- package/dist/internal/slack-provision-client.d.ts +130 -0
- package/dist/internal/slack-provision-client.d.ts.map +1 -0
- package/dist/internal/slack-provision-client.js +284 -0
- package/dist/playground/assets/cursor-icons-16-f_W_ogc-.woff2 +0 -0
- package/dist/playground/assets/index-CidizGZv.css +1 -0
- package/dist/playground/assets/index-DNqirsrK.js +85 -0
- package/dist/playground/index.html +2 -2
- package/dist/types.d.ts +70 -4
- package/dist/types.d.ts.map +1 -1
- package/dist/types.js +26 -0
- package/docs/guides/mcp-oauth.md +10 -5
- package/docs/guides/slack.md +3 -2
- package/docs/reference/connections.md +28 -18
- package/package.json +20 -2
- package/skills/create-agent/SKILL.md +15 -7
- package/skills/debug/SKILL.md +3 -1
- package/skills/mcp-auth/SKILL.md +1 -1
- package/skills/setup-slack/SKILL.md +173 -11
- package/src/bin/agent-serve.ts +63 -2
- package/src/channels/slack/api.ts +16 -3
- package/src/channels/slack/auth.ts +16 -1
- package/src/channels/slack/blocks.ts +139 -0
- package/src/channels/slack/defaults.ts +32 -4
- package/src/channels/slack/dispatch.ts +3 -18
- package/src/channels/slack/index.ts +1 -0
- package/src/channels/slack/interactive.ts +183 -19
- package/src/channels/slack/manifest.ts +3 -0
- package/src/channels/slack/setup.ts +16 -2
- package/src/channels/slack/slack-channel.ts +37 -6
- package/src/channels/slack/types.ts +89 -0
- package/src/connections.ts +4 -1
- package/src/evals/assertions.ts +704 -0
- package/src/evals/expect.ts +272 -0
- package/src/evals/judge.ts +346 -0
- package/src/evals/loaders.ts +113 -0
- package/src/evals/matchers.ts +246 -0
- package/src/evals/reporters.ts +274 -0
- package/src/evals/results.ts +132 -0
- package/src/evals/run-facts.ts +274 -0
- package/src/evals.ts +205 -42
- package/src/index.ts +2 -0
- package/src/internal/cli-ax.ts +134 -15
- package/src/internal/cli-mcp.ts +118 -5
- package/src/internal/cli-slack.ts +993 -15
- package/src/internal/deploy-client.ts +10 -0
- package/src/internal/discovery.ts +31 -5
- package/src/internal/distribution.ts +2 -0
- package/src/internal/eval-judge-model.ts +148 -0
- package/src/internal/eval-run-store.ts +10 -14
- package/src/internal/eval-runner.ts +460 -214
- package/src/internal/evals-client.ts +5 -0
- package/src/internal/local-env.ts +53 -0
- package/src/internal/playground/static.ts +3 -0
- package/src/internal/run-client.ts +17 -1
- package/src/internal/sdk-runner.ts +103 -22
- package/src/internal/server.ts +10 -0
- package/src/internal/session-engine.ts +89 -21
- package/src/internal/slack-provision-client.ts +438 -0
- package/src/types.ts +92 -4
- package/dist/channels/slack/post-update-delivery.d.ts +0 -85
- package/dist/channels/slack/post-update-delivery.d.ts.map +0 -1
- package/dist/docs/assets/chunks/@localSearchIndexroot.DZtu0ZIc.js +0 -1
- package/dist/internal/json-dir-store.d.ts +0 -32
- package/dist/internal/json-dir-store.d.ts.map +0 -1
- package/dist/internal/persistence-coordinator.d.ts +0 -127
- package/dist/internal/persistence-coordinator.d.ts.map +0 -1
- package/dist/persistence.d.ts +0 -184
- package/dist/persistence.d.ts.map +0 -1
- package/dist/playground/assets/cursor-icons-16-CQ50JpfO.woff2 +0 -0
- package/dist/playground/assets/index-Bwl4Onx4.js +0 -85
- package/dist/playground/assets/index-ByV4nfch.css +0 -1
- /package/dist/docs/assets/{guides_slack.md.FWale6Ip.lean.js → guides_slack.md.D_xFKlnF.lean.js} +0 -0
package/dist/evals.d.ts
CHANGED
|
@@ -44,33 +44,62 @@
|
|
|
44
44
|
* ],
|
|
45
45
|
* });
|
|
46
46
|
* ```
|
|
47
|
+
*
|
|
48
|
+
* ## Severity: gates and scores
|
|
49
|
+
*
|
|
50
|
+
* Every assertion returns a chainable handle, so severity rides on the
|
|
51
|
+
* assertion instead of a separate thresholds map. Gates are hard (a miss fails
|
|
52
|
+
* the eval and `eval` exits non-zero); soft assertions are tracked scores that
|
|
53
|
+
* mark the eval `scored` and only fail the build under `--strict`:
|
|
54
|
+
*
|
|
55
|
+
* ```ts
|
|
56
|
+
* t.succeeded(); // gate (default)
|
|
57
|
+
* t.calledTool("get_weather").soft(); // tracked, no gate
|
|
58
|
+
* t.check(t.reply, similarity("Sunny, 72F")).atLeast(0.8);
|
|
59
|
+
* t.judge.closedQA("cites a source").atLeast(0.6); // LLM-graded
|
|
60
|
+
* ```
|
|
61
|
+
*
|
|
62
|
+
* Matchers narrow tool assertions (`{ input, output, status, count }` accept a
|
|
63
|
+
* literal, a RegExp, or a predicate), and each turn returned by `t.send(...)`
|
|
64
|
+
* carries the same vocabulary scoped to that turn.
|
|
47
65
|
*/
|
|
66
|
+
import type { EvalAssertionResult, EvalAssertions, EvalVerdict } from "./evals/assertions.js";
|
|
67
|
+
import type { Expectation } from "./evals/expect.js";
|
|
68
|
+
import type { EvalJudge, EvalJudgeConfig } from "./evals/judge.js";
|
|
69
|
+
import type { EvalReporter } from "./evals/results.js";
|
|
48
70
|
import type { AgentCloudOptions, SessionEvent } from "./types.js";
|
|
49
|
-
export type
|
|
50
|
-
|
|
51
|
-
|
|
52
|
-
}
|
|
53
|
-
|
|
54
|
-
|
|
55
|
-
}
|
|
56
|
-
|
|
57
|
-
|
|
58
|
-
|
|
59
|
-
|
|
60
|
-
|
|
61
|
-
|
|
62
|
-
|
|
63
|
-
|
|
64
|
-
|
|
65
|
-
|
|
66
|
-
|
|
67
|
-
|
|
68
|
-
|
|
71
|
+
export type { EvalAssertionHandle, EvalAssertionResult, EvalAssertionSeverity, EvalAssertions, EvalVerdict, } from "./evals/assertions.js";
|
|
72
|
+
export { computeVerdict, verdictFailsRun } from "./evals/assertions.js";
|
|
73
|
+
export type { EvalSchemaLike, Expectation } from "./evals/expect.js";
|
|
74
|
+
export { equals, includes, matches, normalizedSimilarity, satisfies, similarity, } from "./evals/expect.js";
|
|
75
|
+
export type { EvalJudge, EvalJudgeCallOptions, EvalJudgeConfig, EvalJudgeGraderName, } from "./evals/judge.js";
|
|
76
|
+
export { EVAL_JUDGE_INJECTION_GUARD, fenceUntrusted, } from "./evals/judge.js";
|
|
77
|
+
export type { EvalCountMatcher, EvalEventMatcher, EvalEventOrderMatcher, EvalMatcher, EvalSubagentMatcher, EvalToolCallMatcher, EvalToolCallStatus, } from "./evals/matchers.js";
|
|
78
|
+
export type { EvalReporter, EvalReporterEvalInfo, EvalReporterTarget, EvalRunResult, EvalRunSummary, } from "./evals/results.js";
|
|
79
|
+
export type { EvalInputRequestFact, EvalRunFacts, EvalSubagentFact, EvalToolCallFact, } from "./evals/run-facts.js";
|
|
80
|
+
export { deriveRunFacts } from "./evals/run-facts.js";
|
|
81
|
+
/**
|
|
82
|
+
* One settled turn. Carries the same assertion vocabulary as `t`, narrowed to
|
|
83
|
+
* this turn's events, so an intermediate turn can be graded before the next
|
|
84
|
+
* `send` overwrites `t.reply`:
|
|
85
|
+
*
|
|
86
|
+
* ```ts
|
|
87
|
+
* const draft = await t.send("Draft the summary.");
|
|
88
|
+
* draft.calledTool("read_file");
|
|
89
|
+
* t.check(draft.message, includes("Summary"));
|
|
90
|
+
* ```
|
|
91
|
+
*/
|
|
92
|
+
export interface EvalTurnResult extends EvalAssertions {
|
|
69
93
|
message: string;
|
|
70
94
|
sessionId: string;
|
|
71
95
|
events: SessionEvent[];
|
|
96
|
+
/** Tool names requested during this turn, in order. */
|
|
72
97
|
toolCalls: string[];
|
|
73
98
|
ok: boolean;
|
|
99
|
+
/** 1-based turn index within the case. */
|
|
100
|
+
index: number;
|
|
101
|
+
/** Throws when this turn failed. Only needed when later steps depend on it. */
|
|
102
|
+
expectOk(): EvalTurnResult;
|
|
74
103
|
}
|
|
75
104
|
export interface EvalSendOptions {
|
|
76
105
|
/** Absolute harness `cwd` for a new session (`SendMessageOptions.workspaceDir`). */
|
|
@@ -84,7 +113,15 @@ export interface EvalSendOptions {
|
|
|
84
113
|
*/
|
|
85
114
|
cloud?: AgentCloudOptions;
|
|
86
115
|
}
|
|
87
|
-
|
|
116
|
+
/**
|
|
117
|
+
* The eval driver and assertion surface.
