reeval 0.0.1 → 0.3.0
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/THIRD_PARTY_NOTICES.md +4 -4
- package/assets/package.json +1 -1
- package/assets/pnpm-lock.yaml +84 -33
- package/dist/agent-BYy0Tc6P.js +334 -0
- package/dist/agent-BYy0Tc6P.js.map +1 -0
- package/dist/{agent-DVHkx8aT.js → agent-DcQAvomp.js} +4 -4
- package/dist/{agent-DVHkx8aT.js.map → agent-DcQAvomp.js.map} +1 -1
- package/dist/analysis-CxSCt1YY.js.map +1 -1
- package/dist/assemble-BFsyMhb-.js +2 -0
- package/dist/{assemble-B-nmz8oH.js → assemble-h8s0hzTQ.js} +23 -14
- package/dist/assemble-h8s0hzTQ.js.map +1 -0
- package/dist/{cli--CQ3Iy9q.js → cli-DemDdg8J.js} +165 -54
- package/dist/cli-DemDdg8J.js.map +1 -0
- package/dist/{cli-spec-DKAu09Gi.js → cli-spec-BxLPgNN-.js} +8 -2
- package/dist/cli-spec-BxLPgNN-.js.map +1 -0
- package/dist/cli.js +2 -2
- package/dist/{command-D-jnQWVG.js → command-CxwnRrTs.js} +11 -11
- package/dist/{command-D-jnQWVG.js.map → command-CxwnRrTs.js.map} +1 -1
- package/dist/command-DhgZPhsY.js +2 -0
- package/dist/{command-DQAvbG7d.js → command-PEbIzJj3.js} +12 -12
- package/dist/{command-DQAvbG7d.js.map → command-PEbIzJj3.js.map} +1 -1
- package/dist/{command-BSJzHMtP.js → command-pHfluVCF.js} +11 -9
- package/dist/command-pHfluVCF.js.map +1 -0
- package/dist/{command-CEm8PZ1N.js → command-y91-aNhu.js} +169 -280
- package/dist/command-y91-aNhu.js.map +1 -0
- package/dist/{commands-A4udfyKZ.js → commands-De-SxIgY.js} +1 -1
- package/dist/{commands-CF8mOfiu.js → commands-DhCuEa6z.js} +6 -6
- package/dist/{commands-CF8mOfiu.js.map → commands-DhCuEa6z.js.map} +1 -1
- package/dist/{config-Cuj0kVYj.js → config-CqSLZkwx.js} +39 -14
- package/dist/{config-Cuj0kVYj.js.map → config-CqSLZkwx.js.map} +1 -1
- package/dist/{db-DxDzMhtB.js → db-DA4LBwi_.js} +4 -4
- package/dist/{db-DxDzMhtB.js.map → db-DA4LBwi_.js.map} +1 -1
- package/dist/db-DDfq7Tx_.js +2 -0
- package/dist/{dist-BZXYHN2X.js → dist-BSoys6pg.js} +2 -2
- package/dist/{dist-BZXYHN2X.js.map → dist-BSoys6pg.js.map} +1 -1
- package/dist/{docker-DQDZWXF9.js → docker-DMRwn7yh.js} +2 -2
- package/dist/{docker-DQDZWXF9.js.map → docker-DMRwn7yh.js.map} +1 -1
- package/dist/{docker-oSbCExHL.js → docker-DwG9f76x.js} +4 -4
- package/dist/{docker-oSbCExHL.js.map → docker-DwG9f76x.js.map} +1 -1
- package/dist/{docker-DnWAVdEm.js → docker-jf52ARP7.js} +1 -1
- package/dist/{doctor-CxlH1pDM.js → doctor-DhvVbNKv.js} +23 -18
- package/dist/doctor-DhvVbNKv.js.map +1 -0
- package/dist/{egress-73SQfKdx.js → egress-D9vJCATi.js} +35 -19
- package/dist/egress-D9vJCATi.js.map +1 -0
- package/dist/{environment-Mz7wvrtD.js → environment-ByCF9s3n.js} +6 -16
- package/dist/{environment-Mz7wvrtD.js.map → environment-ByCF9s3n.js.map} +1 -1
- package/dist/files-B16cQ6XX.js +2 -0
- package/dist/files-BC5Yqir-.js +2 -0
- package/dist/{files-bF2fPuT9.js → files-CfsIRN7n.js} +40 -7
- package/dist/files-CfsIRN7n.js.map +1 -0
- package/dist/{files-D2-qn_Dl.js → files-DHBBIYx_.js} +6 -6
- package/dist/{files-D2-qn_Dl.js.map → files-DHBBIYx_.js.map} +1 -1
- package/dist/{git-D6hMeieS.js → git-oX92_Jh-.js} +2 -2
- package/dist/{git-D6hMeieS.js.map → git-oX92_Jh-.js.map} +1 -1
- package/dist/{harness-ls5aX1UO.js → harness-BbMShUC-.js} +4 -4
- package/dist/{harness-ls5aX1UO.js.map → harness-BbMShUC-.js.map} +1 -1
- package/dist/harness.d.ts +2 -2
- package/dist/harness.js +1 -1
- package/dist/{header-DzFDoMCI.js → header-sPQ9Lq2S.js} +2 -2
- package/dist/{header-DzFDoMCI.js.map → header-sPQ9Lq2S.js.map} +1 -1
- package/dist/{images-Cc1-coFc.js → images-C9Y4ijbG.js} +6 -6
- package/dist/{images-Cc1-coFc.js.map → images-C9Y4ijbG.js.map} +1 -1
- package/dist/index.d.ts +4 -4
- package/dist/index.js +23 -23
- package/dist/index.js.map +1 -1
- package/dist/init-DPJv7OmX.js +2 -0
- package/dist/{init-CFljS8BJ.js → init-x55uF8a4.js} +10 -10
- package/dist/{init-CFljS8BJ.js.map → init-x55uF8a4.js.map} +1 -1
- package/dist/{invocations-rXktgSAn.js → invocations-DSMfs3Dx.js} +3 -3
- package/dist/{invocations-rXktgSAn.js.map → invocations-DSMfs3Dx.js.map} +1 -1
- package/dist/{jiti-BOLXpKyG.js → jiti-DWsGEgxD.js} +4 -4
- package/dist/{jiti-BOLXpKyG.js.map → jiti-DWsGEgxD.js.map} +1 -1
- package/dist/judge.d.ts +7 -6
- package/dist/judge.js +8 -3
- package/dist/judge.js.map +1 -1
- package/dist/{live-Bs3LReMV.js → live-DZkCtmNP.js} +2 -2
- package/dist/{live-Bs3LReMV.js.map → live-DZkCtmNP.js.map} +1 -1
- package/dist/{matchers-DyLvm7EG.js → matchers-D51lM8dO.js} +4 -4
- package/dist/{matchers-DyLvm7EG.js.map → matchers-D51lM8dO.js.map} +1 -1
- package/dist/model-CNoc8t2Y.js +2 -0
- package/dist/model-DQQQ3q_n.js +2 -0
- package/dist/model-GFM8iOzM.js +228 -0
- package/dist/model-GFM8iOzM.js.map +1 -0
- package/dist/{model-CDL98xue.js → model-ORtLm2UB.js} +41 -3
- package/dist/{model-CDL98xue.js.map → model-ORtLm2UB.js.map} +1 -1
- package/dist/{node-kG64XxdE.js → node-OXPxBD2N.js} +3 -3
- package/dist/{node-kG64XxdE.js.map → node-OXPxBD2N.js.map} +1 -1
- package/dist/{opencode-providers-DZuD_oYt.js → opencode-providers-DJVocFOU.js} +31 -3
- package/dist/opencode-providers-DJVocFOU.js.map +1 -0
- package/dist/{options-CAQhu--C.js → options-CIutEeuy.js} +86 -46
- package/dist/options-CIutEeuy.js.map +1 -0
- package/dist/{plan-gwk6dJ8c.js → plan-DcHTdiJN.js} +21 -5
- package/dist/plan-DcHTdiJN.js.map +1 -0
- package/dist/{preflight-BtubXKQ8.js → preflight-BQyUr_hW.js} +1 -1
- package/dist/{preflight-BypZMNBG.js → preflight-vISQIiF1.js} +3 -3
- package/dist/{preflight-BypZMNBG.js.map → preflight-vISQIiF1.js.map} +1 -1
- package/dist/{prepare-D6Uzcubb.js → prepare-BTeAiW9Q.js} +5 -5
- package/dist/{prepare-D6Uzcubb.js.map → prepare-BTeAiW9Q.js.map} +1 -1
- package/dist/{prepare-Cwv0bHs_.js → prepare-Ca0Nxd9b.js} +2 -2
- package/dist/{provenance-Dfb8YY5Y.js → provenance-D4vo7RF6.js} +21 -3
- package/dist/provenance-D4vo7RF6.js.map +1 -0
- package/dist/{providers-2p3ylMHI.js → providers-Bjgky1u8.js} +3 -3
- package/dist/{providers-2p3ylMHI.js.map → providers-Bjgky1u8.js.map} +1 -1
- package/dist/{providers-BUOxitcT.js → providers-DfVMglqP.js} +1 -1
- package/dist/query-CWkZnV62.js +2 -0
- package/dist/{query-CFATHZmV.js → query-D2KaStTs.js} +2 -2
- package/dist/{query-CFATHZmV.js.map → query-D2KaStTs.js.map} +1 -1
- package/dist/questions-CzXdVVsh.js +325 -0
- package/dist/questions-CzXdVVsh.js.map +1 -0
- package/dist/{record-DQ8wzyco.js → record-CbuQH4l1.js} +7 -3
- package/dist/record-CbuQH4l1.js.map +1 -0
- package/dist/{results-CeXb8Wde.js → results-BI-zTqQy.js} +4 -4
- package/dist/{results-CeXb8Wde.js.map → results-BI-zTqQy.js.map} +1 -1
- package/dist/{root-CFDjpWuc.js → root-Zeokqt3-.js} +2 -2
- package/dist/{root-CFDjpWuc.js.map → root-Zeokqt3-.js.map} +1 -1
- package/dist/{sandbox-By2gkdkC.js → sandbox-Dd-4Bx6B.js} +3 -3
- package/dist/{sandbox-By2gkdkC.js.map → sandbox-Dd-4Bx6B.js.map} +1 -1
- package/dist/sandbox.d.ts +2 -2
- package/dist/sandbox.js +1 -1
- package/dist/server-BjbgFX28.js +2 -0
- package/dist/{server-Bi61uiY0.js → server-C7t8WYCR.js} +11 -9
- package/dist/{server-Bi61uiY0.js.map → server-C7t8WYCR.js.map} +1 -1
- package/dist/source-CLE8vMNa.js +2 -0
- package/dist/{source-Bsfa3LZH.js → source-fNUCWsCW.js} +2 -2
- package/dist/{source-Bsfa3LZH.js.map → source-fNUCWsCW.js.map} +1 -1
- package/dist/{src-DXx_29pq.js → src-B1OT_Zhf.js} +46 -46
- package/dist/{src-DXx_29pq.js.map → src-B1OT_Zhf.js.map} +1 -1
- package/dist/{src-BQyMiWri.js → src-BAiYD4eC.js} +48 -7
- package/dist/{src-BQyMiWri.js.map → src-BAiYD4eC.js.map} +1 -1
- package/dist/{src-D8KYzxBX.js → src-BmBFlEIe.js} +12 -7
- package/dist/{src-D8KYzxBX.js.map → src-BmBFlEIe.js.map} +1 -1
- package/dist/{src-N7E_mfcZ.js → src-C91RndLw.js} +1 -1
- package/dist/{src-C-LHyTSV.js → src-CEQFcAJH.js} +1 -1
- package/dist/src-CNp0srm9.js +3 -0
- package/dist/{src-DwwiVLqv.js → src-D3eW1e7Y.js} +19 -15
- package/dist/{src-DwwiVLqv.js.map → src-D3eW1e7Y.js.map} +1 -1
- package/dist/src-DJsBrqo6.js +4 -0
- package/dist/src-DTlmkpJe.js +3 -0
- package/dist/{src-Cmp-7YrD.js → src-DWk0e7ix.js} +5 -5
- package/dist/{src-Cmp-7YrD.js.map → src-DWk0e7ix.js.map} +1 -1
- package/dist/{src-AEwxd1UP.js → src-DfEAlC9s.js} +4 -4
- package/dist/{src-AEwxd1UP.js.map → src-DfEAlC9s.js.map} +1 -1
- package/dist/src-DgYURuJ6.js +3 -0
- package/dist/src-sSPmuaAc.js +3 -0
- package/dist/state-Bx-pH2ON.js +2 -0
- package/dist/{state-DZ7eqb1F.js → state-g6ij_40j.js} +14 -14
- package/dist/{state-DZ7eqb1F.js.map → state-g6ij_40j.js.map} +1 -1
- package/dist/studio-app/assets/index-B1Pf3LAT.css +2 -0
- package/dist/studio-app/assets/index-DciOL7BI.js +58 -0
- package/dist/studio-app/index.html +2 -2
- package/dist/summary-DjbxgCDi.js +192 -0
