@cursor/july 0.1.26 → 0.1.27
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/ab.d.ts +6 -0
- package/dist/ab.d.ts.map +1 -1
- package/dist/ab.js +1 -0
- package/dist/artifacts.d.ts +34 -0
- package/dist/artifacts.d.ts.map +1 -0
- package/dist/artifacts.js +57 -0
- package/dist/bin/agent-serve.js +31 -9
- package/dist/channels/github/github-channel.d.ts +1 -1
- package/dist/channels/github/github-channel.d.ts.map +1 -1
- package/dist/channels/github/github-channel.js +14 -1
- package/dist/channels/github/types.d.ts +6 -1
- package/dist/channels/github/types.d.ts.map +1 -1
- package/dist/channels/slack/api.d.ts +2 -1
- package/dist/channels/slack/api.d.ts.map +1 -1
- package/dist/channels/slack/api.js +1 -0
- package/dist/channels/slack/cursor-account.d.ts +2 -1
- package/dist/channels/slack/cursor-account.d.ts.map +1 -1
- package/dist/channels/slack/cursor-account.js +1 -1
- package/dist/channels/slack/dispatch.d.ts +4 -1
- package/dist/channels/slack/dispatch.d.ts.map +1 -1
- package/dist/channels/slack/dispatch.js +3 -3
- package/dist/channels/slack/interactive.d.ts +2 -1
- package/dist/channels/slack/interactive.d.ts.map +1 -1
- package/dist/channels/slack/interactive.js +1 -0
- package/dist/channels/slack/slack-channel.d.ts.map +1 -1
- package/dist/channels/slack/slack-channel.js +4 -3
- package/dist/channels/slack/types.d.ts +6 -1
- package/dist/channels/slack/types.d.ts.map +1 -1
- package/dist/docs/404.html +1 -1
- package/dist/docs/ab.html +2 -2
- package/dist/docs/assets/{app.DYcC9FY-.js → app.BICLoJYF.js} +1 -1
- package/dist/docs/assets/chunks/@localSearchIndexroot.CQJytDwU.js +1 -0
- package/dist/docs/assets/chunks/{VPLocalSearchBox.o4N_knTV.js → VPLocalSearchBox.wN4Ko9dT.js} +1 -1
- package/dist/docs/assets/chunks/{theme.DQ-njyo0.js → theme.dnnTE-Co.js} +2 -2
- package/dist/docs/building-with-agents.html +2 -2
- package/dist/docs/concepts.html +2 -2
- package/dist/docs/deployment.html +2 -2
- package/dist/docs/evals.html +2 -2
- package/dist/docs/example-agents/approval-buddy.html +2 -2
- package/dist/docs/example-agents/benny.html +2 -2
- package/dist/docs/example-agents/bugbot.html +2 -2
- package/dist/docs/example-agents/codebase-wiki.html +2 -2
- package/dist/docs/example-agents/codeowners-review.html +2 -2
- package/dist/docs/example-agents/concierge.html +2 -2
- package/dist/docs/example-agents/fsd.html +2 -2
- package/dist/docs/example-agents/index.html +2 -2
- package/dist/docs/example-agents/knowledge-base.html +2 -2
- package/dist/docs/example-agents/oncall.html +2 -2
- package/dist/docs/example-agents/security-reviewer.html +2 -2
- package/dist/docs/example-agents/slack-agent.html +2 -2
- package/dist/docs/example-agents/weather-agent.html +2 -2
- package/dist/docs/guides/agent-to-agent.html +2 -2
- package/dist/docs/guides/cloud-runtime.html +2 -2
- package/dist/docs/guides/github.html +2 -2
- package/dist/docs/guides/human-in-the-loop.html +2 -2
- package/dist/docs/guides/mcp-oauth.html +2 -2
- package/dist/docs/guides/slack.html +2 -2
- package/dist/docs/guides/webhooks.html +2 -2
- package/dist/docs/hillclimbing.html +2 -2
- package/dist/docs/index.html +2 -2
- package/dist/docs/quickstart.html +2 -2
- package/dist/docs/reference/agent-config.html +2 -2
- package/dist/docs/reference/channels.html +2 -2
- package/dist/docs/reference/cli.html +2 -2
- package/dist/docs/reference/connections.html +2 -2
- package/dist/docs/reference/hooks.html +2 -2
- package/dist/docs/reference/http-api.html +2 -2
- package/dist/docs/reference/instructions.html +2 -2
- package/dist/docs/reference/playground.html +2 -2
- package/dist/docs/reference/project-layout.html +2 -2
- package/dist/docs/reference/prompt.html +2 -2
- package/dist/docs/reference/schedules.html +2 -2
- package/dist/docs/reference/sessions.html +2 -2
- package/dist/docs/reference/skills.html +2 -2
- package/dist/docs/reference/subagents.html +2 -2
- package/dist/docs/reference/tools.html +2 -2
- package/dist/docs/scaffolding-agents.html +2 -2