|
|
118
|
+
*
|
|
119
|
+
* `t` both drives the agent and grades it. Assertions on `t` read the whole run
|
|
120
|
+
* once `test` finishes; assertions on the turn returned by `t.send(...)` read
|
|
121
|
+
* only that turn. Recorded assertions never throw, so one run reports every
|
|
122
|
+
* failure rather than dying on the first.
|
|
123
|
+
*/
|
|
124
|
+
export interface EvalTestContext extends EvalAssertions {
|
|
88
125
|
/** Send a turn and wait until the session parks (or fails). */
|
|
89
126
|
send(message: string, options?: EvalSendOptions): Promise<EvalTurnResult>;
|
|
90
127
|
/** Last assistant text, or null before the first send. */
|
|
@@ -92,6 +129,8 @@ export interface EvalTestContext {
|
|
|
92
129
|
readonly sessionId: string | undefined;
|
|
93
130
|
/** All events captured so far across turns. */
|
|
94
131
|
readonly events: SessionEvent[];
|
|
132
|
+
/** Turns settled so far, oldest first. */
|
|
133
|
+
readonly turns: readonly EvalTurnResult[];
|
|
95
134
|
/**
|
|
96
135
|
* 1-based index of this run when {@link EvalConfig.iterations} /
|
|
97
136
|
* {@link EvalDatapoint.iterations} expands a datapoint. Always `1` when
|
|
@@ -100,16 +139,25 @@ export interface EvalTestContext {
|
|
|
100
139
|
readonly iteration: number;
|
|
101
140
|
/** Total iterations for this datapoint (default `1`). */
|
|
102
141
|
readonly iterations: number;
|
|
103
|
-
/**
|
|
104
|
-
|
|
105
|
-
/**
|
|
106
|
-
|
|
107
|
-
|
|
108
|
-
|
|
109
|
-
|
|
110
|
-
|
|
111
|
-
/**
|
|
112
|
-
|
|
142
|
+
/** Aborts when the case hits its timeout. */
|
|
143
|
+
readonly signal: AbortSignal;
|
|
144
|
+
/**
|
|
145
|
+
* LLM-as-judge assertions. Soft by default (tracked, never fatal); add a bar
|
|
146
|
+
* with `.atLeast(0.7)` or promote with `.gate(0.8)`. Requires a judge model
|
|
147
|
+
* from `defineEvalConfig({ judge })`, `defineEval({ judge })`, or the call.
|
|
148
|
+
*/
|
|
149
|
+
readonly judge: EvalJudge;
|
|
150
|
+
/**
|
|
151
|
+
* Record a gate that must pass before the script can safely continue.
|
|
152
|
+
* Returns `value` when it passes and stops the test body when it does not,
|
|
153
|
+
* without adding a duplicate execution error.
|
|
154
|
+
*/
|
|
155
|
+
require<T>(value: T, assertion: Expectation): Promise<T>;
|
|
156
|
+
/**
|
|
157
|
+
* End the case as skipped. Skipped cases are reported separately and never
|
|
158
|
+
* change the exit code. Call it before sending messages.
|
|
159
|
+
*/
|
|
160
|
+
skip(reason: string): never;
|
|
113
161
|
/** Debug line (surfaced under `agent-serve eval --verbose`). */
|
|
114
162
|
log(message: string): void;
|
|
115
163
|
/**
|
|
@@ -139,6 +187,12 @@ export interface EvalDatapoint {
|
|
|
139
187
|
* concurrent case.
|
|
140
188
|
*/
|
|
141
189
|
iterations?: number;
|
|
190
|
+
/** Free-form data carried onto the result and every reporter. */
|
|
191
|
+
metadata?: Record<string, unknown>;
|
|
192
|
+
/** Judge model for this datapoint (overrides `evals.config.ts`). */
|
|
193
|
+
judge?: EvalJudgeConfig;
|
|
194
|
+
/** Destinations for just this datapoint, in addition to config reporters. */
|
|
195
|
+
reporters?: EvalReporter[];
|
|
142
196
|
test: (t: EvalTestContext) => Promise<void>;
|
|
143
197
|
}
|
|
144
198
|
/**
|
|
@@ -154,6 +208,12 @@ export interface EvalConfig {
|
|
|
154
208
|
* Case-level {@link EvalDatapoint.iterations} overrides this.
|
|
155
209
|
*/
|
|
156
210
|
iterations?: number;
|
|
211
|
+
/** Free-form data carried onto every result in this file and to reporters. */
|
|
212
|
+
metadata?: Record<string, unknown>;
|
|
213
|
+
/** Judge model for this file (overrides `evals.config.ts`). */
|
|
214
|
+
judge?: EvalJudgeConfig;
|
|
215
|
+
/** Destinations for this file, in addition to config reporters. */
|
|
216
|
+
reporters?: EvalReporter[];
|
|
157
217
|
/** Single-datapoint shorthand. Case id = file path under `evals/`. */
|
|
158
218
|
test?: (t: EvalTestContext) => Promise<void>;
|
|
159
219
|
/** Multiple datapoints in this file. Each id = `<fileId>/<case.id>`. */
|
|
@@ -168,6 +228,9 @@ export interface EvalCaseDefinition {
|
|
|
168
228
|
iterations?: number;
|
|
169
229
|
/** 1-based iteration index for this expanded runnable case. */
|
|
170
230
|
iteration?: number;
|
|
231
|
+
metadata?: Record<string, unknown>;
|
|
232
|
+
judge?: EvalJudgeConfig;
|
|
233
|
+
reporters?: EvalReporter[];
|
|
171
234
|
test: (t: EvalTestContext) => Promise<void>;
|
|
172
235
|
readonly __agentServe: "eval";
|
|
173
236
|
}
|
|
@@ -239,6 +302,10 @@ export interface EvalCaseSnapshot {
|
|
|
239
302
|
tags?: string[];
|
|
240
303
|
status: EvalCaseRunStatus;
|
|
241
304
|
ok?: boolean;
|
|
305
|
+
/** Full grade, including `scored` (soft threshold miss) and `skipped`. */
|
|
306
|
+
verdict?: EvalVerdict;
|
|
307
|
+
/** Reason passed to `t.skip(...)`. */
|
|
308
|
+
skipReason?: string;
|
|
242
309
|
assertions?: EvalAssertionResult[];
|
|
243
310
|
sessionId?: string;
|
|
244
311
|
durationMs?: number;
|
|
@@ -246,6 +313,8 @@ export interface EvalCaseSnapshot {
|
|
|
246
313
|
logs?: string[];
|
|
247
314
|
/** Structured scores from {@link EvalTestContext.metric}. */
|
|
248
315
|
metrics?: Record<string, string | number>;
|
|
316
|
+
/** Author-supplied metadata from `defineEval({ metadata })`. */
|
|
317
|
+
metadata?: Record<string, unknown>;
|
|
249
318
|
inputs?: string[];
|
|
250
319
|
finalText?: string;
|
|
251
320
|
tools?: string[];
|
|
@@ -266,6 +335,13 @@ export interface EvalRunSnapshot {
|
|
|
266
335
|
summary: {
|
|
267
336
|
passed: number;
|
|
268
337
|
failed: number;
|
|
338
|
+
/**
|
|
339
|
+
* Cases that only missed a soft threshold (fatal only under `--strict`).
|
|
340
|
+
* Optional: a batch persisted by an older serve process has no such count.
|
|
341
|
+
*/
|
|
342
|
+
scored?: number;
|
|
343
|
+
/** Cases ended by `t.skip(...)`; they never change the exit code. */
|
|
344
|
+
skipped?: number;
|
|
269
345
|
total: number;
|
|
270
346
|
done: number;
|
|
271
347
|
};
|
|
@@ -326,6 +402,17 @@ export interface EvalConfigFile {
|
|
|
326
402
|
* request limits.