- package/dist/summary-DjbxgCDi.js.map +1 -0
- package/dist/{summary-D7Ttky0z.js → summary-gawvEpNu.js} +33 -5
- package/dist/summary-gawvEpNu.js.map +1 -0
- package/dist/tools-B7SoW5kU.js +249 -0
- package/dist/tools-B7SoW5kU.js.map +1 -0
- package/dist/{types-DLec-6Cx.d.ts → types-C5AEPr29.d.ts} +178 -29
- package/dist/view/index.d.ts +63 -1
- package/dist/view/index.js +5 -5
- package/dist/{views-BrHOv6wg.js → views-CHXQfSWQ.js} +4 -4
- package/dist/{views-BrHOv6wg.js.map → views-CHXQfSWQ.js.map} +1 -1
- package/dist/views-DDei2LHa.js +3 -0
- package/dist/{worker-entry-DnYkANPG.js → worker-entry-B6nuDpBX.js} +3 -3
- package/dist/{worker-entry-DnYkANPG.js.map → worker-entry-B6nuDpBX.js.map} +1 -1
- package/dist/{workspace-C2S0Z6gl.js → workspace-DUUfne8b.js} +4 -4
- package/dist/{workspace-C2S0Z6gl.js.map → workspace-DUUfne8b.js.map} +1 -1
- package/dist/{workspace-CyoXG6CY.js → workspace-DXV77ZqW.js} +1 -1
- package/docker/Dockerfile +2 -0
- package/docker/reeval-path.sh +21 -0
- package/package.json +1 -1
- package/dist/assemble-B-nmz8oH.js.map +0 -1
- package/dist/assemble-CmRVDZ2Q.js +0 -2
- package/dist/cli--CQ3Iy9q.js.map +0 -1
- package/dist/cli-spec-DKAu09Gi.js.map +0 -1
- package/dist/command-BSJzHMtP.js.map +0 -1
- package/dist/command-CEm8PZ1N.js.map +0 -1
- package/dist/db-Cjj2ZsfQ.js +0 -2
- package/dist/doctor-CxlH1pDM.js.map +0 -1
- package/dist/egress-73SQfKdx.js.map +0 -1
- package/dist/files--6CEUXm7.js +0 -2
- package/dist/files-Bd25OAQH.js +0 -2
- package/dist/files-bF2fPuT9.js.map +0 -1
- package/dist/init-z2ZPn9Tj.js +0 -2
- package/dist/model-v-0_Ik7S.js +0 -2
- package/dist/opencode-providers-DZuD_oYt.js.map +0 -1
- package/dist/options-CAQhu--C.js.map +0 -1
- package/dist/plan-gwk6dJ8c.js.map +0 -1
- package/dist/provenance-Dfb8YY5Y.js.map +0 -1
- package/dist/query-DR4m6CXf.js +0 -2
- package/dist/record-DQ8wzyco.js.map +0 -1
- package/dist/server-CsSBL8NC.js +0 -2
- package/dist/source-BXzrmof5.js +0 -2
- package/dist/src-55j1zNHH.js +0 -3
- package/dist/src-BDwSuyIo.js +0 -3
- package/dist/src-DEgff9Z7.js +0 -4
- package/dist/src-W9g-y5X5.js +0 -3
- package/dist/src-daS0wKH2.js +0 -3
- package/dist/state-Cyw3QTsN.js +0 -2
- package/dist/studio-app/assets/index-D6XyfFSH.js +0 -58
- package/dist/studio-app/assets/index-xwzWdcE_.css +0 -2
- package/dist/summary-D7Ttky0z.js.map +0 -1
- package/dist/views-CwCWLWqo.js +0 -3
|
@@ -6,8 +6,8 @@
|
|
|
6
6
|
<meta name="referrer" content="no-referrer" />
|
|
7
7
|
<link rel="icon" type="image/svg+xml" href="./favicon.svg" />
|
|
8
8
|
<title>reeval Studio</title>
|
|
9
|
-
<script type="module" crossorigin src="./assets/index-
|
|
10
|
-
<link rel="stylesheet" crossorigin href="./assets/index-
|
|
9
|
+
<script type="module" crossorigin src="./assets/index-DciOL7BI.js"></script>
|
|
10
|
+
<link rel="stylesheet" crossorigin href="./assets/index-B1Pf3LAT.css">
|
|
11
11
|
</head>
|
|
12
12
|
<body>
|
|
13
13
|
<div id="root"></div>
|
|
@@ -0,0 +1,192 @@
|
|
|
1
|
+
import { l as writeCached } from "./files-CfsIRN7n.js";
|
|
2
|
+
import { a as providerModel } from "./model-GFM8iOzM.js";
|
|
3
|
+
import { r as fromRecorded, t as attemptSummary } from "./tools-B7SoW5kU.js";
|
|
4
|
+
import { n as allRuns } from "./commands-DhCuEa6z.js";
|
|
5
|
+
import { a as failedAnalysis, i as defaultBase, o as prepareAnalysis, s as promptOf, t as analyseDigest } from "./command-y91-aNhu.js";
|
|
6
|
+
import { join } from "node:path";
|
|
7
|
+
import { z } from "zod";
|
|
8
|
+
import { generateText, stepCountIs, tool } from "ai";
|
|
9
|
+
/** Items the fake reads of each kind, as a careful model would: a few, not all of them. */
|
|
10
|
+
const FAKE_READS = 3;
|
|
11
|
+
const SUMMARY_TOOLS_PROMPT = `You are about to analyse a reeval run from the DIGEST in the prompt. First, read what the digest leaves out, with the tools:
|
|
12
|
+
- read_evidence: an evidence item dropped from the digest to fit its size cap (the ids in "dropped").
|
|
13
|
+
- open_attempt: one attempt's turns, steps, checks, verdicts and diff, by its id in the attempts table (e.g. r1/skill/control#1).
|
|
14
|
+
|
|
15
|
+
Open failing attempts, and attempts whose facts look odd; at most six items in all, never every attempt. Then reply with one line saying what you read, and stop: you write the analysis itself in the next step.`;
|
|
16
|
+
/** The tools over a prepared digest, with every result counted against `maxChars`. */
|
|
17
|
+
function summaryTools(p, maxChars) {
|
|
18
|
+
let left = maxChars;
|
|
19
|
+
const spend = (value) => {
|
|
20
|
+
const size = JSON.stringify(value).length;
|
|
21
|
+
if (size > left) return { error: "The input budget (analysis.maxInputTokens) is spent: write the analysis from what you have." };
|
|
22
|
+
left -= size;
|
|
23
|
+
return value;
|
|
24
|
+
};
|
|
25
|
+
const runs = [p.headRun, p.baseRun].flatMap((r) => r ? [r] : []);
|
|
26
|
+
return {
|
|
27
|
+
read_evidence: tool({
|
|
28
|
+
description: "An evidence item dropped from the digest to fit its cap: its lines (transcript lines keep their L<n>), what it is and why it was chosen.",
|
|
29
|
+
inputSchema: z.object({ id: z.string().describe("An evidence id from the digest’s dropped") }),
|
|
30
|
+
execute: async ({ id }) => {
|
|
31
|
+
const item = p.dropped.find((e) => e.id === id);
|
|
32
|
+
if (item) {
|
|
33
|
+
const { priority: _priority, ...rest } = item;
|
|
34
|
+
return spend(rest);
|
|
35
|
+
}
|
|
36
|
+
return { error: p.digest.evidence.some((e) => e.id === id) ? `${id} is in the digest already.` : `No dropped evidence ${id}. Dropped: ${p.digest.dropped.join(", ") || "none"}.` };
|
|
37
|
+
}
|
|
38
|
+
}),
|
|
39
|
+
open_attempt: tool({
|
|
40
|
+
description: "One attempt: each turn's prompt, steps (tool calls, edits, errors) and final message, its checks, verdicts and diff.",
|
|
41
|
+
inputSchema: z.object({ id: z.string().describe("An attempt id from the digest’s attempts table") }),
|
|
42
|
+
execute: async ({ id }) => {
|
|
43
|
+
const a = p.digest.attempts[id];
|
|
44
|
+
const run = a && runs.find((r) => r.record.runId === a.runId);
|
|
45
|
+
const record = run?.record.evals.flatMap((e) => e.attempts).find((x) => x.key === a?.key);
|
|
46
|
+
if (!a || !run || !record) return { error: `No attempt ${id}. The attempts are ${Object.keys(p.digest.attempts).join(", ")}.` };
|
|
47
|
+
const view = await fromRecorded(join(run.dir, run.record.runId), record);
|
|
48
|
+
return spend({
|
|
49
|
+
...attemptSummary(view),
|
|
50
|
+
id
|
|
51
|
+
});
|
|
52
|
+
}
|
|
53
|
+
})
|
|
54
|
+
};
|
|
55
|
+
}
|
|
56
|
+
const isError = (v) => typeof v === "object" && v !== null && "error" in v;
|
|
57
|
+
/** What the tools returned, as evidence to cite and as the text the structured call is shown. */
|
|
58
|
+
function toolReads(reads) {
|
|
59
|
+
const ok = reads.filter((r) => !isError(r.output));
|
|
60
|
+
const evidence = ok.filter((r) => r.tool === "read_evidence").map((r) => ({
|
|
61
|
+
...r.output,
|
|
62
|
+
priority: 1
|
|
63
|
+
}));
|
|
64
|
+
if (!ok.length) return {
|
|
65
|
+
evidence,
|
|
66
|
+
text: ""
|
|
67
|
+
};
|
|
68
|
+
return {
|
|
69
|
+
evidence,
|
|
70
|
+
text: `\n\nREAD WITH TOOLS (cite a dropped evidence item read here by its id, as you would the digest's):\n${ok.map((r) => `${r.tool} ${JSON.stringify(r.input)}: ${JSON.stringify(r.output)}`).join("\n")}`
|
|
71
|
+
};
|
|
72
|
+
}
|
|
73
|
+
/** Lets the model read with the tools; what it read, for the structured call. */
|
|
74
|
+
async function investigate(p, model, config) {
|
|
75
|
+
const result = await generateText({
|
|
76
|
+
model: model.fake ? fakeSummaryModel(p) : await providerModel(model),
|
|
77
|
+
system: SUMMARY_TOOLS_PROMPT,
|
|
78
|
+
prompt: promptOf(p.digest),
|
|
79
|
+
tools: summaryTools(p, config.analysis.maxInputTokens * 4),
|
|
80
|
+
stopWhen: stepCountIs(4),
|
|
81
|
+
prepareStep: ({ stepNumber }) => stepNumber === 3 ? { toolChoice: "none" } : void 0,
|
|
82
|
+
temperature: .2,
|
|
83
|
+
maxOutputTokens: config.analysis.maxOutputTokens,
|
|
84
|
+
maxRetries: 3
|
|
85
|
+
});
|
|
86
|
+
return {
|
|
87
|
+
reads: result.steps.flatMap((s) => s.toolResults.map((r) => ({
|
|
88
|
+
tool: r.toolName,
|
|
89
|
+
input: r.input,
|
|
90
|
+
output: r.output
|
|
91
|
+
}))),
|
|
92
|
+
usage: {
|
|
93
|
+
inputTokens: result.totalUsage?.inputTokens ?? 0,
|
|
94
|
+
outputTokens: result.totalUsage?.outputTokens ?? 0
|
|
95
|
+
}
|
|
96
|
+
};
|
|
97
|
+
}
|
|
98
|
+
/**
|
|
99
|
+
* Writes the run's summary into the cache. A failed call is cached as a document with `error`, never
|
|
100
|
+
* a throw; only reading the run (before any call) can throw.