- package/dist/docs/storage.html +2 -2
- package/dist/docs/troubleshooting.html +2 -2
- package/dist/evals/assertions.d.ts +10 -1
- package/dist/evals/assertions.d.ts.map +1 -1
- package/dist/evals/assertions.js +12 -0
- package/dist/evals.d.ts +1 -1
- package/dist/evals.d.ts.map +1 -1
- package/dist/index.d.ts +1 -0
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +1 -0
- package/dist/internal/ab-fold.d.ts.map +1 -1
- package/dist/internal/ab-fold.js +4 -0
- package/dist/internal/ab-snapshot.d.ts.map +1 -1
- package/dist/internal/ab-snapshot.js +1 -0
- package/dist/internal/artifacts-store.d.ts +45 -0
- package/dist/internal/artifacts-store.d.ts.map +1 -0
- package/dist/internal/artifacts-store.js +250 -0
- package/dist/internal/builtin-tools/artifacts.d.ts +10 -0
- package/dist/internal/builtin-tools/artifacts.d.ts.map +1 -0
- package/dist/internal/builtin-tools/artifacts.js +73 -0
- package/dist/internal/builtin-tools/index.d.ts +8 -1
- package/dist/internal/builtin-tools/index.d.ts.map +1 -1
- package/dist/internal/builtin-tools/index.js +15 -0
- package/dist/internal/cli-ax.d.ts +11 -0
- package/dist/internal/cli-ax.d.ts.map +1 -1
- package/dist/internal/cli-ax.js +80 -0
- package/dist/internal/cli-slack.d.ts.map +1 -1
- package/dist/internal/cli-slack.js +14 -6
- package/dist/internal/discovery.d.ts.map +1 -1
- package/dist/internal/discovery.js +83 -2
- package/dist/internal/distribution.d.ts.map +1 -1
- package/dist/internal/distribution.js +1 -0
- package/dist/internal/model-pricing.d.ts +49 -0
- package/dist/internal/model-pricing.d.ts.map +1 -0
- package/dist/internal/model-pricing.js +377 -0
- package/dist/internal/reminder-runner.d.ts.map +1 -1
- package/dist/internal/reminder-runner.js +1 -0
- package/dist/internal/schedule-runner.d.ts.map +1 -1
- package/dist/internal/schedule-runner.js +1 -0
- package/dist/internal/sdk-runner.d.ts.map +1 -1
- package/dist/internal/sdk-runner.js +9 -1
- package/dist/internal/server.d.ts.map +1 -1
- package/dist/internal/server.js +39 -0
- package/dist/internal/session-cost.d.ts +49 -0
- package/dist/internal/session-cost.d.ts.map +1 -0
- package/dist/internal/session-cost.js +175 -0
- package/dist/internal/session-engine.d.ts +35 -1
- package/dist/internal/session-engine.d.ts.map +1 -1
- package/dist/internal/session-engine.js +128 -7
- package/dist/internal/storage-coordinator.d.ts +15 -0
- package/dist/internal/storage-coordinator.d.ts.map +1 -1
- package/dist/internal/storage-coordinator.js +52 -0
- package/dist/internal/trajectory.d.ts +9 -1
- package/dist/internal/trajectory.d.ts.map +1 -1
- package/dist/internal/trajectory.js +32 -5
- package/dist/playground/assets/index-BDCeW4uc.js +86 -0
- package/dist/playground/assets/index-Bul1WfrB.css +1 -0
- package/dist/playground/index.html +2 -2
- package/dist/storage.d.ts +4 -0
- package/dist/storage.d.ts.map +1 -1
- package/dist/storage.js +4 -0
- package/dist/types.d.ts +164 -1
- package/dist/types.d.ts.map +1 -1
- package/package.json +26 -26
- package/src/ab.ts +7 -0
- package/src/bin/agent-serve.ts +30 -2
- package/src/internal/ab-fold.ts +4 -0
- package/src/internal/ab-snapshot.ts +1 -0
- package/src/internal/cli-ax.ts +95 -0
- package/src/internal/model-pricing.ts +426 -0
- package/src/internal/sdk-runner.ts +10 -2
- package/src/internal/session-cost.ts +231 -0
- package/src/internal/session-engine.ts +28 -0
- package/src/internal/trajectory.ts +45 -3
- package/src/types.ts +24 -1
- package/dist/channels/github/instrument.d.ts +0 -20
- package/dist/channels/github/instrument.d.ts.map +0 -1
- package/dist/docs/assets/chunks/@localSearchIndexroot.BQTzJjR_.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/json-dir-store.js +0 -100
- package/dist/playground/assets/index-CjOQ4hN9.css +0 -1
- package/dist/playground/assets/index-DqXdAFGa.js +0 -85
- package/src/bin/agent-serve.version.test.ts +0 -64
- package/src/channels/github/api.test.ts +0 -64
- package/src/channels/github/auth.test.ts +0 -105