|
|
327
403
|
*/
|
|
328
404
|
maxConcurrency: number;
|
|
405
|
+
/**
|
|
406
|
+
* Default judge model for `t.judge.*` across this tree. Optional: a tree of
|
|
407
|
+
* fully deterministic evals never needs one.
|
|
408
|
+
*/
|
|
409
|
+
judge?: EvalJudgeConfig;
|
|
410
|
+
/**
|
|
411
|
+
* Destinations that observe **every** case in the run — the right place for a
|
|
412
|
+
* shared one, so you do not repeat it in each eval file. `--skip-report`
|
|
413
|
+
* suppresses them for a local iteration loop.
|
|
414
|
+
*/
|
|
415
|
+
reporters?: EvalReporter[];
|
|
329
416
|
/**
|
|
330
417
|
* Optional. Max batches kept in the **playground / `/v1/dev/evals*`**
|
|
331
418
|
* history window (in-memory, and pruned from the `defineStorage` evals
|
package/dist/evals.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"evals.d.ts","sourceRoot":"","sources":["../src/evals.ts"],"names":[],"mappings":"AAAA
|
|
1
|
+
{"version":3,"file":"evals.d.ts","sourceRoot":"","sources":["../src/evals.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAgEG;AAEH,OAAO,KAAK,EACV,mBAAmB,EACnB,cAAc,EACd,WAAW,EACZ,MAAM,uBAAuB,CAAC;AAC/B,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AACrD,OAAO,KAAK,EAAE,SAAS,EAAE,eAAe,EAAE,MAAM,kBAAkB,CAAC;AACnE,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAGvD,OAAO,KAAK,EAAE,iBAAiB,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAKlE,YAAY,EACV,mBAAmB,EACnB,mBAAmB,EACnB,qBAAqB,EACrB,cAAc,EACd,WAAW,GACZ,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EAAE,cAAc,EAAE,eAAe,EAAE,MAAM,uBAAuB,CAAC;AACxE,YAAY,EAAE,cAAc,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AACrE,OAAO,EACL,MAAM,EACN,QAAQ,EACR,OAAO,EACP,oBAAoB,EACpB,SAAS,EACT,UAAU,GACX,MAAM,mBAAmB,CAAC;AAC3B,YAAY,EACV,SAAS,EACT,oBAAoB,EACpB,eAAe,EACf,mBAAmB,GACpB,MAAM,kBAAkB,CAAC;AAG1B,OAAO,EACL,0BAA0B,EAC1B,cAAc,GACf,MAAM,kBAAkB,CAAC;AAC1B,YAAY,EACV,gBAAgB,EAChB,gBAAgB,EAChB,qBAAqB,EACrB,WAAW,EACX,mBAAmB,EACnB,mBAAmB,EACnB,kBAAkB,GACnB,MAAM,qBAAqB,CAAC;AAC7B,YAAY,EACV,YAAY,EACZ,oBAAoB,EACpB,kBAAkB,EAClB,aAAa,EACb,cAAc,GACf,MAAM,oBAAoB,CAAC;AAC5B,YAAY,EACV,oBAAoB,EACpB,YAAY,EACZ,gBAAgB,EAChB,gBAAgB,GACjB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AAEtD;;;;;;;;;;GAUG;AACH,MAAM,WAAW,cAAe,SAAQ,cAAc;IACpD,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,YAAY,EAAE,CAAC;IACvB,uDAAuD;IACvD,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,EAAE,EAAE,OAAO,CAAC;IACZ,0CAA0C;IAC1C,KAAK,EAAE,MAAM,CAAC;IACd,+EAA+E;IAC/E,QAAQ,IAAI,cAAc,CAAC;CAC5B;AAED,MAAM,WAAW,eAAe;IAC9B,oFAAoF;IACpF,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,qEAAqE;IACrE,cAAc,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACxC;;;;OAIG;IACH,KAAK,CAAC,EAAE,iBAAiB,CAAC;CAC3B;AAED;;;;;;;GAOG;AACH,MAAM,WAAW,eAAgB,SAAQ,cAAc;IACrD,+DAA+D;IAC/D,IAAI,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,eAAe,GAAG,OAAO,CAAC,cAAc,CAAC,CAAC;IAC1E,0DAA0D;IAC1D,QAAQ,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,QAAQ,CAAC,SAAS,EAAE,MAAM,GAAG,SAAS,CAAC;IACvC,+CAA+C;IAC/C,QAAQ,CAAC,MAAM,EAAE,YAAY,EAAE,CAAC;IAChC,0CAA0C;IAC1C,QAAQ,CAAC,KAAK,EAAE,SAAS,cAAc,EAAE,CAAC;IAC1C;;;;OAIG;IACH,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,yDAAyD;IACzD,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,6CAA6C;IAC7C,QAAQ,CAAC,MAAM,EAAE,WAAW,CAAC;IAC7B;;;;OAIG;IACH,QAAQ,CAAC,KAAK,EAAE,SAAS,CAAC;IAC1B;;;;OAIG;IACH,OAAO,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,SAAS,EAAE,WAAW,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;IACzD;;;OAGG;IACH,IAAI,CAAC,MAAM,EAAE,MAAM,GAAG,KAAK,CAAC;IAC5B,gEAAgE;IAChE,GAAG,CAAC,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B;;;;OAIG;IACH,MAAM,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAAC;CACpD;AAED,0EAA0E;AAC1E,eAAO,MAAM,mBAAmB,MAAM,CAAC;AAEvC,0EAA0E;AAC1E,MAAM,WAAW,aAAa;IAC5B;;;;OAIG;IACH,EAAE,EAAE,MAAM,CAAC;IACX,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;;;OAKG;IACH,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,iEAAiE;IACjE,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACnC,oEAAoE;IACpE,KAAK,CAAC,EAAE,eAAe,CAAC;IACxB,6EAA6E;IAC7E,SAAS,CAAC,EAAE,YAAY,EAAE,CAAC;IAC3B,IAAI,EAAE,CAAC,CAAC,EAAE,eAAe,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;CAC7C;AAED;;;GAGG;AACH,MAAM,WAAW,UAAU;IACzB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;OAGG;IACH,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,8EAA8E;IAC9E,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACnC,+DAA+D;IAC/D,KAAK,CAAC,EAAE,eAAe,CAAC;IACxB,mEAAmE;IACnE,SAAS,CAAC,EAAE,YAAY,EAAE,CAAC;IAC3B,sEAAsE;IACtE,IAAI,CAAC,EAAE,CAAC,CAAC,EAAE,eAAe,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;IAC7C,wEAAwE;IACxE,KAAK,CAAC,EAAE,aAAa,EAAE,CAAC;CACzB;AAED,sEAAsE;AACtE,MAAM,WAAW,kBAAkB;IACjC,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,8DAA8D;IAC9D,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,+DAA+D;IAC/D,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACnC,KAAK,CAAC,EAAE,eAAe,CAAC;IACxB,SAAS,CAAC,EAAE,YAAY,EAAE,CAAC;IAC3B,IAAI,EAAE,CAAC,CAAC,EAAE,eAAe,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;IAC5C,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;CAC/B;AAED,MAAM,MAAM,cAAc,GAAG,UAAU,GAAG;IAAE,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAA;CAAE,CAAC;AAE5E;;;;GAIG;AACH,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,OAAO,GAAG,MAAM,CAK9D;AAED,wBAAgB,UAAU,CAAC,MAAM,EAAE,UAAU,GAAG,cAAc,CAoC7D;AAaD;;;;;;;GAOG;AACH,wBAAgB,oBAAoB,CAClC,MAAM,EAAE,MAAM,EACd,UAAU,EAAE,cAAc,GACzB,KAAK,CAAC;IAAE,EAAE,EAAE,MAAM,CAAC;IAAC,UAAU,EAAE,kBAAkB,CAAA;CAAE,CAAC,CAwCvD;AAqDD;;;;;GAKG;AACH,wBAAgB,mBAAmB,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,GAAG,OAAO,CAE3E;AAED;;;;;GAKG;AACH,eAAO,MAAM,gBAAgB,YAAY,CAAC;AAE1C,sDAAsD;AACtD,wBAAgB,iBAAiB,CAC/B,IAAI,EAAE,SAAS,MAAM,EAAE,GAAG,SAAS,GAClC,OAAO,CAET;AAED;;;;GAIG;AACH,wBAAgB,2BAA2B,CACzC,CAAC,SAAS;IAAE,EAAE,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAA;CAAE,EAExC,GAAG,EAAE,SAAS,CAAC,EAAE,EACjB,QAAQ,EAAE,SAAS,CAAC,EAAE,EACtB,SAAS,EAAE,CAAC,IAAI,EAAE,CAAC,KAAK,OAAO,GAC9B,CAAC,EAAE,CAgBL;AAED;;;;;GAKG;AACH,eAAO,MAAM,0BAA0B,MAAM,CAAC;AAE9C,gGAAgG;AAChG,eAAO,MAAM,gCAAgC,KAAK,CAAC;AAEnD,iEAAiE;AACjE,eAAO,MAAM,wBAAwB,MAAM,CAAC;AAE5C,MAAM,MAAM,iBAAiB,GAAG,SAAS,GAAG,SAAS,GAAG,MAAM,CAAC;AAC/D,MAAM,MAAM,eAAe,GAAG,SAAS,GAAG,WAAW,GAAG,QAAQ,GAAG,WAAW,CAAC;AAE/E,+DAA+D;AAC/D,MAAM,WAAW,gBAAgB;IAC/B,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,MAAM,EAAE,iBAAiB,CAAC;IAC1B,EAAE,CAAC,EAAE,OAAO,CAAC;IACb,0EAA0E;IAC1E,OAAO,CAAC,EAAE,WAAW,CAAC;IACtB,sCAAsC;IACtC,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,UAAU,CAAC,EAAE,mBAAmB,EAAE,CAAC;IACnC,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,6DAA6D;IAC7D,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,CAAC,CAAC;IAC1C,gEAAgE;IAChE,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACnC,MAAM,CAAC,EAAE,MAAM,EAAE,CAAC;IAClB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,MAAM,EAAE,CAAC;IACjB,SAAS,CAAC,EAAE,KAAK,CAAC;QAChB,QAAQ,EAAE,MAAM,CAAC;QACjB,IAAI,CAAC,EAAE,OAAO,CAAC;QACf,MAAM,CAAC,EAAE,OAAO,CAAC;QACjB,OAAO,EAAE,OAAO,CAAC;KAClB,CAAC,CAAC;CACJ;AAED,gFAAgF;AAChF,MAAM,WAAW,eAAe;IAC9B,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,eAAe,CAAC;IACxB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,OAAO,EAAE;QACP,MAAM,EAAE,MAAM,CAAC;QACf,MAAM,EAAE,MAAM,CAAC;QACf;;;WAGG;QACH,MAAM,CAAC,EAAE,MAAM,CAAC;QAChB,qEAAqE;QACrE,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,KAAK,EAAE,MAAM,CAAC;QACd,IAAI,EAAE,MAAM,CAAC;KACd,CAAC;IACF,KAAK,EAAE,gBAAgB,EAAE,CAAC;IAC1B,MAAM,EAAE;QACN,cAAc,CAAC,EAAE,MAAM,CAAC;QACxB,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,iBAAiB,CAAC,EAAE,MAAM,CAAC;QAC3B,WAAW,CAAC,EAAE,OAAO,CAAC;KACvB,CAAC;CACH;AAED,8EAA8E;AAC9E,MAAM,WAAW,gBAAgB;IAC/B,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;CACjB;AAED,4DAA4D;AAC5D,MAAM,WAAW,kBAAkB;IACjC,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,8DAA8D;IAC9D,WAAW,EAAE,OAAO,CAAC;CACtB;AAED,wFAAwF;AACxF,MAAM,WAAW,qBAAqB;IACpC,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,wEAAwE;IACxE,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED;;;;GAIG;AACH,MAAM,WAAW,YAAY;IAC3B,IAAI,IAAI,OAAO,CAAC;QACd,KAAK,EAAE,gBAAgB,EAAE,CAAC;QAC1B,MAAM,EAAE,kBAAkB,CAAC;KAC5B,CAAC,CAAC;IACH,GAAG,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,eAAe,GAAG,SAAS,CAAC,CAAC;IACzD,kBAAkB,IAAI,OAAO,CAAC;QAC5B,IAAI,EAAE,eAAe,EAAE,CAAC;QACxB,WAAW,CAAC,EAAE,MAAM,CAAC;KACtB,CAAC,CAAC;IACH,KAAK,CAAC,KAAK,CAAC,EAAE,qBAAqB,GAAG,OAAO,CAAC,eAAe,CAAC,CAAC;IAC/D,mDAAmD;IACnD,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,eAAe,CAAC,CAAC;CACjD;AAED,MAAM,WAAW,cAAc;IAC7B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;;OAIG;IACH,cAAc,EAAE,MAAM,CAAC;IACvB;;;OAGG;IACH,KAAK,CAAC,EAAE,eAAe,CAAC;IACxB;;;;OAIG;IACH,SAAS,CAAC,EAAE,YAAY,EAAE,CAAC;IAC3B;;;;;;;;OAQG;IACH,iBAAiB,CAAC,EAAE,MAAM,CAAC;CAC5B;AAED;;;;;;GAMG;AACH,wBAAgB,gBAAgB,CAAC,MAAM,EAAE,cAAc,GAAG,cAAc,CAUvE;AAED;;GAEG;AACH,wBAAgB,2BAA2B,CAAC,KAAK,EAAE,OAAO,GAAG,MAAM,CAMlE;AAED,yDAAyD;AACzD,wBAAgB,8BAA8B,CAAC,KAAK,EAAE,OAAO,GAAG,MAAM,CAKrE;AAED;;;GAGG;AACH,wBAAgB,4BAA4B,CAAC,KAAK,EAAE,OAAO,GAAG,MAAM,CASnE"}
|
package/dist/evals.js
CHANGED
|
@@ -44,18 +44,33 @@
|
|
|
44
44
|
* ],
|
|
45
45
|
* });
|
|
46
46
|
* ```
|
|
47
|
+
*
|
|
48
|
+
* ## Severity: gates and scores
|
|
49
|
+
*
|
|
50
|
+
* Every assertion returns a chainable handle, so severity rides on the
|
|
51
|
+
* assertion instead of a separate thresholds map. Gates are hard (a miss fails
|
|
52
|
+
* the eval and `eval` exits non-zero); soft assertions are tracked scores that
|
|
53
|
+
* mark the eval `scored` and only fail the build under `--strict`:
|
|
54
|
+
*
|
|
55
|
+
* ```ts
|
|
56
|
+
* t.succeeded(); // gate (default)
|
|
57
|
+
* t.calledTool("get_weather").soft(); // tracked, no gate
|
|
58
|
+
* t.check(t.reply, similarity("Sunny, 72F")).atLeast(0.8);
|
|
59
|
+
* t.judge.closedQA("cites a source").atLeast(0.6); // LLM-graded
|
|
60
|
+
* ```
|
|
61
|
+
*
|
|
62
|
+
* Matchers narrow tool assertions (`{ input, output, status, count }` accept a
|
|
63
|
+
* literal, a RegExp, or a predicate), and each turn returned by `t.send(...)`
|
|
64
|
+
* carries the same vocabulary scoped to that turn.
|
|
47
65
|
*/
|
|
48
66
|
import { normalizePositiveInt } from "./internal/bounded-int.js";
|
|
49
67
|
import { brandDefinition } from "./internal/brand.js";
|
|
50
|
-
export
|
|
51
|
-
|
|
52
|
-
|
|
53
|
-
|
|
54
|
-
|
|
55
|
-
}
|
|
56
|
-
export function satisfies(predicate, label) {
|
|
57
|
-
return { kind: "satisfies", predicate, label };
|
|
58
|
-
}
|
|
68
|
+
export { computeVerdict, verdictFailsRun } from "./evals/assertions.js";
|
|
69
|
+
export { equals, includes, matches, normalizedSimilarity, satisfies, similarity, } from "./evals/expect.js";
|
|
70
|
+
// A custom judge built on `t.judge.model` embeds the same attacker-controlled
|
|
71
|
+
// agent output the built-in graders do, so it needs the same two protections.