|
|
101
|
+
*/
|
|
102
|
+
async function writeRunSummary(options) {
|
|
103
|
+
const { config, runId, model } = options;
|
|
104
|
+
const entries = await allRuns(config);
|
|
105
|
+
const head = entries.find((e) => e.runId === runId);
|
|
106
|
+
if (!head) throw new Error(`${runId} isn't in .reeval/runs`);
|
|
107
|
+
const prepared = await prepareAnalysis(config, head, await defaultBase(config, entries, head), null);
|
|
108
|
+
const started = Date.now();
|
|
109
|
+
let doc;
|
|
110
|
+
try {
|
|
111
|
+
const { reads, usage } = await investigate(prepared, model, config);
|
|
112
|
+
doc = await analyseDigest(config, prepared, model, toolReads(reads));
|
|
113
|
+
const plus = (a, b) => a === null ? null : a + b;
|
|
114
|
+
doc.usage = {
|
|
115
|
+
inputTokens: plus(doc.usage.inputTokens, usage.inputTokens),
|
|
116
|
+
outputTokens: plus(doc.usage.outputTokens, usage.outputTokens)
|
|
117
|
+
};
|
|
118
|
+
doc.latencyMs = Date.now() - started;
|
|
119
|
+
} catch (error) {
|
|
120
|
+
const message = error instanceof Error ? error.message : String(error);
|
|
121
|
+
doc = await failedAnalysis(config, prepared, model, message, Date.now() - started);
|
|
122
|
+
}
|
|
123
|
+
const path = await writeCached(config.root, prepared.key, doc);
|
|
124
|
+
return {
|
|
125
|
+
doc,
|
|
126
|
+
key: prepared.key,
|
|
127
|
+
path
|
|
128
|
+
};
|
|
129
|
+
}
|
|
130
|
+
/**
|
|
131
|
+
* A language model that reads as the prompt asks a real one to: the dropped evidence and the
|
|
132
|
+
* failing attempts, a few of each, in one round, then says it's done, through the AI SDK's own tool
|
|
133
|
+
* loop, so the tools and the budget run as they would.
|
|
134
|
+
*/
|
|
135
|
+
function fakeSummaryModel(p) {
|
|
136
|
+
const usage = {
|
|
137
|
+
inputTokens: {
|
|
138
|
+
total: 0,
|
|
139
|
+
noCache: 0,
|
|
140
|
+
cacheRead: 0,
|
|
141
|
+
cacheWrite: 0
|
|
142
|
+
},
|
|
143
|
+
outputTokens: {
|
|
144
|
+
total: 0,
|
|
145
|
+
text: 0,
|
|
146
|
+
reasoning: 0
|
|
147
|
+
}
|
|
148
|
+
};
|
|
149
|
+
const answer = (content, finish) => ({
|
|
150
|
+
content,
|
|
151
|
+
finishReason: {
|
|
152
|
+
unified: finish,
|
|
153
|
+
raw: finish === "stop" ? "stop" : "tool_calls"
|
|
154
|
+
},
|
|
155
|
+
usage,
|
|
156
|
+
warnings: [],
|
|
157
|
+
response: { modelId: "fake-analysis" }
|
|
158
|
+
});
|
|
159
|
+
return {
|
|
160
|
+
specificationVersion: "v4",
|
|
161
|
+
provider: "fake",
|
|
162
|
+
modelId: "fake-analysis",
|
|
163
|
+
supportedUrls: {},
|
|
164
|
+
doGenerate: async (options) => {
|
|
165
|
+
const readSome = options.prompt.some((m) => m.role === "tool" && Array.isArray(m.content) && m.content.some((c) => c.type === "tool-result"));
|
|
166
|
+
const calls = [...p.digest.dropped.slice(0, FAKE_READS).map((id) => ({
|
|
167
|
+
name: "read_evidence",
|
|
168
|
+
input: { id }
|
|
169
|
+
})), ...Object.entries(p.digest.attempts).filter(([, a]) => a.outcome === "fail").slice(0, FAKE_READS).map(([id]) => ({
|
|
170
|
+
name: "open_attempt",
|
|
171
|
+
input: { id }
|
|
172
|
+
}))];
|
|
173
|
+
if (readSome || !calls.length) return answer([{
|
|
174
|
+
type: "text",
|
|
175
|
+
text: `Read ${readSome ? "what I needed" : "nothing"}.`
|
|
176
|
+
}], "stop");
|
|
177
|
+
return answer(calls.map((c, i) => ({
|
|
178
|
+
type: "tool-call",
|
|
179
|
+
toolCallId: `fake_${c.name}_${i}`,
|
|
180
|
+
toolName: c.name,
|
|
181
|
+
input: JSON.stringify(c.input)
|
|
182
|
+
})), "tool-calls");
|
|
183
|
+
},
|
|
184
|
+
doStream: async () => {
|
|
185
|
+
throw new Error("the fake analysis model does not stream");
|
|
186
|
+
}
|
|
187
|
+
};
|
|
188
|
+
}
|
|
189
|
+
//#endregion
|
|
190
|
+
export { writeRunSummary };
|
|
191
|
+
|
|
192
|
+
//# sourceMappingURL=summary-DjbxgCDi.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"summary-DjbxgCDi.js","names":[],"sources":["../src/analysis/agent/summary.ts"],"mappings":";;;;;;;;;AA4BA,MAAM,aAAa;AAEnB,MAAa,uBAAuB;;;;;;AAOpC,SAAgB,aAAa,GAAqB,UAAkB;CAClE,IAAI,OAAO;CACX,MAAM,SAAY,UAAoC;EACpD,MAAM,OAAO,KAAK,UAAU,KAAK,CAAC,CAAC;EACnC,IAAI,OAAO,MACT,OAAO,EACL,OACE,8FACJ;EAEF,QAAQ;EACR,OAAO;CACT;CACA,MAAM,OAAO,CAAC,EAAE,SAAS,EAAE,OAAO,CAAC,CAAC,SAAS,MAAO,IAAI,CAAC,CAAC,IAAI,CAAC,CAAE;CACjE,OAAO;EACL,eAAe,KAAK;GAClB,aACE;GACF,aAAa,EAAE,OAAO,EACpB,IAAI,EAAE,OAAO,CAAC,CAAC,SAAS,0CAA+C,EACzE,CAAC;GACD,SAAS,OAAO,EAAE,SAAS;IACzB,MAAM,OAAO,EAAE,QAAQ,MAAM,MAAM,EAAE,OAAO,EAAE;IAC9C,IAAI,MAAM;KACR,MAAM,EAAE,UAAU,WAAW,GAAG,SAAS;KACzC,OAAO,MAAM,IAAI;IACnB;IACA,OAAO,EACL,OAAO,EAAE,OAAO,SAAS,MAAM,MAAM,EAAE,OAAO,EAAE,IAC5C,GAAG,GAAG,8BACN,uBAAuB,GAAG,aAAa,EAAE,OAAO,QAAQ,KAAK,IAAI,KAAK,OAAO,GACnF;GACF;EACF,CAAC;EACD,cAAc,KAAK;GACjB,aACE;GACF,aAAa,EAAE,OAAO,EACpB,IAAI,EAAE,OAAO,CAAC,CAAC,SAAS,gDAAqD,EAC/E,CAAC;GACD,SAAS,OAAO,EAAE,SAAS;IACzB,MAAM,IAAI,EAAE,OAAO,SAAS;IAC5B,MAAM,MAAM,KAAK,KAAK,MAAM,MAAM,EAAE,OAAO,UAAU,EAAE,KAAK;IAC5D,MAAM,SAAS,KAAK,OAAO,MAAM,SAAS,MAAM,EAAE,QAAQ,CAAC,CAAC,MAAM,MAAM,EAAE,QAAQ,GAAG,GAAG;IACxF,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,QACjB,OAAO,EACL,OAAO,cAAc,GAAG,qBAAqB,OAAO,KAAK,EAAE,OAAO,QAAQ,CAAC,CAAC,KAAK,IAAI,EAAE,GACzF;IAEF,MAAM,OAAO,MAAM,aAAa,KAAK,IAAI,KAAK,IAAI,OAAO,KAAK,GAAG,MAAM;IACvE,OAAO,MAAM;KAAE,GAAG,eAAe,IAAI;KAAG;IAAG,CAAC;GAC9C;EACF,CAAC;CACH;AACF;AAIA,MAAM,WAAW,MAAe,OAAO,MAAM,YAAY,MAAM,QAAQ,WAAW;;AAGlF,SAAgB,UAAU,OAA8B;CACtD,MAAM,KAAK,MAAM,QAAQ,MAAM,CAAC,QAAQ,EAAE,MAAM,CAAC;CACjD,MAAM,WAAW,GACd,QAAQ,MAAM,EAAE,SAAS,eAAe,CAAC,CACzC,KAAK,OAAO;EAAE,GAAI,EAAE;EAA6C,UAAU;CAAE,EAAE;CAClF,IAAI,CAAC,GAAG,QAAQ,OAAO;EAAE;EAAU,MAAM;CAAG;CAC5C,OAAO;EACL;EACA,MAAM,uGAAuG,GAC1G,KAAK,MAAM,GAAG,EAAE,KAAK,GAAG,KAAK,UAAU,EAAE,KAAK,EAAE,IAAI,KAAK,UAAU,EAAE,MAAM,GAAG,CAAC,CAC/E,KAAK,IAAI;CACd;AACF;;AAGA,eAAsB,YACpB,GACA,OACA,QACsF;CACtF,MAAM,SAAS,MAAM,aAAa;EAChC,OAAO,MAAM,OAAO,iBAAiB,CAAC,IAAI,MAAM,cAAc,KAAK;EACnE,QAAQ;EACR,QAAQ,SAAS,EAAE,MAAM;EACzB,OAAO,aAAa,GAAG,OAAO,SAAS,iBAAiB,CAAC;EACzD,UAAU,YAAA,CAAyB;EACnC,cAAc,EAAE,iBACd,eAAe,IAAoB,EAAE,YAAY,OAAgB,IAAI,KAAA;EACvE,aAAa;EACb,iBAAiB,OAAO,SAAS;EACjC,YAAY;CACd,CAAC;CACD,OAAO;EACL,OAAO,OAAO,MAAM,SAAS,MAC3B,EAAE,YAAY,KAAK,OAAO;GAAE,MAAM,EAAE;GAAU,OAAO,EAAE;GAAO,QAAQ,EAAE;EAAO,EAAE,CACnF;EACA,OAAO;GACL,aAAa,OAAO,YAAY,eAAe;GAC/C,cAAc,OAAO,YAAY,gBAAgB;EACnD;CACF;AACF;;;;;AAQA,eAAsB,gBAAgB,SAId;CACtB,MAAM,EAAE,QAAQ,OAAO,UAAU;CACjC,MAAM,UAAU,MAAM,QAAQ,MAAM;CACpC,MAAM,OAAO,QAAQ,MAAM,MAAM,EAAE,UAAU,KAAK;CAClD,IAAI,CAAC,MAAM,MAAM,IAAI,MAAM,GAAG,MAAM,uBAAuB;CAC3D,MAAM,WAAW,MAAM,gBACrB,QACA,MACA,MAAM,YAAY,QAAQ,SAAS,IAAI,GACvC,IACF;CACA,MAAM,UAAU,KAAK,IAAI;CACzB,IAAI;CACJ,IAAI;EACF,MAAM,EAAE,OAAO,UAAU,MAAM,YAAY,UAAU,OAAO,MAAM;EAClE,MAAM,MAAM,cAAc,QAAQ,UAAU,OAAO,UAAU,KAAK,CAAC;EACnE,MAAM,QAAQ,GAAkB,MAAe,MAAM,OAAO,OAAO,IAAI;EACvE,IAAI,QAAQ;GACV,aAAa,KAAK,IAAI,MAAM,aAAa,MAAM,WAAW;GAC1D,cAAc,KAAK,IAAI,MAAM,cAAc,MAAM,YAAY;EAC/D;EACA,IAAI,YAAY,KAAK,IAAI,IAAI;CAC/B,SAAS,OAAO;EACd,MAAM,UAAU,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK;EACrE,MAAM,MAAM,eAAe,QAAQ,UAAU,OAAO,SAAS,KAAK,IAAI,IAAI,OAAO;CACnF;CACA,MAAM,OAAO,MAAM,YAAY,OAAO,MAAM,SAAS,KAAK,GAAG;CAC7D,OAAO;EAAE;EAAK,KAAK,SAAS;EAAK;CAAK;AACxC;;;;;;AASA,SAAS,iBAAiB,GAAoC;CAC5D,MAAM,QAAQ;EACZ,aAAa;GAAE,OAAO;GAAG,SAAS;GAAG,WAAW;GAAG,YAAY;EAAE;EACjE,cAAc;GAAE,OAAO;GAAG,MAAM;GAAG,WAAW;EAAE;CAClD;CACA,MAAM,UAAU,SAAoB,YAAmC;EACrE;EACA,cAAc;GAAE,SAAS;GAAQ,KAAK,WAAW,SAAS,SAAS;EAAa;EAChF;EACA,UAAU,CAAC;EACX,UAAU,EAAE,SAAS,gBAAgB;CACvC;CACA,OAAO;EACL,sBAAsB;EACtB,UAAU;EACV,SAAS;EACT,eAAe,CAAC;EAChB,YAAY,OAAO,YAAmE;GACpF,MAAM,WAAW,QAAQ,OAAO,MAC7B,MACC,EAAE,SAAS,UACX,MAAM,QAAQ,EAAE,OAAO,KACtB,EAAE,QAAyB,MAAM,MAAM,EAAE,SAAS,aAAa,CACpE;GACA,MAAM,QAAQ,CACZ,GAAG,EAAE,OAAO,QACT,MAAM,GAAG,UAAU,CAAC,CACpB,KAAK,QAAQ;IAAE,MAAM;IAAiB,OAAO,EAAE,GAAG;GAAE,EAAE,GACzD,GAAG,OAAO,QAAQ,EAAE,OAAO,QAAQ,CAAC,CACjC,QAAQ,GAAG,OAAO,EAAE,YAAY,MAAM,CAAC,CACvC,MAAM,GAAG,UAAU,CAAC,CACpB,KAAK,CAAC,SAAS;IAAE,MAAM;IAAgB,OAAO,EAAE,GAAG;GAAE,EAAE,CAC5D;GACA,IAAI,YAAY,CAAC,MAAM,QACrB,OAAO,OACL,CAAC;IAAE,MAAM;IAAQ,MAAM,QAAQ,WAAW,kBAAkB,UAAU;GAAG,CAAC,GAC1E,MACF;GAEF,OAAO,OACL,MAAM,KAAK,GAAG,OAAO;IACnB,MAAM;IACN,YAAY,QAAQ,EAAE,KAAK,GAAG;IAC9B,UAAU,EAAE;IACZ,OAAO,KAAK,UAAU,EAAE,KAAK;GAC/B,EAAE,GACF,YACF;EACF;EACA,UAAU,YAAY;GACpB,MAAM,IAAI,MAAM,yCAAyC;EAC3D;CACF;AACF"}
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
import { n as clip, r as formatDuration } from "./text-gEcKjAyp.js";
|
|
2
|
-
import {
|
|
3
|
-
import { o as warnings } from "./state-
|
|
2
|
+
import { I as effortLabel, b as languageLabel, h as harnessLabel, j as cellContrast, w as runtimeLabel, y as isUngatedMiss } from "./model-ORtLm2UB.js";
|
|
3
|
+
import { o as warnings } from "./state-g6ij_40j.js";
|
|
4
4
|
//#region src/history/summary.ts
|
|
5
5
|
const OUTCOME_MARK = {
|
|
6
6
|
pass: "✓",
|
|
@@ -36,6 +36,29 @@ function liftLabel(lift) {
|
|
|
36
36
|
const counts = `${variant.pass}/${variant.counted} vs ${control.pass}/${control.counted}`;
|
|
37
37
|
return `lift ${signedPoints(lift.points)} pts (${range}, ${counts})${lift.withinNoise ? ", within noise" : ""}`;
|
|
38
38
|
}
|
|
39
|
+
const rateText = (r) => `${r.name} ${r.wins}/${r.n} (${Math.round(r.wins / r.n * 100)}%)`;
|
|
40
|
+
/**
|
|
41
|
+
* One compare question's answer: `"Which is cleaner?" within: with-skill 2/2 (100%), plain 0/2
|
|
42
|
+
* (0%) · 2 votes, 0 ties · google/gemini-3.8-flash`, or why it has no votes.
|
|
43
|
+
*/
|
|
44
|
+
function answersLabel(answer) {
|
|
45
|
+
const q = `"${clip(answer.question, 80)}"`;
|
|
46
|
+
const ties = answer.votes.filter((v) => v.winner === "tie").length;
|
|
47
|
+
const cuts = ["within", "across"].flatMap((cut) => {
|
|
48
|
+
const rates = answer.rates.filter((r) => r.cut === cut);
|
|
49
|
+
return rates.length ? [`${cut}: ${rates.map(rateText).join(", ")}`] : [];
|
|
50
|
+
});
|
|
51
|
+
const votes = answer.votes.length;
|
|
52
|
+
const tally = `${votes} vote${votes === 1 ? "" : "s"}, ${ties} tie${ties === 1 ? "" : "s"}`;
|
|
53
|
+
const by = answer.resolvedModel ?? answer.model;
|
|
54
|
+
const failed = answer.error !== void 0 ? ` · failed: ${clip(answer.error, 200)}` : "";
|
|
55
|
+
if (votes === 0 && answer.error === void 0) return `${q} no pairs to judge: attempts that failed their checks are left out`;
|
|
56
|
+
return `${q} ${[
|
|
57
|
+
...cuts,
|
|
58
|
+
tally,
|
|
59
|
+
by
|
|
60
|
+
].join(" · ")}${failed}`;
|
|
61
|
+
}
|
|
39
62
|
/** `fake/-/fake 0/2 (control, not gated) · fake/with-skill/fake 2/2 (lift +100 pts …)`. */
|
|
40
63
|
function cellCounts(e) {
|
|
41
64
|
return e.cells.map((c) => {
|
|
@@ -73,7 +96,8 @@ function stateRows(run) {
|
|
|
73
96
|
rows.push(["Judge", `${env.judge.model ?? "none by default"}${env.judge.resolvedModels.length ? ` (served ${env.judge.resolvedModels.join(", ")})` : ""} · threshold ${env.judge.threshold} · version ${env.judge.version}`]);
|
|
74
97
|
rows.push(["Context", env.context.length ? `${env.context.length} file${env.context.length === 1 ? "" : "s"}: ${env.context.map((c) => `${c.path} ${short(c.sha256)}`).join(", ")}` : "no skills, rules or MCP config found in the sandbox"]);
|
|
75
98
|
rows.push(["Packages", Object.entries(env.packages).map(([name, v]) => `${name} ${v}`).join(" · ")]);
|
|
76
|
-
|
|
99
|
+
const language = languageLabel(env);
|
|
100
|
+
rows.push(["reeval", `${env.es} · ${runtimeLabel(env)}${language ? ` · ${language}` : ""} · ${env.os}/${env.arch}`]);
|
|
77
101
|
rows.push(["Runner", run.runner.kind === "ci" ? `CI (${run.runner.provider})${run.runner.runUrl ? ` ${run.runner.runUrl}` : ""}` : `local · ${run.author}`]);
|
|
78
102
|
return rows;
|
|
79
103
|
}
|
|
@@ -111,6 +135,10 @@ function renderSummary(run) {
|
|
|
111
135
|
lines.push("", `### ${OUTCOME_MARK[e.outcome]} ${e.id}`, "");
|
|
112
136
|
const multi = e.attempts.length > 1;
|
|
113
137
|
if (multi) lines.push(`Cells: ${cellCounts(e)}`, "");
|
|
138
|
+
if (e.answers?.length) {
|
|
139
|
+
for (const a of e.answers) lines.push(`- Question: ${cell(answersLabel(a))}`);
|
|
140
|
+
lines.push("");
|
|
141
|
+
}
|
|
114
142
|
const attemptRank = (a) => isUngatedMiss(e, a) ? 3 : rank(a.outcome);
|
|
115
143
|
for (const a of [...e.attempts].sort((x, y) => attemptRank(x) - attemptRank(y))) {
|
|
116
144
|
if (a.outcome === "skipped") continue;
|
|
@@ -161,6 +189,6 @@ function attemptDetail(lines, a) {
|
|
|
161
189
|
if (links.length) lines.push(`- Files: ${links.join(" · ")}`);
|
|
162
190
|
}
|
|
163
191
|
//#endregion
|
|
164
|
-
export {
|
|
192
|
+
export { liftLabel as a, stateRows as c, gateLabel as i, summaryCounts as l, answersLabel as n, renderSummary as o, cellCounts as r, repeatsVerdict as s, OUTCOME_MARK as t };
|
|
165
193
|
|
|
166
|
-
//# sourceMappingURL=summary-
|
|
194
|
+