- package/src/channels/github/cursor-account.test.ts +0 -204
- package/src/channels/github/forward.test.ts +0 -457
- package/src/channels/github/github.test.ts +0 -985
- package/src/channels/github/replay.test.ts +0 -179
- package/src/channels/slack/api.post-message.test.ts +0 -148
- package/src/channels/slack/approvals.test.ts +0 -328
- package/src/channels/slack/block-actions.test.ts +0 -469
- package/src/channels/slack/bot-mentions.test.ts +0 -267
- package/src/channels/slack/channel-watch.test.ts +0 -363
- package/src/channels/slack/cursor-account.test.ts +0 -301
- package/src/channels/slack/defaults.final-post.test.ts +0 -182
- package/src/channels/slack/dispatch.test.ts +0 -875
- package/src/channels/slack/eval-directive.test.ts +0 -273
- package/src/channels/slack/message-body.test.ts +0 -54
- package/src/channels/slack/nudge-store.test.ts +0 -143
- package/src/channels/slack/slack.test.ts +0 -391
- package/src/channels/slack/stop.test.ts +0 -23
- package/src/channels/slack/thread-context.test.ts +0 -202
- package/src/evals/assertions.test.ts +0 -617
- package/src/evals/expect.test.ts +0 -144
- package/src/evals/judge.test.ts +0 -181
- package/src/evals/loaders.test.ts +0 -132
- package/src/evals/matchers.test.ts +0 -95
- package/src/evals/reporters.test.ts +0 -303
- package/src/evals/run-facts.test.ts +0 -259
- package/src/internal/ab-snapshot.test.ts +0 -325
- package/src/internal/approval-gate.test.ts +0 -49
- package/src/internal/approvals.integration.test.ts +0 -383
- package/src/internal/artifacts-store.test.ts +0 -283
- package/src/internal/authored-loaders.test.ts +0 -31
- package/src/internal/builtin-tools/artifacts.test.ts +0 -247
- package/src/internal/builtin-tools/reminders.test.ts +0 -201
- package/src/internal/channel-route-schema.test.ts +0 -294
- package/src/internal/chat-attach.test.ts +0 -262
- package/src/internal/cli-deploy.test.ts +0 -1991
- package/src/internal/cli-docs.test.ts +0 -161
- package/src/internal/cli-mcp.test.ts +0 -789
- package/src/internal/cli-skills.test.ts +0 -133
- package/src/internal/cli-slack.test.ts +0 -1654
- package/src/internal/cloud-merge.test.ts +0 -74
- package/src/internal/cron.test.ts +0 -22
- package/src/internal/cursor/account-mcp.test.ts +0 -807
- package/src/internal/cursor/backend-client.test.ts +0 -591
- package/src/internal/cursor/credentials.test.ts +0 -351
- package/src/internal/cursor/github-credentials.test.ts +0 -136
- package/src/internal/cursor-account-mcp-auth.test.ts +0 -310
- package/src/internal/cursor-account.integration.test.ts +0 -441
- package/src/internal/cursor-event-relay.test.ts +0 -746
- package/src/internal/cursor-github-credentials.integration.test.ts +0 -271
- package/src/internal/cursor-slack-relay.test.ts +0 -525
- package/src/internal/deploy-source.test.ts +0 -111
- package/src/internal/discovery.artifact-tool.test.ts +0 -136
- package/src/internal/discovery.artifacts.test.ts +0 -119
- package/src/internal/discovery.builtin-tools.test.ts +0 -94
- package/src/internal/discovery.concurrency.test.ts +0 -60
- package/src/internal/discovery.cursor-account.test.ts +0 -133
- package/src/internal/discovery.cwd.test.ts +0 -83
- package/src/internal/discovery.hosting.test.ts +0 -80
- package/src/internal/discovery.identity.test.ts +0 -44
- package/src/internal/docs-site.test.ts +0 -66
- package/src/internal/duration.test.ts +0 -29
- package/src/internal/eval-judge-model.test.ts +0 -187
- package/src/internal/eval-run-store.cancel.test.ts +0 -142
- package/src/internal/eval-run-store.storage.test.ts +0 -211
- package/src/internal/eval-runner.http.test.ts +0 -403
- package/src/internal/eval-runner.run.test.ts +0 -928
- package/src/internal/evals-client.test.ts +0 -307
- package/src/internal/event-mapper.test.ts +0 -243
- package/src/internal/github-fanout.test.ts +0 -213
- package/src/internal/handleAgentServeTrigger.test.ts +0 -185
- package/src/internal/host-kv.test.ts +0 -82
- package/src/internal/host-platforms.test.ts +0 -126