|
|
72
|
+
export { EVAL_JUDGE_INJECTION_GUARD, fenceUntrusted, } from "./evals/judge.js";
|
|
73
|
+
export { deriveRunFacts } from "./evals/run-facts.js";
|
|
59
74
|
/** Hard ceiling for {@link EvalConfig.iterations} / case `iterations`. */
|
|
60
75
|
export const EVAL_MAX_ITERATIONS = 100;
|
|
61
76
|
/**
|
|
@@ -122,16 +137,22 @@ export function expandEvalDefinition(fileId, definition) {
|
|
|
122
137
|
var _a;
|
|
123
138
|
if (definition.cases !== undefined) {
|
|
124
139
|
return definition.cases.flatMap((datapoint) => {
|
|
125
|
-
var _a, _b, _c, _d, _e;
|
|
140
|
+
var _a, _b, _c, _d, _e, _f;
|
|
126
141
|
const description = (_a = datapoint.description) !== null && _a !== void 0 ? _a : definition.description;
|
|
127
142
|
const timeoutMs = (_b = datapoint.timeoutMs) !== null && _b !== void 0 ? _b : definition.timeoutMs;
|
|
128
143
|
const iterations = normalizeEvalIterations((_d = (_c = datapoint.iterations) !== null && _c !== void 0 ? _c : definition.iterations) !== null && _d !== void 0 ? _d : 1);
|
|
144
|
+
const metadata = datapoint.metadata === undefined && definition.metadata === undefined
|
|
145
|
+
? undefined
|
|
146
|
+
: Object.assign(Object.assign({}, definition.metadata), datapoint.metadata);
|
|
129
147
|
return expandIterations({
|
|
130
148
|
baseId: `${fileId}/${datapoint.id}`,
|
|
131
149
|
description,
|
|
132
150
|
tags: (_e = datapoint.tags) !== null && _e !== void 0 ? _e : definition.tags,
|
|
133
151
|
timeoutMs,
|
|
134
152
|
iterations,
|
|
153
|
+
metadata,
|
|
154
|
+
judge: (_f = datapoint.judge) !== null && _f !== void 0 ? _f : definition.judge,
|
|
155
|
+
reporters: mergeReporters(definition.reporters, datapoint.reporters),
|
|
135
156
|
test: datapoint.test,
|
|
136
157
|
});
|
|
137
158
|
});
|
|
@@ -146,9 +167,22 @@ export function expandEvalDefinition(fileId, definition) {
|
|
|
146
167
|
tags: definition.tags,
|
|
147
168
|
timeoutMs: definition.timeoutMs,
|
|
148
169
|
iterations,
|
|
170
|
+
metadata: definition.metadata,
|
|
171
|
+
judge: definition.judge,
|
|
172
|
+
reporters: definition.reporters,
|
|
149
173
|
test: definition.test,
|
|
150
174
|
});
|
|
151
175
|
}
|
|
176
|
+
/** Union file-level and case-level reporters, keeping each instance once. */
|
|
177
|
+
function mergeReporters(fileLevel, caseLevel) {
|
|
178
|
+
if (fileLevel === undefined) {
|
|
179
|
+
return caseLevel;
|
|
180
|
+
}
|
|
181
|
+
if (caseLevel === undefined) {
|
|
182
|
+
return fileLevel;
|
|
183
|
+
}
|
|
184
|
+
return [...new Set([...fileLevel, ...caseLevel])];
|
|
185
|
+
}
|
|
152
186
|
function expandIterations(args) {
|
|
153
187
|
const out = [];
|
|
154
188
|
for (let iteration = 1; iteration <= args.iterations; iteration++) {
|
|
@@ -158,7 +192,7 @@ function expandIterations(args) {
|
|
|
158
192
|
: `${args.description} (iteration ${iteration}/${args.iterations})`;
|
|
159
193
|
out.push({
|
|
160
194
|
id,
|
|
161
|
-
definition: brandDefinition("eval", Object.assign(Object.assign(Object.assign(Object.assign({}, (description !== undefined ? { description } : {})), (args.tags !== undefined ? { tags: args.tags } : {})), (args.timeoutMs !== undefined ? { timeoutMs: args.timeoutMs } : {})), { iterations: args.iterations, iteration, test: args.test })),
|
|
195
|
+
definition: brandDefinition("eval", Object.assign(Object.assign(Object.assign(Object.assign({}, (description !== undefined ? { description } : {})), (args.tags !== undefined ? { tags: args.tags } : {})), (args.timeoutMs !== undefined ? { timeoutMs: args.timeoutMs } : {})), { metadata: args.metadata, judge: args.judge, reporters: args.reporters, iterations: args.iterations, iteration, test: args.test })),
|
|
162
196
|
});
|
|
163
197
|
}
|
|
164
198
|
return out;
|
|
@@ -222,7 +256,7 @@ export const EVAL_MAX_PLAYGROUND_RUNS = 500;
|
|
|
222
256
|
*/
|
|
223
257
|
export function defineEvalConfig(config) {
|
|
224
258
|
var _a;
|
|
225
|
-
return Object.assign(Object.assign({}, (config.timeoutMs === undefined ? {} : { timeoutMs: config.timeoutMs })), { maxConcurrency: normalizeEvalMaxConcurrency(config.maxConcurrency), maxPlaygroundRuns: normalizeEvalMaxPlaygroundRuns((_a = config.maxPlaygroundRuns) !== null && _a !== void 0 ? _a : EVAL_DEFAULT_MAX_PLAYGROUND_RUNS) });
|
|
259
|
+
return Object.assign(Object.assign({}, (config.timeoutMs === undefined ? {} : { timeoutMs: config.timeoutMs })), { maxConcurrency: normalizeEvalMaxConcurrency(config.maxConcurrency), maxPlaygroundRuns: normalizeEvalMaxPlaygroundRuns((_a = config.maxPlaygroundRuns) !== null && _a !== void 0 ? _a : EVAL_DEFAULT_MAX_PLAYGROUND_RUNS), judge: config.judge, reporters: config.reporters });
|
|
226
260
|
}
|
|
227
261
|
/**
|
|
228
262
|
* Validate and enforce the hard concurrency ceiling.
|
package/dist/index.d.ts
CHANGED
|
@@ -35,7 +35,7 @@ export type { StorageABTable, StorageConfig, StorageContext, StorageDefinition,
|
|
|
35
35
|
export { defineStorage, STORAGE_KEY_ROOT, storageKeys, } from "./storage.js";
|
|
36
36
|
export { defineTool } from "./tools.js";
|
|
37
37
|
export type * from "./types.js";
|
|
38
|
-
export { DEFAULT_MODEL, DIRECT_TOOL_CALL_CHANNEL_ID, isCursorAccountTransport, isPeerConnectionTransport, isSymbolicConnectionTransport, MAX_RUNNING_TURNS_LIMIT, } from "./types.js";
|
|
38
|
+
export { cursorAccountServersHasMixedWildcard, DEFAULT_MODEL, DIRECT_TOOL_CALL_CHANNEL_ID, isCursorAccountTransport, isPeerConnectionTransport, isSymbolicConnectionTransport, MAX_RUNNING_TURNS_LIMIT, normalizeCursorAccountServers, } from "./types.js";
|
|
39
39
|
/**
|
|
40
40
|
* Runtime config for an agent — authored at `agent/agent.ts` (the root
|
|
41
41
|
* agent) or `agent/subagents/<id>/agent.ts` (a subagent).
|
package/dist/index.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAKH,OAAO,EACL,KAAK,uBAAuB,EAE7B,MAAM,yBAAyB,CAAC;AAEjC,OAAO,KAAK,EACV,WAAW,EACX,qBAAqB,EACrB,YAAY,EACZ,gBAAgB,EAChB,kBAAkB,EAClB,sBAAsB,EACtB,YAAY,EACb,MAAM,YAAY,CAAC;AAEpB,OAAO,EAAE,QAAQ,EAAE,MAAM,SAAS,CAAC;AAEnC,OAAO,EACL,QAAQ,EACR,UAAU,EACV,MAAM,EACN,aAAa,EACb,GAAG,EACH,WAAW,EACX,QAAQ,EACR,cAAc,EACd,iBAAiB,EACjB,YAAY,EACZ,KAAK,EACL,IAAI,EACJ,GAAG,GACJ,MAAM,eAAe,CAAC;AACvB,OAAO,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAC;AACpD,OAAO,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACxC,YAAY,EAAE,uBAAuB,EAAE,MAAM,yBAAyB,CAAC;AACvE,OAAO,EACL,iBAAiB,EACjB,mBAAmB,EACnB,iBAAiB,EACjB,gBAAgB,EAChB,iBAAiB,EACjB,WAAW,EACX,kBAAkB,GACnB,MAAM,yBAAyB,CAAC;AACjC,YAAY,EACV,4BAA4B,EAC5B,6BAA6B,EAC7B,sBAAsB,EACtB,uBAAuB,EACvB,6BAA6B,GAC9B,MAAM,uCAAuC,CAAC;AAC/C,OAAO,EACL,2BAA2B,EAC3B,uBAAuB,GACxB,MAAM,uCAAuC,CAAC;AAC/C,OAAO,EACL,cAAc,EACd,iBAAiB,EACjB,oBAAoB,GACrB,MAAM,4BAA4B,CAAC;AACpC,OAAO,EACL,uBAAuB,EACvB,oBAAoB,GACrB,MAAM,+BAA+B,CAAC;AACvC,YAAY,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AACvD,OAAO,EACL,gBAAgB,EAChB,cAAc,EACd,4BAA4B,EAC5B,mBAAmB,EACnB,mBAAmB,EACnB,gBAAgB,GACjB,MAAM,8BAA8B,CAAC;AACtC,YAAY,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AAC7C,OAAO,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC;AACrC,YAAY,EACV,kBAAkB,EAClB,mBAAmB,EACnB,mBAAmB,EACnB,kBAAkB,EAClB,eAAe,EACf,YAAY,EACZ,kBAAkB,EAClB,oBAAoB,EACpB,cAAc,EACd,cAAc,EACd,oBAAoB,GACrB,MAAM,gBAAgB,CAAC;AACxB,OAAO,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AAChD,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAC1C,YAAY,EACV,cAAc,EACd,aAAa,EACb,cAAc,EACd,iBAAiB,EACjB,YAAY,EACZ,iBAAiB,EACjB,aAAa,EACb,oBAAoB,GACrB,MAAM,cAAc,CAAC;AACtB,OAAO,EACL,aAAa,EACb,gBAAgB,EAChB,WAAW,GACZ,MAAM,cAAc,CAAC;AACtB,OAAO,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACxC,mBAAmB,YAAY,CAAC;AAChC,OAAO,EACL,aAAa,EACb,2BAA2B,EAC3B,wBAAwB,EACxB,yBAAyB,EACzB,6BAA6B,EAC7B,uBAAuB,