//# sourceMappingURL=summary-gawvEpNu.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"summary-gawvEpNu.js","names":[],"sources":["../src/history/summary.ts"],"mappings":";;;;AAOA,MAAa,eAAwC;CACnD,MAAM;CACN,MAAM;CACN,aAAa;CACb,SAAS;AACX;;AAGA,MAAM,qBAAqB;AAE3B,MAAM,QAAQ,MAAc,EAAE,QAAQ,OAAO,KAAK,CAAC,CAAC,QAAQ,OAAO,GAAG;AAEtE,SAAgB,cAAc,GAAiC;CAC7D,MAAM,QAAQ;EACZ,EAAE,QAAQ,GAAG,EAAE,KAAK;EACpB,EAAE,OAAO,EAAE,WAAW,KAAK,GAAG,EAAE,OAAO,EAAE,SAAS;EAClD,EAAE,WAAW,KAAK,GAAG,EAAE,SAAS;EAChC,EAAE,cAAc,GAAG,EAAE,WAAW;EAChC,EAAE,WAAW,GAAG,EAAE,QAAQ;CAC5B,CAAC,CAAC,OAAO,OAAO;CAChB,IAAI,MAAM,WAAW,GAAG,OAAO;CAC/B,MAAM,QAAQ,EAAE,aAAa,EAAE,QAAQ,KAAK,KAAK,EAAE,SAAS,aAAa,EAAE,MAAM;CACjF,OAAO,GAAG,MAAM,KAAK,KAAK,IAAI;AAChC;AAEA,MAAM,gBAAgB,MAAc,GAAG,KAAK,IAAI,MAAM,MAAM,KAAK,MAAM,KAAK,IAAI,CAAC,IAAI,GAAG;;AAGxF,MAAa,aAAa,SACxB,SAAS,SAAS,SAAS,SAAS,OAAO,YAAY,KAAK;;;;;AAM9D,SAAgB,UAAU,MAA+C;CACvE,MAAM,EAAE,UAAU,SAAS,YAAY;CACvC,MAAM,QAAQ,OAAO,aAAa,SAAS,EAAE,EAAE,GAAG,aAAa,SAAS,EAAE;CAC1E,MAAM,SAAS,GAAG,QAAQ,KAAK,GAAG,QAAQ,QAAQ,MAAM,QAAQ,KAAK,GAAG,QAAQ;CAChF,OAAO,QAAQ,aAAa,KAAK,MAAM,EAAE,QAAQ,MAAM,IAAI,OAAO,GAAG,KAAK,cAAc,mBAAmB;AAC7G;AAEA,MAAM,YAAY,MAChB,GAAG,EAAE,KAAK,GAAG,EAAE,KAAK,GAAG,EAAE,EAAE,IAAI,KAAK,MAAO,EAAE,OAAO,EAAE,IAAK,GAAG,EAAE;;;;;AAMlE,SAAgB,aAAa,QAA+B;CAC1D,MAAM,IAAI,IAAI,KAAK,OAAO,UAAU,EAAE,EAAE;CACxC,MAAM,OAAO,OAAO,MAAM,QAAQ,MAAM,EAAE,WAAW,KAAK,CAAC,CAAC;CAC5D,MAAM,OAAQ,CAAC,UAAU,QAAQ,CAAC,CAAW,SAAS,QAAQ;EAC5D,MAAM,QAAQ,OAAO,MAAM,QAAQ,MAAM,EAAE,QAAQ,GAAG;EACtD,OAAO,MAAM,SAAS,CAAC,GAAG,IAAI,IAAI,MAAM,IAAI,QAAQ,CAAC,CAAC,KAAK,IAAI,GAAG,IAAI,CAAC;CACzE,CAAC;CACD,MAAM,QAAQ,OAAO,MAAM;CAC3B,MAAM,QAAQ,GAAG,MAAM,OAAO,UAAU,IAAI,KAAK,IAAI,IAAI,KAAK,MAAM,SAAS,IAAI,KAAK;CACtF,MAAM,KAAK,OAAO,iBAAiB,OAAO;CAC1C,MAAM,SAAS,OAAO,UAAU,KAAA,IAAY,cAAc,KAAK,OAAO,OAAO,GAAG,MAAM;CACtF,IAAI,UAAU,KAAK,OAAO,UAAU,KAAA,GAClC,OAAO,GAAG,EAAE;CAEd,OAAO,GAAG,EAAE,GAAG;EAAC,GAAG;EAAM;EAAO;CAAE,CAAC,CAAC,KAAK,KAAK,IAAI;AACpD;;AAGA,SAAgB,WAAW,GAAsB;CAC/C,OAAO,EAAE,MACN,KAAK,MAAM;EACV,MAAM,QAAQ;GACZ,EAAE,SAAS,aAAa,EAAE,MAAM,MAAM,MAAM,EAAE,SAAS,SAAS,IAAI,YAAY;GAChF,EAAE,SAAS,SAAS,cAAc;GAClC,EAAE,SAAS,SAAS,EAAE,SAAS,SAAS,QAAQ,UAAU,EAAE,IAAI,MAAM;GACtE,EAAE,OAAO,UAAU,EAAE,IAAI,IAAI;GAC7B,GAAG,CAAC,aAAa,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC,SAAS,WACnC,SAAS,CAAC,UAAU,YAAY,MAAM,GAAG,IAAI,CAAC,CAChD;GACA,EAAE,UAAU,SAAS,aAAa,EAAE,UAAU,KAAK,IAAI,MAAM;EAC/D,CAAC,CAAC,OAAO,OAAO;EAChB,OAAO,GAAG,EAAE,IAAI,GAAG,EAAE,KAAK,GAAG,EAAE,OAAO,EAAE,OAAO,MAAM,SAAS,KAAK,MAAM,KAAK,IAAI,EAAE,KAAK;CAC3F,CAAC,CAAC,CACD,KAAK,KAAK;AACf;AAEA,SAAS,UAAU,GAAsB;CACvC,MAAM,aAAa,EAAE,SAAS,SAAS,MAAM,EAAE,UAAU;CACzD,IAAI,WAAW,WAAW,GAAG,OAAO;CACpC,OAAO,WAAW,KAAK,MAAM,GAAG,EAAE,QAAQ,GAAG,EAAE,MAAM,QAAQ,CAAC,EAAE,GAAG,EAAE,WAAW,CAAC,CAAC,KAAK,IAAI;AAC7F;;AAGA,MAAM,SAAS,SAAiB,KAAK,QAAQ,YAAY,EAAE,CAAC,CAAC,MAAM,GAAG,EAAE;;;;;AAMxE,SAAgB,UAAU,KAAyC;CACjE,MAAM,EAAE,KAAK,OAAO,QAAQ;CAC5B,MAAM,OAAgC,CAAC;CACvC,KAAK,KAAK,CACR,eACA,MACI,GAAG,IAAI,IAAI,MAAM,GAAG,EAAE,IAAI,IAAI,SAAS,OAAO,IAAI,WAAW,KAAK,IAAI,QAAQ,YAAY,UAAU,gDACpG,sBACN,CAAC;CACD,IAAI,KAAK,OACP,KAAK,KAAK,CACR,eACA,IAAI,gBACA,mBAAmB,IAAI,cAAc,MAAM,iBAAiB,MAAM,IAAI,cAAc,MAAM,MAC1F,cAAc,IAAI,qBAAqB,YAAY,IAAI,WAAW,YAAY,MAAM,IAAI,QAAQ,MAAM,IAC5G,CAAC;CAEH,KAAK,KAAK,CACR,aACA,IAAI,aAAa,UAAU,KAAK,MAAM,GAAG,EAAE,KAAK,GAAG,MAAM,EAAE,MAAM,GAAG,CAAC,CAAC,KAAK,IAAI,KAC7E,YACJ,CAAC;CACD,KAAK,KAAK,CACR,UACA,IAAI,OAAO,SACP,IAAI,OACD,KACE,MACC,GAAG,EAAE,IAAI,QAAQ,MAAM,EAAE,EAAE,EAAE,KAAK,EAAE,YAAY,MAAM,qCAAqC,SAAS,EAAE,YAAY,QAAQ,EAAE,aAAa,iBAAiB,EAAE,WAAW,KAAK,GAAG,MAAM,EAAE,WAAW,MAAM,MAAM,KAAK,EAAE,SAAS,SAAS,cAAc,EAAE,QAAQ,OAAO,MAAM,EAAE,SAAS,IAC3R,CAAC,CACA,KAAK,IAAI,IACZ,qBACN,CAAC;CACD,KAAK,KAAK,CAAC,aAAa,IAAI,UAAU,IAAI,YAAY,CAAC,CAAC,KAAK,IAAI,KAAK,MAAM,CAAC;CAC7E,KAAK,KAAK,CACR,gBACA,IAAI,aACD,KACE,MACC,GAAG,EAAE,KAAK,GAAG,MAAM,EAAE,MAAM,IAAI,EAAE,OAAO,SAAS,aAAa,EAAE,OAAO,KAAK,MAAM,GAAG,EAAE,OAAO,EAAE,SAAS,IAAI,EAAE,OAAO,MAAM,GAAG,CAAC,MAAM,IAAI,CAAC,CAAC,KAAK,IAAI,MAAM,KAAK,EAAE,IAAI,SAAS,UAAU,EAAE,IAAI,KAAK,IAAI,MAAM,KAAK,EAAE,QAAQ,SAAS,cAAc,EAAE,QAAQ,KAAK,IAAI,MAAM,IACjR,CAAC,CACA,KAAK,IAAI,KAAK,MACnB,CAAC;CACD,KAAK,KAAK,CACR,SACA,GAAG,IAAI,MAAM,SAAS,oBAAoB,IAAI,MAAM,eAAe,SAAS,YAAY,IAAI,MAAM,eAAe,KAAK,IAAI,EAAE,KAAK,GAAG,eAAe,IAAI,MAAM,UAAU,aAAa,IAAI,MAAM,SAChM,CAAC;CACD,KAAK,KAAK,CACR,WACA,IAAI,QAAQ,SACR,GAAG,IAAI,QAAQ,OAAO,OAAO,IAAI,QAAQ,WAAW,IAAI,KAAK,IAAI,IAAI,IAAI,QAAQ,KAAK,MAAM,GAAG,EAAE,KAAK,GAAG,MAAM,EAAE,MAAM,GAAG,CAAC,CAAC,KAAK,IAAI,MACrI,qDACN,CAAC;CACD,KAAK,KAAK,CACR,YACA,OAAO,QAAQ,IAAI,QAAQ,CAAC,CACzB,KAAK,CAAC,MAAM,OAAO,GAAG,KAAK,GAAG,GAAG,CAAC,CAClC,KAAK,KAAK,CACf,CAAC;CACD,MAAM,WAAW,cAAc,GAAG;CAClC,KAAK,KAAK,CACR,UACA,GAAG,IAAI,GAAG,KAAK,aAAa,GAAG,IAAI,WAAW,MAAM,aAAa,GAAG,KAAK,IAAI,GAAG,GAAG,IAAI,MACzF,CAAC;CACD,KAAK,KAAK,CACR,UACA,IAAI,OAAO,SAAS,OAChB,OAAO,IAAI,OAAO,SAAS,GAAG,IAAI,OAAO,SAAS,IAAI,IAAI,OAAO,WAAW,OAC5E,WAAW,IAAI,QACrB,CAAC;CACD,OAAO;AACT;AAEA,SAAgB,cAAc,KAAwB;CACpD,MAAM,QAAQ;EACZ,gBAAgB,IAAI;EACpB;EACA,GAAG,cAAc,IAAI,OAAO,IAAI,IAAI,WAAW,aAAa,KAAK,QAAQ,IAAI,WAAW,WAAW,mBAAmB,IAAI,OAAO,IAAI,aAAa,IAAI,UAAU,IAAI,eAAe,IAAI,UAAU,EAAE;EACnM;EACA,SAAS,IAAI,UAAU,WAAW,IAAI,QAAQ,KAAK,IAAI,IAAI,QAAQ,OAAO,QAAQ,KAAK,IAAI,KAAK,MAAM,WAAW,IAAI,KAAK,MAAM,WAAW,IAAI,KAAK,SAAS,WAAW,IAAI,SAAS,aAAa,CAAC,IAAI,OAAO,aAAa,KAAA,IAAY,GAAG,IAAI,OAAO,SAAS,aAAa,IAAI,IAAI,OAAO,YAAY,EAAE,CAAC,CAAC,OAAO,OAAO,CAAC,CAAC,KAAK,IAAI,MAAM;EACzU;EACA,GAAG,IAAI,KAAK,gBAAgB,KAAK,MAAM,kBAAkB,EAAE,QAAQ,GAAG;EACtE;EACA;EACA;EACA;EACA,GAAG,UAAU,GAAG,CAAC,CAAC,KAAK,CAAC,GAAG,OAAO,KAAK,EAAE,KAAK,KAAK,CAAC,EAAE,GAAG;EACzD;EACA,GAAG,SAAS,GAAG,CAAC,CAAC,KAAK,MAAM,kBAAkB,EAAE,GAAG;EACnD,YAAY,IAAI,MAAM,YAAY,IAAI,MAAM,sBAAsB,IAAI;EACtE;EACA;EACA;EACA;EACA;CACF;CACA,KAAK,MAAM,KAAK,IAAI,OAAO;EACzB,MAAM,IAAI,EAAE,SAAS;EACrB,MAAM,OAAO,EAAE,OACX,SACA,EAAE,SAAS,SAAS,IAClB,WAAW,CAAC,IACZ,IACE,eAAe,EAAE,UAAU,IAC3B;EACR,MAAM,KAAK,KAAK,aAAa,EAAE,SAAS,KAAK,KAAK,EAAE,EAAE,EAAE,KAAK,KAAK,IAAI,EAAE,KAAK,UAAU,CAAC,EAAE,GAAG;CAC/F;CAGA,MAAM,QAAQ,MAAgB,MAAM,UAAU,MAAM,gBAAgB,IAAI,MAAM,SAAS,IAAI;CAC3F,KAAK,MAAM,KAAK,CAAC,GAAG,IAAI,KAAK,CAAC,CAAC,MAAM,GAAG,MAAM,KAAK,EAAE,OAAO,IAAI,KAAK,EAAE,OAAO,CAAC,GAAG;EAChF,IAAI,EAAE,YAAY,WAAW;EAC7B,MAAM,KAAK,IAAI,OAAO,aAAa,EAAE,SAAS,GAAG,EAAE,MAAM,EAAE;EAC3D,MAAM,QAAQ,EAAE,SAAS,SAAS;EAClC,IAAI,OAAO,MAAM,KAAK,UAAU,WAAW,CAAC,KAAK,EAAE;EACnD,IAAI,EAAE,SAAS,QAAQ;GACrB,KAAK,MAAM,KAAK,EAAE,SAAS,MAAM,KAAK,eAAe,KAAK,aAAa,CAAC,CAAC,GAAG;GAC5E,MAAM,KAAK,EAAE;EACf;EACA,MAAM,eAAe,MAAsB,cAAc,GAAG,CAAC,IAAI,IAAI,KAAK,EAAE,OAAO;EACnF,KAAK,MAAM,KAAK,CAAC,GAAG,EAAE,QAAQ,CAAC,CAAC,MAAM,GAAG,MAAM,YAAY,CAAC,IAAI,YAAY,CAAC,CAAC,GAAG;GAC/E,IAAI,EAAE,YAAY,WAAW;GAC7B,IAAI,cAAc,GAAG,CAAC,GAAG;IACvB,IAAI,OAAO,MAAM,KAAK,UAAU,EAAE,GAAG,uBAAuB,EAAE;IAC9D,MAAM,KAAK,6CAA6C,EAAE;IAC1D,cAAc,OAAO,CAAC;IACtB,MAAM,KAAK,IAAI,YAAY;IAC3B;GACF;GACA,IAAI,OAAO,MAAM,KAAK,QAAQ,aAAa,EAAE,SAAS,GAAG,EAAE,MAAM,EAAE;GACnE,cAAc,OAAO,CAAC;EACxB;CACF;CACA,MAAM,KAAK,EAAE;CACb,OAAO,MAAM,KAAK,IAAI;AACxB;;AAGA,MAAa,kBAAkB,MAC7B,EAAE,OAAO,SAAS,WAAW,EAAE,WAAW,MAAM,MAAM,EAAE,YAAY,MAAM;AAE5E,SAAS,cAAc,OAAiB,GAAwB;CAC9D,IAAI