- package/src/internal/http-channel.test.ts +0 -402
- package/src/internal/init-project.test.ts +0 -270
- package/src/internal/install-cursor-skills.test.ts +0 -262
- package/src/internal/local-env.test.ts +0 -120
- package/src/internal/log-ring.test.ts +0 -31
- package/src/internal/logs-client.test.ts +0 -350
- package/src/internal/mcp-endpoint.test.ts +0 -436
- package/src/internal/mcp-host.test.ts +0 -298
- package/src/internal/mcp-oauth.test.ts +0 -148
- package/src/internal/net.test.ts +0 -17
- package/src/internal/peer-connections.test.ts +0 -128
- package/src/internal/peer-mcp.integration.test.ts +0 -289
- package/src/internal/playground/toolchain.test.ts +0 -53
- package/src/internal/playground-cli.test.ts +0 -187
- package/src/internal/playground-proxy.test.ts +0 -376
- package/src/internal/prompt-context.integration.test.ts +0 -232
- package/src/internal/prompt-context.test.ts +0 -127
- package/src/internal/reminder-runner.test.ts +0 -432
- package/src/internal/reminder-store.test.ts +0 -53
- package/src/internal/request-headers.test.ts +0 -27
- package/src/internal/resolve-prod-target.test.ts +0 -787
- package/src/internal/resolved-connections.test.ts +0 -295
- package/src/internal/router.test.ts +0 -57
- package/src/internal/sdk-runner.test.ts +0 -290
- package/src/internal/server.artifacts.test.ts +0 -334
- package/src/internal/session-engine.artifacts.test.ts +0 -293
- package/src/internal/session-engine.coalesce.test.ts +0 -169
- package/src/internal/session-engine.concurrency.test.ts +0 -250
- package/src/internal/session-engine.host-oauth-mcp.test.ts +0 -110
- package/src/internal/session-engine.interrupt.test.ts +0 -577
- package/src/internal/session-engine.storage.test.ts +0 -547
- package/src/internal/session-urls.test.ts +0 -28
- package/src/internal/sessions-client.test.ts +0 -518
- package/src/internal/storage-coordinator.test.ts +0 -517
- package/src/internal/tool-call.test.ts +0 -458
- package/src/internal/tool-result.test.ts +0 -52
- package/src/internal/trajectory.approvals.test.ts +0 -83
- package/src/internal/trajectory.subagents.test.ts +0 -198
- package/src/internal/turn-governor.test.ts +0 -137
- package/src/internal/update-check.test.ts +0 -485
- package/src/internal/workspace.test.ts +0 -207
- package/src/storage-backends/cursor-hosted.test.ts +0 -121
|
@@ -1,273 +0,0 @@
|
|
|
1
|
-
import { describe, expect, it } from "vitest";
|
|
2
|
-
import type { EvalRunSnapshot, HostEvalsApi } from "../../evals.js";
|
|
3
|
-
import {
|
|
4
|
-
EvalRunConflictError,
|
|
5
|
-
EvalRunEmptyError,
|
|
6
|
-
} from "../../internal/eval-run-store.js";
|
|
7
|
-
import {
|
|
8
|
-
formatSlackEvalHelp,
|
|
9
|
-
formatSlackEvalList,
|
|
10
|
-
formatSlackEvalRunSummary,
|
|
11
|
-
handleSlackEvalDirective,
|
|
12
|
-
normalizeSlackEvalDirectiveBody,
|
|
13
|
-
parseSlackEvalDirective,
|
|
14
|
-
} from "./eval-directive.js";
|
|
15
|
-
import type { SlackBinding } from "./types.js";
|
|
16
|
-
|
|
17
|
-
describe("normalizeSlackEvalDirectiveBody", () => {
|
|
18
|
-
it("strips mentions and cursorAccount agent-name prefix", () => {
|
|
19
|
-
expect(normalizeSlackEvalDirectiveBody("<@Ubot> eval list")).toBe(
|
|
20
|
-
"eval list"
|
|
21
|
-
);
|
|
22
|
-
expect(normalizeSlackEvalDirectiveBody("@Ubot eval status")).toBe(
|
|
23
|
-
"eval status"
|
|
24
|
-
);
|
|
25
|
-
expect(
|
|
26
|
-
normalizeSlackEvalDirectiveBody(
|
|
27
|
-
"OncallBuddy eval triage/open-incidents",
|
|
28
|
-
{
|
|
29
|
-
agentName: "OncallBuddy",
|
|
30
|
-
}
|
|
31
|
-
)
|
|
32
|
-
).toBe("eval triage/open-incidents");
|
|
33
|
-
expect(
|
|
34
|
-
normalizeSlackEvalDirectiveBody("benny eval --tag smoke", {
|
|
35
|
-
agentName: "Benny",
|
|
36
|
-
})
|
|
37
|
-
).toBe("eval --tag smoke");
|
|
38
|
-
// Do not strip arbitrary leading words ("please eval …" is not a directive).
|
|
39
|
-
expect(normalizeSlackEvalDirectiveBody("please eval this")).toBe(
|
|
40
|
-
"please eval this"
|
|
41
|
-
);
|
|
42
|
-
});
|
|
43
|
-
});
|
|
44
|
-
|
|