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAKH,OAAO,EACL,KAAK,uBAAuB,EAE7B,MAAM,yBAAyB,CAAC;AAEjC,OAAO,KAAK,EACV,WAAW,EACX,qBAAqB,EACrB,YAAY,EACZ,gBAAgB,EAChB,kBAAkB,EAClB,sBAAsB,EACtB,YAAY,EACb,MAAM,YAAY,CAAC;AAEpB,OAAO,EAAE,QAAQ,EAAE,MAAM,SAAS,CAAC;AAEnC,OAAO,EACL,QAAQ,EACR,UAAU,EACV,MAAM,EACN,aAAa,EACb,GAAG,EACH,WAAW,EACX,QAAQ,EACR,cAAc,EACd,iBAAiB,EACjB,YAAY,EACZ,KAAK,EACL,IAAI,EACJ,GAAG,GACJ,MAAM,eAAe,CAAC;AACvB,OAAO,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAC;AACpD,OAAO,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACxC,YAAY,EAAE,uBAAuB,EAAE,MAAM,yBAAyB,CAAC;AACvE,OAAO,EACL,iBAAiB,EACjB,mBAAmB,EACnB,iBAAiB,EACjB,gBAAgB,EAChB,iBAAiB,EACjB,WAAW,EACX,kBAAkB,GACnB,MAAM,yBAAyB,CAAC;AACjC,YAAY,EACV,4BAA4B,EAC5B,6BAA6B,EAC7B,sBAAsB,EACtB,uBAAuB,EACvB,6BAA6B,GAC9B,MAAM,uCAAuC,CAAC;AAC/C,OAAO,EACL,2BAA2B,EAC3B,uBAAuB,GACxB,MAAM,uCAAuC,CAAC;AAC/C,OAAO,EACL,cAAc,EACd,iBAAiB,EACjB,oBAAoB,GACrB,MAAM,4BAA4B,CAAC;AACpC,OAAO,EACL,uBAAuB,EACvB,oBAAoB,GACrB,MAAM,+BAA+B,CAAC;AACvC,YAAY,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AACvD,OAAO,EACL,gBAAgB,EAChB,cAAc,EACd,4BAA4B,EAC5B,mBAAmB,EACnB,mBAAmB,EACnB,gBAAgB,GACjB,MAAM,8BAA8B,CAAC;AACtC,YAAY,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AAC7C,OAAO,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC;AACrC,YAAY,EACV,kBAAkB,EAClB,mBAAmB,EACnB,mBAAmB,EACnB,kBAAkB,EAClB,eAAe,EACf,YAAY,EACZ,kBAAkB,EAClB,oBAAoB,EACpB,cAAc,EACd,cAAc,EACd,oBAAoB,GACrB,MAAM,gBAAgB,CAAC;AACxB,OAAO,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AAChD,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAC1C,YAAY,EACV,cAAc,EACd,aAAa,EACb,cAAc,EACd,iBAAiB,EACjB,YAAY,EACZ,iBAAiB,EACjB,aAAa,EACb,oBAAoB,GACrB,MAAM,cAAc,CAAC;AACtB,OAAO,EACL,aAAa,EACb,gBAAgB,EAChB,WAAW,GACZ,MAAM,cAAc,CAAC;AACtB,OAAO,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACxC,mBAAmB,YAAY,CAAC;AAChC,OAAO,EACL,oCAAoC,EACpC,aAAa,EACb,2BAA2B,EAC3B,wBAAwB,EACxB,yBAAyB,EACzB,6BAA6B,EAC7B,uBAAuB,EACvB,6BAA6B,GAC9B,MAAM,YAAY,CAAC;AAEpB;;;GAGG;AACH,wBAAgB,WAAW,CAAC,MAAM,EAAE,WAAW,GAAG,qBAAqB,CAEtE;AAED;;;GAGG;AACH,wBAAgB,kBAAkB,CAChC,MAAM,EAAE,kBAAkB,GACzB,sBAAsB,CAExB;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAsB,KAAK,CACzB,YAAY,EAAE,MAAM,GAAG,YAAY,EACnC,OAAO,GAAE,YAAY,GAAG;IACtB,SAAS,CAAC,EAAE,uBAAuB,CAAC;IACpC;;;;OAIG;IACH,IAAI,CAAC,EAAE,QAAQ,GAAG,OAAO,CAAC;CACtB,GACL,OAAO,CAAC,gBAAgB,CAAC,CAqB3B"}
|
package/dist/index.js
CHANGED
|
@@ -52,7 +52,7 @@ export { defineSchedule } from "./schedules.js";
|
|
|
52
52
|
export { defineSkill } from "./skills.js";
|
|
53
53
|
export { defineStorage, STORAGE_KEY_ROOT, storageKeys, } from "./storage.js";
|
|
54
54
|
export { defineTool } from "./tools.js";
|
|
55
|
-
export { DEFAULT_MODEL, DIRECT_TOOL_CALL_CHANNEL_ID, isCursorAccountTransport, isPeerConnectionTransport, isSymbolicConnectionTransport, MAX_RUNNING_TURNS_LIMIT, } from "./types.js";
|
|
55
|
+
export { cursorAccountServersHasMixedWildcard, DEFAULT_MODEL, DIRECT_TOOL_CALL_CHANNEL_ID, isCursorAccountTransport, isPeerConnectionTransport, isSymbolicConnectionTransport, MAX_RUNNING_TURNS_LIMIT, normalizeCursorAccountServers, } from "./types.js";
|
|
56
56
|
/**
|
|
57
57
|
* Runtime config for an agent — authored at `agent/agent.ts` (the root
|
|
58
58
|
* agent) or `agent/subagents/<id>/agent.ts` (a subagent).
|
|
@@ -6,6 +6,7 @@
|
|
|
6
6
|
* filesystem-first evals under evals/ (*.eval.ts). Live A/B metrics
|
|
7
7
|
* (`defineAB` under `agent/ab`) plug into serve — there is no `ab` CLI.
|
|
8
8
|
*/
|
|
9
|
+
import type { EvalAssertionResult } from "../evals.js";
|
|
9
10
|
export interface AxCliOptions {
|
|
10
11
|
cwd?: string;
|
|
11
12
|
dir: string;
|
|
@@ -36,6 +37,19 @@ export interface AxCliOptions {
|
|
|
36
37
|
tags?: string[];
|
|
37
38
|
/** Eval ids / path prefixes. */
|
|
38
39
|
evalIds?: string[];
|
|
40
|
+
/** `eval`: a soft threshold miss (`scored`) also fails the exit code. */
|
|
41
|
+
strict?: boolean;
|
|
42
|
+
/** `eval`: write JUnit XML to this path. */
|
|
43
|
+
junit?: string;
|
|
44
|
+
/**
|
|
45
|
+
* `eval`: run-artifact directory. Defaults to a timestamped directory under
|
|
46
|
+
* `<dir>/.agent-serve/evals/`; `false` disables artifacts entirely.
|
|
47
|
+
*/
|
|
48
|
+
artifacts?: string | false;
|
|
49
|
+
/** `eval`: skip reporters declared in `evals.config.ts` and eval files. */
|
|
50
|
+
skipReport?: boolean;
|
|
51
|
+
/** `eval`: override `evals.config.ts` `maxConcurrency`. */
|
|
52
|
+
maxConcurrency?: number;
|
|
39
53
|
/** Scaffold target directory for `init`. */
|
|
40
54
|
initTarget?: string;
|
|
41
55
|
/** Tool name for `call`. */
|
|
@@ -145,6 +159,19 @@ export declare function cmdEvalCancel(runId: string | undefined, options: AxCliO
|
|
|
145
159
|
* the batch is still running (for poll loops after `--no-wait`).
|
|
146
160
|
*/
|
|
147
161
|
export declare function cmdEvalStatus(runId: string | undefined, options: AxCliOptions): Promise<number>;
|
|
162
|
+
/**
|
|
163
|
+
* Where a local run writes its artifacts.
|
|
164
|
+
*
|
|
165
|
+
* Defaults to a timestamped directory under the project's gitignored
|
|
166
|
+
* `.agent-serve/evals/`, so the full story of a failure is always on disk
|
|
167
|
+
* without the eval author opting in. `--no-artifacts` turns it off.
|
|
168
|
+
*/
|
|
169
|
+
export declare function resolveEvalArtifactsDir(projectRoot: string, options: Pick<AxCliOptions, "artifacts">): string | undefined;
|
|
170
|
+
/**
|
|
171
|
+
* One assertion line. Soft assertions read `~` rather than `✗` when they miss,
|
|
172
|
+
* so a tracked score is visibly different from a failed gate.
|
|
173
|
+
*/
|
|
174
|
+
export declare function formatEvalAssertionLine(assertion: EvalAssertionResult): string;
|
|
148
175
|
/**
|
|
149
176
|
* `logs` — follow deploy progress (`--prod`) then runtime agent logs.
|
|
150
177
|
* Local / `--url` tails `/v1/logs` only.