,EAAE,SAAS,CAAC,eAAe,CAAC,GAC9B,MAAM,KACJ,KAAK,EAAE,MAAM,KAAK,KAClB,IACA,OACA,EAAE,MAAM,QAAQ,MAAM,IAAI,CAAC,CAAC,MAAM,GAAG,EAAE,CAAC,CAAC,KAAK,IAAI,GAClD,OACA,EACF;CAEF,KAAK,MAAM,QAAQ,EAAE,OAAO;EAC1B,MAAM,KACJ,aAAa,KAAK,OAAO,IAAI,eAAe,KAAK,UAAU,EAAE,KAAK,KAAK,KAAK,KAAK,MAAM,GAAG,GAAG,GAC/F;EACA,IAAI,KAAK,OAAO,MAAM,KAAK,cAAc,KAAK,KAAK,KAAK,KAAK,GAAG,GAAG,GAAG;CACxE;CACA,MAAM,IAAI,EAAE;CACZ,MAAM,KACJ,iBAAiB,eAAe,EAAE,OAAO,EAAE,WAAW,eAAe,EAAE,OAAO,EAAE,YAAY,eAAe,EAAE,QAAQ,EAAE,WAAW,eAAe,EAAE,OAAO,GAC5J;CACA,MAAM,MAAM,EAAE;CACd,MAAM,OAAO,GAAkB,SAC7B,MAAM,OAAO,GAAG,KAAK,iBAAiB,GAAG,EAAE,GAAG;CAChD,MAAM,KACJ,YAAY,IAAI,UAAU,gBAAgB,IAAI,MAAM,sBAAsB,IAAI,IAAI,aAAa,cAAc,EAAE,KAAK,IAAI,IAAI,YAAY,SAAS,EAAE,KAAK,eAAe,IAAI,WAAW,EAAE,WAC1L;CACA,KAAK,MAAM,KAAK,EAAE,YAAY;EAC5B,MAAM,UAAU,EAAE,SAAS,SACvB,WAAW,EAAE,QAAQ,OAAO,GAAG,EAAE,QAAQ,WAAW,IAAI,UAAU,cAClE;EACJ,MAAM,KACJ,aAAa,EAAE,QAAQ,KAAK,EAAE,MAAM,QAAQ,CAAC,EAAE,cAAc,EAAE,UAAU,IAAI,EAAE,QAAQ,QAAQ,KAAK,EAAE,UACxG;CACF;CACA,KAAK,MAAM,KAAK,EAAE,QAChB,MAAM,KACJ,gBAAgB,EAAE,SAAS,MAAM,KAAK,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,EAAE,IAAI,KAAK,MAAM,IAC3F;CAEF,IAAI,EAAE,aAAa,QACjB,MAAM,KACJ,cAAc,EAAE,aAAa,KAAK,MAAO,EAAE,QAAQ,KAAK,EAAE,KAAK,MAAM,EAAE,MAAM,WAAW,GAAG,EAAE,OAAO,KAAK,EAAE,KAAK,GAAI,CAAC,CAAC,KAAK,IAAI,GACjI;CAEF,MAAM,OAAO,EAAE,MAAM,GAAG,EAAE,CAAC,EAAE;CAC7B,IAAI,MAAM;EACR,MAAM,QAAQ,KAAK,SAAS,qBAAqB,GAAG,KAAK,MAAM,GAAG,kBAAkB,EAAE,KAAK;EAC3F,MAAM,KACJ,IACA,iBACA,IACA,GAAG,MACA,QAAQ,CAAC,CACT,MAAM,IAAI,CAAC,CACX,KAAK,MAAM,KAAK,GAAG,GACtB,EACF;CACF;CACA,MAAM,QAAQ;EACZ,EAAE,MAAM,YAAY,eAAe,EAAE,MAAM,SAAS,KAAK;EACzD,EAAE,MAAM,cAAc,gBAAgB,EAAE,MAAM,WAAW,KAAK;EAC9D,EAAE,MAAM,QAAQ,UAAU,EAAE,MAAM,KAAK,KAAK;EAC5C,EAAE,MAAM,cAAc,iBAAiB,EAAE,MAAM,WAAW,KAAK;CACjE,CAAC,CAAC,OAAO,OAAO;CAChB,IAAI,MAAM,QAAQ,MAAM,KAAK,YAAY,MAAM,KAAK,KAAK,GAAG;AAC9D"}
|
|
@@ -0,0 +1,249 @@
|
|
|
1
|
+
import { s as recordFileName } from "./ids-Bp6ut7ZD.js";
|
|
2
|
+
import { n as clip } from "./text-gEcKjAyp.js";
|
|
3
|
+
import { x as parseTranscript } from "./model-ORtLm2UB.js";
|
|
4
|
+
import { s as mergeOutcome } from "./assemble-h8s0hzTQ.js";
|
|
5
|
+
import { join } from "node:path";
|
|
6
|
+
import { readFile } from "node:fs/promises";
|
|
7
|
+
//#region src/analysis/agent/tools.ts
|
|
8
|
+
const INPUT_KEYS = [
|
|
9
|
+
"command",
|
|
10
|
+
"cmd",
|
|
11
|
+
"path",
|
|
12
|
+
"file_path",
|
|
13
|
+
"filePath",
|
|
14
|
+
"target_file",
|
|
15
|
+
"pattern",
|
|
16
|
+
"query",
|
|
17
|
+
"url"
|
|
18
|
+
];
|
|
19
|
+
/** A tool call's input in a few words: its command or path when it has one. */
|
|
20
|
+
function inputSummary(input) {
|
|
21
|
+
if (typeof input === "string") return clip(input.replace(/\s+/g, " "), 120);
|
|
22
|
+
if (typeof input !== "object" || input === null) return "";
|
|
23
|
+
const fields = input;
|
|
24
|
+
for (const key of INPUT_KEYS) {
|
|
25
|
+
const value = fields[key];
|
|
26
|
+
if (typeof value === "string" && value) return clip(value.replace(/\s+/g, " "), 120);
|
|
27
|
+
if (Array.isArray(value) && value.every((v) => typeof v === "string")) return clip(value.join(" "), 120);
|
|
28
|
+
}
|
|
29
|
+
return clip(JSON.stringify(input), 120);
|
|
30
|
+
}
|
|
31
|
+
/** One transcript event as a step, or null for text and reasoning (a turn's text is kept whole). */
|
|
32
|
+
function stepOf(event) {
|
|
33
|
+
switch (event.type) {
|
|
34
|
+
case "tool": {
|
|
35
|
+
const input = inputSummary(event.input);
|
|
36
|
+
return `${event.tool}${input ? `: ${input}` : ""}${event.ok ? "" : " (failed)"}`;
|
|
37
|
+
}
|
|
38
|
+
case "file-edit": return `edit (${event.kind}): ${event.path}`;
|
|
39
|
+
case "web-search": return `web-search: ${clip(event.query, 120)}`;
|
|
40
|
+
case "web-fetch": return `web-fetch: ${event.url}${event.ok ? "" : " (failed)"}`;
|
|
41
|
+
case "error": return `error: ${clip(event.message, 200)}`;
|
|
42
|
+
case "retry": return `retry: ${clip(event.message, 200)}`;
|
|
43
|
+
default: return null;
|
|
44
|
+
}
|
|
45
|
+
}
|
|
46
|
+
const steps = (events) => events.flatMap((e) => stepOf(e) ?? []);
|
|
47
|
+
const tail = (text, max) => text.length <= max ? text : `…${text.slice(-max)}`;
|
|
48
|
+
const head = (text, max) => text.length <= max ? text : `${text.slice(0, max)}\n… ${text.length - max} more characters`;
|
|
49
|
+
/** A worker's staged record, read the way the run will assemble it. */
|
|
50
|
+
function fromStaged(result, record) {
|
|
51
|
+
const attempt = result.attempt;
|
|
52
|
+
if (!attempt) return null;
|
|
53
|
+
const { outcome, error } = mergeOutcome(result, record);
|
|
54
|
+
return {
|
|
55
|
+
id: `${attempt.cell.key}#${attempt.n}`,
|
|
56
|
+
cell: attempt.cell.key,
|
|
57
|
+
n: attempt.n,
|
|
58
|
+
outcome,
|
|
59
|
+
error: error ? {
|
|
60
|
+
kind: error.kind,
|
|
61
|
+
message: error.message
|
|
62
|
+
} : null,
|
|
63
|
+
checks: (record?.checks ?? []).map((c) => ({
|
|
64
|
+
command: c.command,
|
|
65
|
+
exitCode: c.exitCode,
|
|
66
|
+
output: tail(`${c.stdout}\n${c.stderr}`.trim(), 600)
|
|
67
|
+
})),
|
|
68
|
+
judgements: (record?.judgements ?? []).map((j) => ({
|
|
69
|
+
question: j.question,
|
|
70
|
+
verdict: j.verdict,
|
|
71
|
+
score: j.score
|
|
72
|
+
})),
|
|
73
|
+
turns: (record?.turns ?? []).map((t) => ({
|
|
74
|
+
prompt: t.prompt,
|
|
75
|
+
text: t.text,
|
|
76
|
+
status: t.status,
|
|
77
|
+
steps: steps(t.transcript)
|
|
78
|
+
})),
|
|
79
|
+
files: (record?.diff?.files ?? []).map((f) => f.path),
|
|
80
|
+
patch: record?.diff?.patch ?? ""
|
|
81
|
+
};
|
|
82
|
+
}
|
|
83
|
+
/** The staged attempts of one eval on one harness, from `.reeval/staging/<run>/evals`. */
|
|
84
|
+
async function stagedAttempts(stagingDir, results) {
|
|
85
|
+
const views = [];
|
|
86
|
+
for (const result of results) {
|
|
87
|
+
const path = join(stagingDir, "evals", recordFileName(result.key));
|
|
88
|
+
let record = null;
|
|
89
|
+
try {
|
|
90
|
+
record = JSON.parse(await readFile(path, "utf8"));
|
|
91
|
+
} catch {}
|
|
92
|
+
const view = fromStaged(result, record);
|
|
93
|
+
if (view) views.push(view);
|
|
94
|
+
}
|
|
95
|
+
return views;
|
|
96
|
+
}
|
|
97
|
+
/** A recorded attempt, read from its run's artifact directory (`<history>/<run>/`, or `.reeval/runs/<run>/`). */
|
|
98
|
+
async function fromRecorded(runDir, attempt) {
|
|
99
|
+
const read = (path) => path ? readFile(join(runDir, path), "utf8").catch(() => "") : Promise.resolve("");
|
|
100
|
+
const lines = parseTranscript(await read(attempt.files.transcript?.path));
|
|
101
|
+