45
|
-
describe("parseSlackEvalDirective", () => {
|
|
46
|
-
it("parses run / list / status / help and filters", () => {
|
|
47
|
-
expect(parseSlackEvalDirective("<@U> eval")).toEqual({ kind: "run" });
|
|
48
|
-
expect(parseSlackEvalDirective("eval list")).toEqual({ kind: "list" });
|
|
49
|
-
expect(parseSlackEvalDirective("EVAL status")).toEqual({ kind: "status" });
|
|
50
|
-
expect(parseSlackEvalDirective("eval cancel evalrun_abc")).toEqual({
|
|
51
|
-
kind: "cancel",
|
|
52
|
-
runId: "evalrun_abc",
|
|
53
|
-
});
|
|
54
|
-
expect(parseSlackEvalDirective("eval cancel")).toEqual({ kind: "help" });
|
|
55
|
-
expect(parseSlackEvalDirective("eval help")).toEqual({ kind: "help" });
|
|
56
|
-
expect(parseSlackEvalDirective("eval smoke triage/open-incidents")).toEqual(
|
|
57
|
-
{
|
|
58
|
-
kind: "run",
|
|
59
|
-
filterIds: ["smoke", "triage/open-incidents"],
|
|
60
|
-
}
|
|
61
|
-
);
|
|
62
|
-
expect(parseSlackEvalDirective("eval --tag oncall -t smoke")).toEqual({
|
|
63
|
-
kind: "run",
|
|
64
|
-
tags: ["oncall", "smoke"],
|
|
65
|
-
});
|
|
66
|
-
expect(parseSlackEvalDirective("eval --tag=booking smoke")).toEqual({
|
|
67
|
-
kind: "run",
|
|
68
|
-
filterIds: ["smoke"],
|
|
69
|
-
tags: ["booking"],
|
|
70
|
-
});
|
|
71
|
-
});
|
|
72
|
-
|
|
73
|
-
it("returns null for non-directives and help for unknown flags", () => {
|
|
74
|
-
expect(parseSlackEvalDirective("please eval this")).toBeNull();
|
|
75
|
-
expect(parseSlackEvalDirective("stop")).toBeNull();
|
|
76
|
-
expect(parseSlackEvalDirective("")).toBeNull();
|
|
77
|
-
expect(parseSlackEvalDirective("eval --unknown")).toEqual({ kind: "help" });
|
|
78
|
-
expect(parseSlackEvalDirective("eval --tag")).toEqual({ kind: "help" });
|
|
79
|
-
});
|
|
80
|
-
});
|
|
81
|
-
|
|
82
|
-
describe("formatSlackEval*", () => {
|
|
83
|
-
it("formats help, list, and run summaries", () => {
|
|
84
|
-
expect(formatSlackEvalHelp()).toContain("eval list");
|
|
85
|
-
expect(formatSlackEvalList([])).toContain("No evals found");
|
|
86
|
-
expect(
|
|
87
|
-
formatSlackEvalList([
|
|
88
|
-
{
|
|
89
|
-
id: "smoke",
|
|
90
|
-
description: "Smoke check",
|
|
91
|
-
tags: ["ci"],
|
|
92
|
-
},
|
|
93
|
-
])
|
|
94
|
-
).toContain("`smoke`");
|
|
95
|
-
const run: EvalRunSnapshot = {
|
|
96
|
-
runId: "evalrun_abc",
|
|
97
|
-
status: "completed",
|
|
98
|
-
startedAt: "2026-07-22T00:00:00.000Z",
|
|
99
|
-
finishedAt: "2026-07-22T00:00:05.000Z",
|
|
100
|
-
summary: { passed: 1, failed: 0, total: 1, done: 1 },
|
|
101
|
-
cases: [
|
|
102
|
-
{
|
|
103
|
-
id: "smoke",
|
|
104
|
-
fileId: "smoke",
|
|
105
|
-
status: "done",
|
|
106
|
-
ok: true,
|
|
107
|
-
durationMs: 5000,
|
|
108
|
-
},
|
|
109
|
-
],
|
|
110
|
-
config: { maxPlaygroundRuns: 20, durableRuns: false },
|
|
111
|
-
};
|
|
112
|
-
const text = formatSlackEvalRunSummary(run);
|
|
113
|
-
expect(text).toContain("evalrun_abc");
|
|
114
|
-
expect(text).toContain("`PASS`");
|
|
115
|
-
expect(text).toContain("1 passed");
|
|
116
|
-
});
|
|
117
|
-
});
|
|
118
|
-
|
|
119
|
-
function fakeBinding(): {
|
|
120
|
-
binding: SlackBinding;
|
|
121
|
-
posts: string[];
|
|
122
|
-
typing: string[];
|
|
123
|
-
} {
|
|
124
|
-
const posts: string[] = [];
|
|
125
|
-
const typing: string[] = [];
|
|
126
|
-
const binding = {
|
|
127
|
-
thread: {
|
|
128
|
-
post: async (text: string) => {
|
|
129
|
-
posts.push(text);
|
|
130
|
-
return { ts: "1.0" };
|
|
131
|
-
},
|
|
132
|
-
startTyping: async (status: string) => {
|
|
133
|
-
typing.push(status);
|
|
134
|
-
},
|
|
135
|
-
setTitle: async () => {},
|
|
136
|
-
setSuggestedPrompts: async () => {},
|
|
137
|
-
},
|
|
138
|
-
} as unknown as SlackBinding;
|
|
139
|
-
return { binding, posts, typing };
|
|
140
|
-
}
|
|
141
|
-
|
|
142
|
-
function fakeEvals(overrides: Partial<HostEvalsApi> = {}): HostEvalsApi {
|
|
143
|
-
return {
|
|
144
|
-
list: async () => ({
|
|
145
|
-
evals: [{ id: "smoke", fileId: "smoke", tags: ["ci"] }],
|
|
146
|
-