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"cli-ax.d.ts","sourceRoot":"","sources":["../../src/internal/cli-ax.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;
|
|
1
|
+
{"version":3,"file":"cli-ax.d.ts","sourceRoot":"","sources":["../../src/internal/cli-ax.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAMH,OAAO,KAAK,EACV,mBAAmB,EAIpB,MAAM,aAAa,CAAC;AAwDrB,MAAM,WAAW,YAAY;IAC3B,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,GAAG,EAAE,MAAM,CAAC;IACZ;;;;OAIG;IACH,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,8BAA8B;IAC9B,QAAQ,CAAC,EAAE,MAAM,EAAE,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;;;OAKG;IACH,MAAM,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,KAAK,CAAC;IAC/B,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,gCAAgC;IAChC,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;IACnB,yEAAyE;IACzE,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,4CAA4C;IAC5C,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;;OAGG;IACH,SAAS,CAAC,EAAE,MAAM,GAAG,KAAK,CAAC;IAC3B,2EAA2E;IAC3E,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,2DAA2D;IAC3D,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,4CAA4C;IAC5C,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,4BAA4B;IAC5B,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,6BAA6B;IAC7B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,iEAAiE;IACjE,GAAG,CAAC,EAAE,MAAM,CAAC;IACb;;;;;;OAMG;IACH,IAAI,CAAC,EAAE,OAAO,CAAC;IACf;;;;OAIG;IACH,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,oEAAoE;IACpE,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,6DAA6D;IAC7D,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,yCAAyC;IACzC,SAAS,CAAC,EAAE,OAAO,KAAK,CAAC;IACzB;;;OAGG;IACH,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,6CAA6C;IAC7C,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,4DAA4D;IAC5D,IAAI,CAAC,EAAE,OAAO,CAAC;IACf;;;OAGG;IACH,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB;;;OAGG;IACH,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB;;;OAGG;IACH,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,sEAAsE;IACtE,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB;AAED,MAAM,WAAW,mBAAmB;IAClC,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC,KAAK,CAAC,EAAE,MAAM,OAAO,CAAC,IAAI,CAAC,CAAC;CAC7B;AAsED,wBAAsB,aAAa,CACjC,OAAO,EAAE,YAAY,GACpB,OAAO,CAAC,mBAAmB,CAAC,CAmB9B;AAED,wBAAsB,OAAO,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAoFpE;AAED;;;GAGG;AACH;;;;;;;;GAQG;AACH,wBAAsB,SAAS,CAC7B,SAAS,EAAE,MAAM,GAAG,SAAS,EAC7B,OAAO,EAAE,YAAY,GACpB,OAAO,CAAC,MAAM,CAAC,CAcjB;AAUD,wBAAsB,OAAO,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CA8GpE;AAED,wBAAsB,MAAM,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CA4GnE;AAED;;;GAGG;AACH,wBAAsB,OAAO,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CA0DpE;AAED,wBAAsB,aAAa,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAe1E;AAED;;;;GAIG;AACH,wBAAgB,qBAAqB,CACnC,OAAO,EAAE,IAAI,CAAC,YAAY,EAAE,MAAM,GAAG,KAAK,CAAC,GAC1C,OAAO,GAAG,KAAK,GAAG,YAAY,GAAG;IAAE,KAAK,EAAE,MAAM,CAAA;CAAE,CAepD;AAiDD,wBAAsB,OAAO,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAkMpE;AAED;;;GAGG;AACH,wBAAgB,qBAAqB,CACnC,OAAO,EAAE,IAAI,CAAC,YAAY,EAAE,MAAM,GAAG,KAAK,GAAG,MAAM,GAAG,MAAM,CAAC,GAC5D,MAAM,CAcR;AAsID;;GAEG;AACH,wBAAsB,aAAa,CACjC,KAAK,EAAE,MAAM,GAAG,SAAS,EACzB,OAAO,EAAE,YAAY,GACpB,OAAO,CAAC,MAAM,CAAC,CAwDjB;AAED;;;;;GAKG;AACH,wBAAsB,aAAa,CACjC,KAAK,EAAE,MAAM,GAAG,SAAS,EACzB,OAAO,EAAE,YAAY,GACpB,OAAO,CAAC,MAAM,CAAC,CA8FjB;AAmGD;;;;;;GAMG;AACH,wBAAgB,uBAAuB,CACrC,WAAW,EAAE,MAAM,EACnB,OAAO,EAAE,IAAI,CAAC,YAAY,EAAE,WAAW,CAAC,GACvC,MAAM,GAAG,SAAS,CASpB;AA8HD;;;GAGG;AACH,wBAAgB,uBAAuB,CACrC,SAAS,EAAE,mBAAmB,GAC7B,MAAM,CAgBR;AAED;;;GAGG;AACH,wBAAsB,OAAO,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAgFpE;AAmCD;;GAEG;AACH,wBAAsB,WAAW,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAkCxE;AAED;;;;GAIG;AACH,wBAAsB,UAAU,CAC9B,SAAS,EAAE,MAAM,GAAG,SAAS,EAC7B,OAAO,EAAE,YAAY,GACpB,OAAO,CAAC,MAAM,CAAC,CAkDjB;AAED;;;;;;GAMG;AACH,wBAAsB,aAAa,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAoF1E"}
|
package/dist/internal/cli-ax.js
CHANGED
|
@@ -18,6 +18,7 @@ var __awaiter = (this && this.__awaiter) || function (thisArg, _arguments, P, ge
|
|
|
18
18
|
import { mkdir, mkdtemp, writeFile } from "node:fs/promises";
|
|
19
19
|
import { tmpdir } from "node:os";
|
|
20
20
|
import { join, resolve } from "node:path";
|
|
21
|
+
import { Artifacts, JUnit } from "../evals/reporters.js";
|
|
21
22
|
import { serve } from "../index.js";
|
|
22
23
|
import { lookupSessionContinuation, runChat } from "./chat-client.js";
|
|
23
24
|
import { cmdLogin } from "./cli-cursor.js";
|
|
@@ -616,6 +617,7 @@ export function cmdEval(options) {
|
|
|
616
617
|
// batch lands in playground history and defineStorage evals (when set).
|
|
617
618
|
if (mode === "local") {
|
|
618
619
|
const progress = stream ? createStreamProgress() : undefined;
|
|
620
|
+
const artifactsDir = resolveEvalArtifactsDir(projectRoot, options);
|
|
619
621
|
try {
|
|
620
622
|
const results = yield runDiscoveredEvals({
|
|
621
623
|
projectRoot,
|
|
@@ -623,6 +625,11 @@ export function cmdEval(options) {
|
|
|
623
625
|
filterIds,
|
|
624
626
|
tags,
|
|
625
627
|
timeoutMs: options.timeoutMs,
|
|
628
|
+
maxConcurrency: options.maxConcurrency,
|
|
629
|
+
strict: options.strict === true,
|
|
630
|
+
skipReport: options.skipReport === true,
|
|
631
|
+
reporters: buildCliEvalReporters(options, artifactsDir),
|
|
632
|
+
target: { baseUrl: target.agentUrl, mode: "local" },
|
|
626
633
|
bearerToken: options.bearerToken,
|
|
627
634
|
headers: target.headers,
|
|
628
635
|
verbose: Boolean(options.verbose) || stream,
|
|
@@ -632,7 +639,7 @@ export function cmdEval(options) {
|
|
|
632
639
|
process.stderr.write("No matching evals found.\n");
|
|
633
640
|
return 2;
|
|
634
641
|
}
|
|
635
|
-
return printEvalResults(results, options);
|
|
642
|
+
return printEvalResults(results, options, undefined, artifactsDir);
|
|
636
643
|
}
|
|
637
644
|
finally {
|
|
638
645
|
progress === null || progress === void 0 ? void 0 : progress.close();
|
|
@@ -1031,33 +1038,76 @@ function createRemoteEvalProgress(enabled) {
|
|
|
1031
1038
|
},
|
|
1032
1039
|
};
|
|
1033
1040
|
}
|
|
1034
|
-
|
|
1041
|
+
/**
|
|
1042
|
+
* Where a local run writes its artifacts.