return {
|
|
102
|
+
id: attempt.id,
|
|
103
|
+
cell: attempt.cell,
|
|
104
|
+
n: attempt.n,
|
|
105
|
+
outcome: attempt.outcome,
|
|
106
|
+
error: attempt.error ? {
|
|
107
|
+
kind: attempt.error.kind,
|
|
108
|
+
message: attempt.error.message
|
|
109
|
+
} : null,
|
|
110
|
+
checks: await Promise.all(attempt.checks.map(async (c) => ({
|
|
111
|
+
command: c.command,
|
|
112
|
+
exitCode: c.exitCode,
|
|
113
|
+
output: tail((await read(c.log?.path)).trim(), 600)
|
|
114
|
+
}))),
|
|
115
|
+
judgements: attempt.judgements.map((j) => ({
|
|
116
|
+
question: j.question,
|
|
117
|
+
verdict: j.verdict,
|
|
118
|
+
score: j.score
|
|
119
|
+
})),
|
|
120
|
+
turns: attempt.turns.map((t, i) => ({
|
|
121
|
+
prompt: t.prompt,
|
|
122
|
+
text: t.text,
|
|
123
|
+
status: t.status,
|
|
124
|
+
steps: steps(lines.filter((l) => l.turn === i && l.type !== "prompt" && l.type !== "turn-end"))
|
|
125
|
+
})),
|
|
126
|
+
files: attempt.changedFiles.map((f) => f.path),
|
|
127
|
+
patch: await read(attempt.files.diff?.path)
|
|
128
|
+
};
|
|
129
|
+
}
|
|
130
|
+
/** Each failed attempt beside the nearest passing run of its cell, else of another cell. */
|
|
131
|
+
function failedVsPassed(attempts) {
|
|
132
|
+
const passed = attempts.filter((a) => a.outcome === "pass");
|
|
133
|
+
const near = (a) => (b, c) => Math.abs(b.n - a.n) - Math.abs(c.n - a.n);
|
|
134
|
+
return attempts.filter((a) => a.outcome === "fail").flatMap((a) => {
|
|
135
|
+
const sameCell = passed.filter((p) => p.cell === a.cell).sort(near(a))[0];
|
|
136
|
+
const other = passed.filter((p) => p.cell !== a.cell).sort(near(a))[0];
|
|
137
|
+
const match = sameCell ?? other;
|
|
138
|
+
return match ? [{
|
|
139
|
+
failed: a.id,
|
|
140
|
+
passed: match.id,
|
|
141
|
+
match: sameCell ? "cell" : "harness"
|
|
142
|
+
}] : [];
|
|
143
|
+
});
|
|
144
|
+
}
|
|
145
|
+
const stepCount = (a) => a.turns.reduce((n, t) => n + t.steps.length, 0);
|
|
146
|
+
/** `list_attempts`: every attempt's outcome in a line, and the failed-vs-passed pairs to open. */
|
|
147
|
+
function listOutcomes(attempts) {
|
|
148
|
+
return {
|
|
149
|
+
attempts: attempts.map((a) => ({
|
|
150
|
+
id: a.id,
|
|
151
|
+
outcome: a.outcome,
|
|
152
|
+
error: a.error ? `${a.error.kind}: ${clip(a.error.message, 300)}` : null,
|
|
153
|
+
failedChecks: a.checks.filter((c) => c.exitCode !== 0).map((c) => c.command),
|
|
154
|
+
judgements: a.judgements.map((j) => `${j.verdict} ${j.score.toFixed(2)}: ${clip(j.question, 120)}`),
|
|
155
|
+
turns: a.turns.length,
|
|
156
|
+
steps: stepCount(a),
|
|
157
|
+
files: a.files.length
|
|
158
|
+
})),
|
|
159
|
+
pairs: failedVsPassed(attempts)
|
|
160
|
+
};
|
|
161
|
+
}
|
|
162
|
+
/** `open_attempt`: one attempt's turns, steps (the first 40 and last 20), checks, verdicts and diff. */
|
|
163
|
+
function attemptSummary(a) {
|
|
164
|
+
return {
|
|
165
|
+
id: a.id,
|
|
166
|
+
outcome: a.outcome,
|
|
167
|
+
error: a.error ? {
|
|
168
|
+
kind: a.error.kind,
|
|
169
|
+
message: clip(a.error.message, 1e3)
|
|
170
|
+
} : null,
|
|
171
|
+
turns: a.turns.map((t, i) => ({
|
|
172
|
+
turn: i + 1,
|
|
173
|
+
prompt: clip(t.prompt, 400),
|
|
174
|
+
status: t.status,
|
|
175
|
+
steps: t.steps.length > 60 ? [
|
|
176
|
+
...t.steps.slice(0, 40),
|
|
177
|
+
`… ${t.steps.length - 60} steps …`,
|
|
178
|
+
...t.steps.slice(-20)
|
|
179
|
+
] : t.steps,
|
|
180
|
+
text: clip(t.text, 800)
|
|
181
|
+
})),
|
|
182
|
+
checks: a.checks.map((c) => ({
|
|
183
|
+
command: c.command,
|
|
184
|
+
exitCode: c.exitCode,
|
|
185
|
+
...c.exitCode !== 0 ? { output: c.output } : {}
|
|
186
|
+
})),
|
|
187
|
+
judgements: a.judgements,
|
|
188
|
+
files: a.files.slice(0, 50),
|
|
189
|
+
diff: head(a.patch, 3e3)
|
|
190
|
+
};
|
|
191
|
+
}
|
|
192
|
+
/**
|
|
193
|
+
* `compare_attempts`: where a failed attempt's steps first differ from a passing one's, a few
|
|
194
|
+
* steps either side, the checks, verdicts and files that differ, and the start of each diff.
|
|
195
|
+
*/
|
|
196
|
+
function divergence(failed, passed, window = 6) {
|
|
197
|
+
const a = failed.turns.flatMap((t) => t.steps);
|
|
198
|
+
const b = passed.turns.flatMap((t) => t.steps);
|
|
199
|
+
let same = 0;
|
|
200
|
+
while (same < a.length && same < b.length && a[same] === b[same]) same++;
|
|
201
|
+
const diverged = same < a.length || same < b.length;
|
|
202
|
+
const from = Math.max(0, same - 2);
|
|
203
|
+
const commands = [...new Set([...failed.checks, ...passed.checks].map((c) => c.command))];
|
|
204
|
+
const exit = (v, command) => v.checks.find((c) => c.command === command)?.exitCode ?? null;
|
|
205
|
+
const verdict = (v, question) => v.judgements.find((j) => j.question === question)?.verdict ?? null;
|
|
206
|
+
const questions = [...new Set([...failed.judgements, ...passed.judgements].map((j) => j.question))];
|
|
207
|
+
const last = (v) => clip(v.turns.at(-1)?.text ?? "", 600);
|
|
208
|
+
return {
|
|
209
|
+
failed: failed.id,
|
|
210
|
+
passed: passed.id,
|
|
211
|
+
error: failed.error ? `${failed.error.kind}: ${clip(failed.error.message, 600)}` : null,
|
|
212
|
+
stepsInCommon: same,
|
|
213
|
+
/** The first step (from 1) where they differ; null when the steps are the same. */
|
|
214
|
+
divergedAt: diverged ? same + 1 : null,
|
|
215
|
+
steps: {
|
|
216
|
+
failed: a.length,
|
|
217
|
+
passed: b.length
|
|
218
|
+
},
|
|
219
|
+
failedSteps: a.slice(from, same + window),
|
|
220
|
+
passedSteps: b.slice(from, same + window),
|
|
221
|
+
checks: commands.filter((c) => exit(failed, c) !== exit(passed, c)).map((c) => ({
|
|
222
|
+
command: c,
|
|
223
|
+
failed: exit(failed, c),
|
|
224
|
+
passed: exit(passed, c),
|
|
225
|
+
failedOutput: failed.checks.find((x) => x.command === c)?.output ?? null
|
|
226
|
+
})),
|
|
227
|
+
judgements: questions.filter((q) => verdict(failed, q) !== verdict(passed, q)).map((q) => ({
|
|
228
|
+
question: clip(q, 200),
|
|
229
|
+
failed: verdict(failed, q),
|
|
230
|
+
passed: verdict(passed, q)
|
|
231
|
+
})),
|
|
232
|
+
files: {
|
|
233
|
+
onlyFailed: failed.files.filter((f) => !passed.files.includes(f)).slice(0, 30),
|
|
234
|
+
onlyPassed: passed.files.filter((f) => !failed.files.includes(f)).slice(0, 30)
|
|
235
|
+
},
|
|
236
|
+
finalText: {
|
|
237
|
+
failed: last(failed),
|
|
238
|
+
passed: last(passed)
|
|
239
|
+
},
|
|
240
|
+
diff: {
|
|
241
|
+
failed: head(failed.patch, 1500),
|
|
242
|
+
passed: head(passed.patch, 1500)
|
|
243
|
+
}
|
|
244
|
+
};
|
|
245
|
+
}
|
|
246
|
+