config: { maxPlaygroundRuns: 20, durableRuns: false },
|
|
147
|
-
}),
|
|
148
|
-
get: async () => undefined,
|
|
149
|
-
listRunsWithActive: async () => ({ runs: [] }),
|
|
150
|
-
start: async () => {
|
|
151
|
-
throw new Error("start not stubbed");
|
|
152
|
-
},
|
|
153
|
-
cancel: async () => {
|
|
154
|
-
throw new Error("cancel not stubbed");
|
|
155
|
-
},
|
|
156
|
-
...overrides,
|
|
157
|
-
};
|
|
158
|
-
}
|
|
159
|
-
|
|
160
|
-
describe("handleSlackEvalDirective", () => {
|
|
161
|
-
it("lists evals and reports empty status", async () => {
|
|
162
|
-
const { binding, posts } = fakeBinding();
|
|
163
|
-
await handleSlackEvalDirective({
|
|
164
|
-
directive: { kind: "list" },
|
|
165
|
-
evals: fakeEvals(),
|
|
166
|
-
binding,
|
|
167
|
-
});
|
|
168
|
-
expect(posts[0]).toContain("`smoke`");
|
|
169
|
-
|
|
170
|
-
await handleSlackEvalDirective({
|
|
171
|
-
directive: { kind: "status" },
|
|
172
|
-
evals: fakeEvals(),
|
|
173
|
-
binding,
|
|
174
|
-
});
|
|
175
|
-
expect(posts[1]).toContain("No eval runs yet");
|
|
176
|
-
});
|
|
177
|
-
|
|
178
|
-
it("starts a run, polls, and posts the summary", async () => {
|
|
179
|
-
const running: EvalRunSnapshot = {
|
|
180
|
-
runId: "evalrun_1",
|
|
181
|
-
status: "running",
|
|
182
|
-
startedAt: "2026-07-22T00:00:00.000Z",
|
|
183
|
-
summary: { passed: 0, failed: 0, total: 1, done: 0 },
|
|
184
|
-
cases: [{ id: "smoke", fileId: "smoke", status: "running" }],
|
|
185
|
-
config: { maxPlaygroundRuns: 20, durableRuns: false },
|
|
186
|
-
};
|
|
187
|
-
const done: EvalRunSnapshot = {
|
|
188
|
-
...running,
|
|
189
|
-
status: "completed",
|
|
190
|
-
finishedAt: "2026-07-22T00:00:02.000Z",
|
|
191
|
-
summary: { passed: 1, failed: 0, total: 1, done: 1 },
|
|
192
|
-
cases: [
|
|
193
|
-
{
|
|
194
|
-
id: "smoke",
|
|
195
|
-
fileId: "smoke",
|
|
196
|
-
status: "done",
|
|
197
|
-
ok: true,
|
|
198
|
-
durationMs: 1200,
|
|
199
|
-
},
|
|
200
|
-
],
|
|
201
|
-
};
|
|
202
|
-
let gets = 0;
|
|
203
|
-
const { binding, posts } = fakeBinding();
|
|
204
|
-
await handleSlackEvalDirective({
|
|
205
|
-
directive: { kind: "run", filterIds: ["smoke"] },
|
|
206
|
-
evals: fakeEvals({
|
|
207
|
-
start: async (input) => {
|
|
208
|
-
expect(input?.filterIds).toEqual(["smoke"]);
|
|
209
|
-
return running;
|
|
210
|
-
},
|
|
211
|
-
get: async () => {
|
|
212
|
-
gets += 1;
|
|
213
|
-
return gets === 1 ? running : done;
|
|
214
|
-
},
|
|
215
|
-
}),
|
|
216
|
-
binding,
|
|
217
|
-
pollMs: 1,
|
|
218
|
-
});
|
|
219
|
-
expect(posts.some((p) => p.includes("Started Eval ID"))).toBe(true);
|
|
220
|
-
expect(posts.some((p) => p.includes("`PASS`") && p.includes("smoke"))).toBe(
|
|
221
|
-
true
|
|
222
|
-
);
|
|
223
|
-
});
|
|
224
|
-
|
|
225
|
-
it("surfaces conflict and empty-match errors", async () => {
|
|
226
|
-
const { binding, posts } = fakeBinding();
|
|
227
|
-
await handleSlackEvalDirective({
|
|
228
|
-
directive: { kind: "run" },
|
|
229
|
-
evals: fakeEvals({
|
|
230
|
-
start: async () => {
|
|
231
|
-
throw new EvalRunConflictError("evalrun_busy");
|
|
232
|
-
},
|
|
233
|
-
get: async () => undefined,
|
|
234
|
-
}),
|
|
235
|
-
binding,
|
|
236
|
-
});
|
|
237
|
-
expect(posts[0]).toContain("already in progress");
|
|
238
|
-
|
|
239
|
-
await handleSlackEvalDirective({
|
|
240
|
-
directive: { kind: "run", tags: ["missing"] },
|
|
241
|
-
evals: fakeEvals({
|
|
242
|
-
start: async () => {
|
|
243
|
-
throw new EvalRunEmptyError();
|
|
244
|
-
},
|
|
245
|
-
}),
|
|
246
|
-
binding,
|
|
247
|
-
});
|
|
248
|
-
expect(posts[1]).toContain("No matching evals");
|
|
249
|
-
});
|
|
250
|
-
|
|
251
|
-
it("posts help text", async () => {
|
|
252
|
-
const { binding, posts } = fakeBinding();
|
|
253
|
-
await handleSlackEvalDirective({
|
|
254
|
-
directive: { kind: "help" },
|
|
255
|
-
evals: fakeEvals(),
|
|
256
|
-
binding,
|
|
257
|
-
});
|
|
258
|
-
expect(posts[0]).toContain("eval --tag");
|
|
259
|
-
});
|
|
260
|
-
});
|
|
261
|
-
|
|
262
|
-