|
|
1043
|
+
*
|
|
1044
|
+
* Defaults to a timestamped directory under the project's gitignored
|
|
1045
|
+
* `.agent-serve/evals/`, so the full story of a failure is always on disk
|
|
1046
|
+
* without the eval author opting in. `--no-artifacts` turns it off.
|
|
1047
|
+
*/
|
|
1048
|
+
export function resolveEvalArtifactsDir(projectRoot, options) {
|
|
1049
|
+
if (options.artifacts === false) {
|
|
1050
|
+
return undefined;
|
|
1051
|
+
}
|
|
1052
|
+
if (typeof options.artifacts === "string" && options.artifacts !== "") {
|
|
1053
|
+
return resolve(options.artifacts);
|
|
1054
|
+
}
|
|
1055
|
+
const stamp = new Date().toISOString().replace(/[:.]/g, "-");
|
|
1056
|
+
return join(projectRoot, ".agent-serve", "evals", stamp);
|
|
1057
|
+
}
|
|
1058
|
+
/** JUnit / artifacts reporters requested via CLI flags. */
|
|
1059
|
+
function buildCliEvalReporters(options, artifactsDir) {
|
|
1060
|
+
const reporters = [];
|
|
1061
|
+
if (artifactsDir !== undefined) {
|
|
1062
|
+
reporters.push(Artifacts({ dir: artifactsDir }));
|
|
1063
|
+
}
|
|
1064
|
+
if (options.junit !== undefined && options.junit !== "") {
|
|
1065
|
+
reporters.push(JUnit({ filePath: options.junit }));
|
|
1066
|
+
}
|
|
1067
|
+
return reporters;
|
|
1068
|
+
}
|
|
1069
|
+
function evalVerdictOf(result) {
|
|
1070
|
+
var _a;
|
|
1071
|
+
return (_a = result.verdict) !== null && _a !== void 0 ? _a : (result.ok ? "passed" : "failed");
|
|
1072
|
+
}
|
|
1073
|
+
const EVAL_VERDICT_LABEL = {
|
|
1074
|
+
passed: "PASS",
|
|
1075
|
+
failed: "FAIL",
|
|
1076
|
+
scored: "SCORED",
|
|
1077
|
+
skipped: "SKIP",
|
|
1078
|
+
};
|
|
1079
|
+
function printEvalResults(results, options, batch, artifactsDir) {
|
|
1035
1080
|
var _a, _b;
|
|
1036
|
-
const
|
|
1037
|
-
const
|
|
1081
|
+
const strict = options.strict === true;
|
|
1082
|
+
const counts = { passed: 0, failed: 0, scored: 0, skipped: 0 };
|
|
1083
|
+
for (const result of results) {
|
|
1084
|
+
counts[evalVerdictOf(result)]++;
|
|
1085
|
+
}
|
|
1038
1086
|
const batchFailed = (batch === null || batch === void 0 ? void 0 : batch.status) === "failed";
|
|
1039
|
-
const ok = failed === 0 && !batchFailed;
|
|
1087
|
+
const ok = counts.failed === 0 && !(strict && counts.scored > 0) && !batchFailed;
|
|
1040
1088
|
if ((batch === null || batch === void 0 ? void 0 : batch.error) !== undefined && batch.error !== "" && !options.json) {
|
|
1041
1089
|
// Always surface harness-level failures (not only --verbose).
|
|
1042
1090
|
process.stderr.write(`eval run error: ${batch.error}\n`);
|
|
1043
1091
|
}
|
|
1044
1092
|
if (options.json) {
|
|
1045
|
-
console.log(JSON.stringify(Object.assign(Object.assign(Object.assign({ ok,
|
|
1046
|
-
|
|
1047
|
-
failed,
|
|
1048
|
-
results }, ((batch === null || batch === void 0 ? void 0 : batch.runId) === undefined ? {} : { runId: batch.runId })), ((batch === null || batch === void 0 ? void 0 : batch.status) === undefined ? {} : { status: batch.status })), ((batch === null || batch === void 0 ? void 0 : batch.error) === undefined || batch.error === ""
|
|
1093
|
+
console.log(JSON.stringify(Object.assign(Object.assign(Object.assign(Object.assign({ ok, passed: counts.passed, failed: counts.failed, scored: counts.scored, skipped: counts.skipped, strict,
|
|
1094
|
+
results }, (artifactsDir === undefined ? {} : { artifactsDir })), ((batch === null || batch === void 0 ? void 0 : batch.runId) === undefined ? {} : { runId: batch.runId })), ((batch === null || batch === void 0 ? void 0 : batch.status) === undefined ? {} : { status: batch.status })), ((batch === null || batch === void 0 ? void 0 : batch.error) === undefined || batch.error === ""
|
|
1049
1095
|
? {}
|
|
1050
1096
|
: { error: batch.error })), null, 2));
|
|
1051
1097
|
}
|
|
1052
1098
|
else {
|
|
1053
1099
|
for (const r of results) {
|
|
1054
|
-
|
|
1100
|
+
const verdict = evalVerdictOf(r);
|
|
1101
|
+
process.stdout.write(`${EVAL_VERDICT_LABEL[verdict]} ${r.id} (${(r.durationMs / 1000).toFixed(1)}s)`);
|
|
1055
1102
|
if (r.description !== undefined) {
|
|
1056
1103
|
process.stdout.write(` — ${r.description}`);
|
|
1057
1104
|
}
|
|
1058
1105
|
process.stdout.write("\n");
|
|
1106
|
+
if (r.skipReason !== undefined) {
|
|
1107
|
+
process.stdout.write(` skipped: ${r.skipReason}\n`);
|
|
1108
|
+
}
|
|
1059
1109
|
for (const a of r.assertions) {
|
|
1060
|
-
process.stdout.write(` ${
|
|
1110
|
+
process.stdout.write(` ${formatEvalAssertionLine(a)}\n`);
|
|
1061
1111
|
}
|
|
1062
1112
|
for (const [name, value] of Object.entries((_a = r.metrics) !== null && _a !== void 0 ? _a : {})) {
|
|
1063
1113
|
process.stdout.write(` metric ${name}=${String(value)}\n`);
|
|
@@ -1072,10 +1122,38 @@ function printEvalResults(results, options, batch) {
|
|
|
1072
1122
|
process.stdout.write(` reply: ${r.finalText.slice(0, 200).replace(/\s+/g, " ")}\n`);
|
|
1073
1123
|
}
|
|
1074
1124
|
}
|
|
1075
|
-
|
|
1125
|
+
const extra = [];
|
|
1126
|
+
if (counts.scored > 0) {
|
|
1127
|
+
extra.push(`${counts.scored} scored${strict ? " (fatal: --strict)" : ""}`);
|
|
1128
|
+
}
|
|
1129
|
+
if (counts.skipped > 0) {
|
|
1130
|
+
extra.push(`${counts.skipped} skipped`);
|
|
1131
|
+
}
|
|
1132
|
+
process.stdout.write(`\n${counts.passed} passed, ${counts.failed} failed${extra.length === 0 ? "" : `, ${extra.join(", ")}`}, ${results.length} total\n`);
|
|
1133
|
+
if (artifactsDir !== undefined) {
|
|
1134
|
+
process.stdout.write(`artifacts: ${artifactsDir}\n`);
|
|
1135
|
+
}
|
|
1076
1136
|
}
|
|
1077
1137
|
return ok ? 0 : 1;
|
|
1078
1138
|
}
|
|
1139
|
+
/**
|
|
1140
|
+
* One assertion line. Soft assertions read `~` rather than `✗` when they miss,
|
|
1141
|
+
* so a tracked score is visibly different from a failed gate.
|
|
1142
|
+
*/
|
|
1143
|
+
export function formatEvalAssertionLine(assertion) {
|
|
1144
|
+
var _a;
|
|
1145
|
+
const soft = ((_a = assertion.severity) !== null && _a !== void 0 ? _a : "gate") === "soft";
|
|
1146
|
+
const mark = assertion.passed ? "✓" : soft ? "~" : "✗";
|
|
1147
|
+
const parts = [];
|
|
1148
|
+
if (assertion.score !== undefined) {
|
|
1149
|
+
parts.push(`score=${assertion.score.toFixed(3)}${assertion.threshold === undefined ? "" : ` >= ${assertion.threshold}`}`);
|
|
1150
|
+
}
|
|
1151
|
+
if (assertion.detail !== undefined) {
|
|
1152
|
+
parts.push(assertion.detail);
|
|
1153
|
+
}
|
|
1154
|
+
const suffix = parts.length === 0 ? "" : ` (${parts.join(", ")})`;
|
|
1155
|
+
return `${mark} ${assertion.name}${soft ? " [soft]" : ""}${suffix}`;
|
|
1156
|
+
}
|
|
1079
1157
|
/**
|
|
1080
1158
|
* `logs` — follow deploy progress (`--prod`) then runtime agent logs.
|
|
1081
1159
|
* Local / `--url` tails `/v1/logs` only.
|