describe("parseSlackEvalDirective cursorAccount shapes", () => {
|
|
263
|
-
it("accepts agent-name-prefixed eval commands", () => {
|
|
264
|
-
expect(
|
|
265
|
-
parseSlackEvalDirective("OncallBuddy eval triage/open-incidents", {
|
|
266
|
-
agentName: "OncallBuddy",
|
|
267
|
-
})
|
|
268
|
-
).toEqual({
|
|
269
|
-
kind: "run",
|
|
270
|
-
filterIds: ["triage/open-incidents"],
|
|
271
|
-
});
|
|
272
|
-
});
|
|
273
|
-
});
|
|
@@ -1,54 +0,0 @@
|
|
|
1
|
-
import { describe, expect, it } from "vitest";
|
|
2
|
-
import { slackApiMessageBody } from "./message-body.js";
|
|
3
|
-
|
|
4
|
-
describe("slackApiMessageBody", () => {
|
|
5
|
-
it("prefers text and dedupes attachment fallback", () => {
|
|
6
|
-
const body = slackApiMessageBody({
|
|
7
|
-
text: "Alert: rate limit",
|
|
8
|
-
attachments: [
|
|
9
|
-
{
|
|
10
|
-
title: "Bugbot rate limit",
|
|
11
|
-
text: "Alert: rate limit",
|
|
12
|
-
fallback: "Alert: rate limit",
|
|
13
|
-
title_link: "https://app.datadoghq.com/monitors/1",
|
|
14
|
-
},
|
|
15
|
-
],
|
|
16
|
-
});
|
|
17
|
-
expect(body).toContain("Alert: rate limit");
|
|
18
|
-
expect(body).toContain("Bugbot rate limit");
|
|
19
|
-
expect(body).toContain("https://app.datadoghq.com/monitors/1");
|
|
20
|
-
expect(body.split("Alert: rate limit").length - 1).toBe(1);
|
|
21
|
-
});
|
|
22
|
-
|
|
23
|
-
it("reads Block Kit section text", () => {
|
|
24
|
-
const body = slackApiMessageBody({
|
|
25
|
-
blocks: [
|
|
26
|
-
{
|
|
27
|
-
type: "section",
|
|
28
|
-
text: { type: "mrkdwn", text: "*Monitor* fired" },
|
|
29
|
-
},
|
|
30
|
-
],
|
|
31
|
-
});
|
|
32
|
-
expect(body).toBe("*Monitor* fired");
|
|
33
|
-
});
|
|
34
|
-
|
|
35
|
-
it("recovers a Datadog notification posted with empty top-level text", () => {
|
|
36
|
-
const body = slackApiMessageBody({
|
|
37
|
-
text: "",
|
|
38
|
-
attachments: [
|
|
39
|
-
{
|
|
40
|
-
title:
|
|
41
|
-
"Triggered: [service-monitor prod:temporal-bugbot] High cache error rate (client=main, region=us-east-1)",
|
|
42
|
-
title_link: "https://us5.datadoghq.com/monitors/19610160",
|
|
43
|
-
text: "Service temporal-bugbot has a high cache error rate.",
|
|
44
|
-
},
|
|
45
|
-
],
|
|
46
|
-
});
|
|
47
|
-
expect(body).toContain("temporal-bugbot");
|
|
48
|
-
expect(body).toContain("https://us5.datadoghq.com/monitors/19610160");
|
|
49
|
-
});
|
|
50
|
-
|
|
51
|
-
it("returns an empty string when nothing readable is present", () => {
|
|
52
|
-
expect(slackApiMessageBody({ text: " ", attachments: [{}] })).toBe("");
|
|
53
|
-
});
|
|
54
|
-
});
|
|
@@ -1,143 +0,0 @@
|
|
|
1
|
-
import { mkdtemp, rm } from "node:fs/promises";
|
|
2
|
-
import { tmpdir } from "node:os";
|
|
3
|
-
import { join } from "node:path";
|
|
4
|
-
import { afterEach, describe, expect, it } from "vitest";
|
|
5
|
-
import {
|
|
6
|
-
formatSlackFollowupNudgeBlock,
|
|
7
|
-
SlackAskDedupeStore,
|
|
8
|
-
SlackNudgeStore,
|
|
9
|
-
} from "./nudge-store.js";
|
|
10
|
-
|
|
11
|
-
const cleanups: Array<() => Promise<void>> = [];
|
|
12
|
-
|
|
13
|
-
afterEach(async () => {
|
|
14
|
-
while (cleanups.length > 0) {
|
|
15
|
-
await cleanups.pop()?.();
|
|
16
|
-
}
|
|
17
|
-
});
|
|
18
|
-
|
|
19
|
-
async function tempRoot(): Promise<string> {
|
|
20
|
-
const root = await mkdtemp(join(tmpdir(), "slack-nudge-"));
|
|
21
|
-
cleanups.push(() => rm(root, { recursive: true, force: true }));
|
|
22
|
-
return root;
|
|
23
|
-
}
|
|
24
|
-
|
|
25
|
-
describe("SlackNudgeStore", () => {
|
|
26
|
-
it("appends, drains exactly once, and skips delivered source ts", async () => {
|
|
27
|
-
const store = new SlackNudgeStore(await tempRoot());
|
|
28
|
-
const key = "C1:1.0";
|
|
29
|
-
expect(
|
|
30
|
-
await store.append({
|
|
31
|
-
continuationKey: key,
|
|
32
|
-
sourceTs: "2.0",
|
|
33
|
-
text: "any leads?",
|
|
34
|
-
author: "<@U1>",
|
|
35
|
-
})
|
|
36
|
-
).toBe(true);
|
|
37
|
-
expect(
|
|
38
|
-
await store.append({
|
|
39
|
-
continuationKey: key,
|
|
40
|
-
sourceTs: "2.0",
|
|
41
|
-
text: "any leads?",
|
|
42
|
-
author: "<@U1>",
|
|
43
|
-
})
|
|
44
|
-
).toBe(false);
|
|
45
|
-
|
|
46
|
-
const first = await store.drainForInject(key);
|
|
47
|
-
expect(first).toHaveLength(1);
|
|
48
|
-
expect(first[0]?.sourceTs).toBe("2.0");
|
|
49
|
-
expect(await store.isDelivered(key, "2.0")).toBe(true);
|
|
50
|
-
expect(await store.hasUndelivered(key)).toBe(false);
|
|
51
|
-
|
|
52
|
-
const second = await store.drainForInject(key);
|
|
53
|
-
expect(second).toHaveLength(0);
|
|
54
|
-
});
|
|
55
|
-
|
|
56
|
-
it("content-dedupes pending nudges", async () => {
|
|
57
|
-
const store = new SlackNudgeStore(await tempRoot());
|
|
58
|
-
const key = "C1:1.0";
|
|
59
|
-
expect(
|
|
60
|
-
await store.append({
|
|
61
|
-
continuationKey: key,
|
|
62
|
-
sourceTs: "2.0",
|
|
63
|
-
text: "same body",
|
|
64
|
-
})
|
|
65
|
-
).toBe(true);
|
|
66
|
-
expect(
|
|
67
|
-
await store.append({
|
|
68
|
-
continuationKey: key,
|
|
69
|
-
sourceTs: "3.0",
|
|
70
|
-
text: "same body",
|
|
71
|
-
})
|
|
72
|
-
).toBe(false);
|
|
73
|
-
});
|
|
74
|
-
|
|
75
|
-
it("clear removes pending and delivered", async () => {
|
|
76
|
-
const store = new SlackNudgeStore(await tempRoot());
|
|
77
|
-
const key = "C1:1.0";
|
|
78
|
-
await store.append({
|
|
79
|
-
continuationKey: key,
|
|
80
|
-
sourceTs: "2.0",
|
|
81
|
-
text: "stop me",
|
|
82
|
-
});
|
|
83
|
-
await store.drainForInject(key);
|
|
84
|
-
await store.clear(key);
|
|
85
|
-
expect(await store.isDelivered(key, "2.0")).toBe(false);
|
|
86
|
-
expect(await store.hasUndelivered(key)).toBe(false);
|
|
87
|
-
});
|
|
88
|
-
|
|
89
|
-
it("maps path-traversal keys to _empty so clear cannot escape root", async () => {
|
|
90
|
-
const root = await tempRoot();
|
|
91
|
-
const store = new SlackNudgeStore(root);
|
|
92
|
-
expect(
|
|
93
|
-
await store.append({
|
|
94
|
-
continuationKey: "..",
|
|
95
|
-
sourceTs: "2.0",
|
|
96
|
-
text: "escape attempt",
|
|
97
|
-
})
|
|
98
|
-
).toBe(true);
|
|
99
|
-
const { readdir } = await import("node:fs/promises");
|
|
100
|
-
expect(await readdir(root)).toContain("_empty");
|
|
101
|
-
await store.clear("..");
|
|
102
|
-
expect(await readdir(root)).not.toContain("..");
|
|
103
|
-
// Parent of the store root must still be readable after clear("..").
|
|
104
|
-
await expect(readdir(join(root, ".."))).resolves.toBeDefined();
|
|
105
|
-
});
|
|
106
|
-
});
|
|
107
|
-
|
|
108
|
-
describe("formatSlackFollowupNudgeBlock", () => {
|
|
109
|
-
it("formats structured follow-up block", () => {
|
|
110
|
-
const block = formatSlackFollowupNudgeBlock([
|
|
111
|
-
{
|
|
112
|
-
id: "n1",
|
|
113
|
-
continuationKey: "C:1",
|
|
114
|
-
sourceTs: "2.0",
|
|
115
|
-
text: "any leads?",
|
|
116
|
-
author: "<@U1>",
|
|
117
|
-
createdAt: "2026-01-01T00:00:00.000Z",
|
|
118
|
-
contentHash: "abc",
|
|
119
|
-
},
|
|
120
|
-
]);
|
|
121
|
-
expect(block).toContain("<slack_followup>");
|
|
122
|
-
expect(block).toContain("<@U1> @ 2.0:");
|
|
123
|
-
expect(block).toContain("any leads?");
|
|
124
|
-
});
|
|
125
|
-
});
|
|
126
|
-
|
|
127
|
-
describe("SlackAskDedupeStore", () => {
|
|
128
|
-
it("claims each channel:ts once", async () => {
|
|
129
|
-
const store = new SlackAskDedupeStore(await tempRoot());
|
|
130
|
-
expect(await store.tryClaim("C1", "10.0")).toBe(true);
|
|
131
|
-
expect(await store.tryClaim("C1", "10.0")).toBe(false);
|
|
132
|
-
expect(await store.wasClaimed("C1", "10.0")).toBe(true);
|
|
133
|
-
expect(await store.tryClaim("C1", "11.0")).toBe(true);
|
|
134
|
-
});
|
|
135
|
-
|
|
136
|
-
it("release allows a later claim of the same channel:ts", async () => {
|
|
137
|
-
const store = new SlackAskDedupeStore(await tempRoot());
|
|
138
|
-
expect(await store.tryClaim("C1", "10.0")).toBe(true);
|
|
139
|
-
await store.release("C1", "10.0");
|
|
140
|
-
expect(await store.wasClaimed("C1", "10.0")).toBe(false);
|
|
141
|
-
expect(await store.tryClaim("C1", "10.0")).toBe(true);
|
|
142
|
-
});
|
|
143
|
-
});
|