//#endregion
|
|
247
|
+
export { stagedAttempts as a, listOutcomes as i, divergence as n, fromRecorded as r, attemptSummary as t };
|
|
248
|
+
|
|
249
|
+
//# sourceMappingURL=tools-B7SoW5kU.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"tools-B7SoW5kU.js","names":[],"sources":["../src/analysis/agent/tools.ts"],"mappings":";;;;;;;AA8BA,MAAM,aAAa;CACjB;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;AACF;;AAGA,SAAS,aAAa,OAAwB;CAC5C,IAAI,OAAO,UAAU,UAAU,OAAO,KAAK,MAAM,QAAQ,QAAQ,GAAG,GAAG,GAAG;CAC1E,IAAI,OAAO,UAAU,YAAY,UAAU,MAAM,OAAO;CACxD,MAAM,SAAS;CACf,KAAK,MAAM,OAAO,YAAY;EAC5B,MAAM,QAAQ,OAAO;EACrB,IAAI,OAAO,UAAU,YAAY,OAAO,OAAO,KAAK,MAAM,QAAQ,QAAQ,GAAG,GAAG,GAAG;EACnF,IAAI,MAAM,QAAQ,KAAK,KAAK,MAAM,OAAO,MAAM,OAAO,MAAM,QAAQ,GAClE,OAAO,KAAK,MAAM,KAAK,GAAG,GAAG,GAAG;CACpC;CACA,OAAO,KAAK,KAAK,UAAU,KAAK,GAAG,GAAG;AACxC;;AAGA,SAAgB,OAAO,OAAuC;CAC5D,QAAQ,MAAM,MAAd;EACE,KAAK,QAAQ;GACX,MAAM,QAAQ,aAAa,MAAM,KAAK;GACtC,OAAO,GAAG,MAAM,OAAO,QAAQ,KAAK,UAAU,KAAK,MAAM,KAAK,KAAK;EACrE;EACA,KAAK,aACH,OAAO,SAAS,MAAM,KAAK,KAAK,MAAM;EACxC,KAAK,cACH,OAAO,eAAe,KAAK,MAAM,OAAO,GAAG;EAC7C,KAAK,aACH,OAAO,cAAc,MAAM,MAAM,MAAM,KAAK,KAAK;EACnD,KAAK,SACH,OAAO,UAAU,KAAK,MAAM,SAAS,GAAG;EAC1C,KAAK,SACH,OAAO,UAAU,KAAK,MAAM,SAAS,GAAG;EAC1C,SACE,OAAO;CACX;AACF;AAEA,MAAM,SAAS,WAA8B,OAAO,SAAS,MAAM,OAAO,CAAC,KAAK,CAAC,CAAC;AAElF,MAAM,QAAQ,MAAc,QAAiB,KAAK,UAAU,MAAM,OAAO,IAAI,KAAK,MAAM,CAAC,GAAG;AAC5F,MAAM,QAAQ,MAAc,QAC1B,KAAK,UAAU,MAAM,OAAO,GAAG,KAAK,MAAM,GAAG,GAAG,EAAE,MAAM,KAAK,SAAS,IAAI;;AAG5E,SAAgB,WAAW,QAAoB,QAA+C;CAC5F,MAAM,UAAU,OAAO;CACvB,IAAI,CAAC,SAAS,OAAO;CACrB,MAAM,EAAE,SAAS,UAAU,aAAa,QAAQ,MAAM;CACtD,OAAO;EACL,IAAI,GAAG,QAAQ,KAAK,IAAI,GAAG,QAAQ;EACnC,MAAM,QAAQ,KAAK;EACnB,GAAG,QAAQ;EACX;EACA,OAAO,QAAQ;GAAE,MAAM,MAAM;GAAM,SAAS,MAAM;EAAQ,IAAI;EAC9D,SAAS,QAAQ,UAAU,CAAC,EAAA,CAAG,KAAK,OAAO;GACzC,SAAS,EAAE;GACX,UAAU,EAAE;GACZ,QAAQ,KAAK,GAAG,EAAE,OAAO,IAAI,EAAE,SAAS,KAAK,GAAG,GAAG;EACrD,EAAE;EACF,aAAa,QAAQ,cAAc,CAAC,EAAA,CAAG,KAAK,OAAO;GACjD,UAAU,EAAE;GACZ,SAAS,EAAE;GACX,OAAO,EAAE;EACX,EAAE;EACF,QAAQ,QAAQ,SAAS,CAAC,EAAA,CAAG,KAAK,OAAO;GACvC,QAAQ,EAAE;GACV,MAAM,EAAE;GACR,QAAQ,EAAE;GACV,OAAO,MAAM,EAAE,UAAU;EAC3B,EAAE;EACF,QAAQ,QAAQ,MAAM,SAAS,CAAC,EAAA,CAAG,KAAK,MAAM,EAAE,IAAI;EACpD,OAAO,QAAQ,MAAM,SAAS;CAChC;AACF;;AAGA,eAAsB,eACpB,YACA,SACwB;CACxB,MAAM,QAAuB,CAAC;CAC9B,KAAK,MAAM,UAAU,SAAS;EAC5B,MAAM,OAAO,KAAK,YAAY,SAAS,eAAe,OAAO,GAAG,CAAC;EACjE,IAAI,SAA4B;EAChC,IAAI;GACF,SAAS,KAAK,MAAM,MAAM,SAAS,MAAM,MAAM,CAAC;EAClD,QAAQ,CAER;EACA,MAAM,OAAO,WAAW,QAAQ,MAAM;EACtC,IAAI,MAAM,MAAM,KAAK,IAAI;CAC3B;CACA,OAAO;AACT;;AAGA,eAAsB,aAAa,QAAgB,SAA8C;CAC/F,MAAM,QAAQ,SACZ,OAAO,SAAS,KAAK,QAAQ,IAAI,GAAG,MAAM,CAAC,CAAC,YAAY,EAAE,IAAI,QAAQ,QAAQ,EAAE;CAClF,MAAM,QAAQ,gBAAgB,MAAM,KAAK,QAAQ,MAAM,YAAY,IAAI,CAAC;CACxE,OAAO;EACL,IAAI,QAAQ;EACZ,MAAM,QAAQ;EACd,GAAG,QAAQ;EACX,SAAS,QAAQ;EACjB,OAAO,QAAQ,QAAQ;GAAE,MAAM,QAAQ,MAAM;GAAM,SAAS,QAAQ,MAAM;EAAQ,IAAI;EACtF,QAAQ,MAAM,QAAQ,IACpB,QAAQ,OAAO,IAAI,OAAO,OAAO;GAC/B,SAAS,EAAE;GACX,UAAU,EAAE;GACZ,QAAQ,MAAM,MAAM,KAAK,EAAE,KAAK,IAAI,EAAA,CAAG,KAAK,GAAG,GAAG;EACpD,EAAE,CACJ;EACA,YAAY,QAAQ,WAAW,KAAK,OAAO;GACzC,UAAU,EAAE;GACZ,SAAS,EAAE;GACX,OAAO,EAAE;EACX,EAAE;EACF,OAAO,QAAQ,MAAM,KAAK,GAAG,OAAO;GAClC,QAAQ,EAAE;GACV,MAAM,EAAE;GACR,QAAQ,EAAE;GACV,OAAO,MACL,MAAM,QACH,MACC,EAAE,SAAS,KAAK,EAAE,SAAS,YAAY,EAAE,SAAS,UACtD,CACF;EACF,EAAE;EACF,OAAO,QAAQ,aAAa,KAAK,MAAM,EAAE,IAAI;EAC7C,OAAO,MAAM,KAAK,QAAQ,MAAM,MAAM,IAAI;CAC5C;AACF;;AAMA,SAAgB,eAAe,UAAuC;CACpE,MAAM,SAAS,SAAS,QAAQ,MAAM,EAAE,YAAY,MAAM;CAC1D,MAAM,QAAQ,OAAoB,GAAgB,MAChD,KAAK,IAAI,EAAE,IAAI,EAAE,CAAC,IAAI,KAAK,IAAI,EAAE,IAAI,EAAE,CAAC;CAC1C,OAAO,SACJ,QAAQ,MAAM,EAAE,YAAY,MAAM,CAAC,CACnC,SAAS,MAAM;EACd,MAAM,WAAW,OAAO,QAAQ,MAAM,EAAE,SAAS,EAAE,IAAI,CAAC,CAAC,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC;EACvE,MAAM,QAAQ,OAAO,QAAQ,MAAM,EAAE,SAAS,EAAE,IAAI,CAAC,CAAC,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC;EACpE,MAAM,QAAQ,YAAY;EAC1B,OAAO,QACH,CACE;GACE,QAAQ,EAAE;GACV,QAAQ,MAAM;GACd,OAAO,WAAY,SAAoB;EACzC,CACF,IACA,CAAC;CACP,CAAC;AACL;AAEA,MAAM,aAAa,MAAmB,EAAE,MAAM,QAAQ,GAAG,MAAM,IAAI,EAAE,MAAM,QAAQ,CAAC;;AAGpF,SAAgB,aAAa,UAAyB;CACpD,OAAO;EACL,UAAU,SAAS,KAAK,OAAO;GAC7B,IAAI,EAAE;GACN,SAAS,EAAE;GACX,OAAO,EAAE,QAAQ,GAAG,EAAE,MAAM,KAAK,IAAI,KAAK,EAAE,MAAM,SAAS,GAAG,MAAM;GACpE,cAAc,EAAE,OAAO,QAAQ,MAAM,EAAE,aAAa,CAAC,CAAC,CAAC,KAAK,MAAM,EAAE,OAAO;GAC3E,YAAY,EAAE,WAAW,KACtB,MAAM,GAAG,EAAE,QAAQ,GAAG,EAAE,MAAM,QAAQ,CAAC,EAAE,IAAI,KAAK,EAAE,UAAU,GAAG,GACpE;GACA,OAAO,EAAE,MAAM;GACf,OAAO,UAAU,CAAC;GAClB,OAAO,EAAE,MAAM;EACjB,EAAE;EACF,OAAO,eAAe,QAAQ;CAChC;AACF;;AAGA,SAAgB,eAAe,GAAgB;CAC7C,OAAO;EACL,IAAI,EAAE;EACN,SAAS,EAAE;EACX,OAAO,EAAE,QAAQ;GAAE,MAAM,EAAE,MAAM;GAAM,SAAS,KAAK,EAAE,MAAM,SAAS,GAAI;EAAE,IAAI;EAChF,OAAO,EAAE,MAAM,KAAK,GAAG,OAAO;GAC5B,MAAM,IAAI;GACV,QAAQ,KAAK,EAAE,QAAQ,GAAG;GAC1B,QAAQ,EAAE;GACV,OACE,EAAE,MAAM,SAAS,KACb;IAAC,GAAG,EAAE,MAAM,MAAM,GAAG,EAAE;IAAG,KAAK,EAAE,MAAM,SAAS,GAAG;IAAW,GAAG,EAAE,MAAM,MAAM,GAAG;GAAC,IACnF,EAAE;GACR,MAAM,KAAK,EAAE,MAAM,GAAG;EACxB,EAAE;EACF,QAAQ,EAAE,OAAO,KAAK,OAAO;GAC3B,SAAS,EAAE;GACX,UAAU,EAAE;GACZ,GAAI,EAAE,aAAa,IAAI,EAAE,QAAQ,EAAE,OAAO,IAAI,CAAC;EACjD,EAAE;EACF,YAAY,EAAE;EACd,OAAO,EAAE,MAAM,MAAM,GAAG,EAAE;EAC1B,MAAM,KAAK,EAAE,OAAO,GAAI;CAC1B;AACF;;;;;AAMA,SAAgB,WAAW,QAAqB,QAAqB,SAAS,GAAG;CAC/E,MAAM,IAAI,OAAO,MAAM,SAAS,MAAM,EAAE,KAAK;CAC7C,MAAM,IAAI,OAAO,MAAM,SAAS,MAAM,EAAE,KAAK;CAC7C,IAAI,OAAO;CACX,OAAO,OAAO,EAAE,UAAU,OAAO,EAAE,UAAU,EAAE,UAAU,EAAE,OAAO;CAClE,MAAM,WAAW,OAAO,EAAE,UAAU,OAAO,EAAE;CAC7C,MAAM,OAAO,KAAK,IAAI,GAAG,OAAO,CAAC;CACjC,MAAM,WAAW,CAAC,GAAG,IAAI,IAAI,CAAC,GAAG,OAAO,QAAQ,GAAG,OAAO,MAAM,CAAC,CAAC,KAAK,MAAM,EAAE,OAAO,CAAC,CAAC;CACxF,MAAM,QAAQ,GAAgB,YAC5B,EAAE,OAAO,MAAM,MAAM,EAAE,YAAY,OAAO,CAAC,EAAE,YAAY;CAC3D,MAAM,WAAW,GAAgB,aAC/B,EAAE,WAAW,MAAM,MAAM,EAAE,aAAa,QAAQ,CAAC,EAAE,WAAW;CAChE,MAAM,YAAY,CAChB,GAAG,IAAI,IAAI,CAAC,GAAG,OAAO,YAAY,GAAG,OAAO,UAAU,CAAC,CAAC,KAAK,MAAM,EAAE,QAAQ,CAAC,CAChF;CACA,MAAM,QAAQ,MAAmB,KAAK,EAAE,MAAM,GAAG,EAAE,CAAC,EAAE,QAAQ,IAAI,GAAG;CACrE,OAAO;EACL,QAAQ,OAAO;EACf,QAAQ,OAAO;EACf,OAAO,OAAO,QAAQ,GAAG,OAAO,MAAM,KAAK,IAAI,KAAK,OAAO,MAAM,SAAS,GAAG,MAAM;EACnF,eAAe;;EAEf,YAAY,WAAW,OAAO,IAAI;EAClC,OAAO;GAAE,QAAQ,EAAE;GAAQ,QAAQ,EAAE;EAAO;EAC5C,aAAa,EAAE,MAAM,MAAM,OAAO,MAAM;EACxC,aAAa,EAAE,MAAM,MAAM,OAAO,MAAM;EACxC,QAAQ,SACL,QAAQ,MAAM,KAAK,QAAQ,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,CAClD,KAAK,OAAO;GACX,SAAS;GACT,QAAQ,KAAK,QAAQ,CAAC;GACtB,QAAQ,KAAK,QAAQ,CAAC;GACtB,cAAc,OAAO,OAAO,MAAM,MAAM,EAAE,YAAY,CAAC,CAAC,EAAE,UAAU;EACtE,EAAE;EACJ,YAAY,UACT,QAAQ,MAAM,QAAQ,QAAQ,CAAC,MAAM,QAAQ,QAAQ,CAAC,CAAC,CAAC,CACxD,KAAK,OAAO;GACX,UAAU,KAAK,GAAG,GAAG;GACrB,QAAQ,QAAQ,QAAQ,CAAC;GACzB,QAAQ,QAAQ,QAAQ,CAAC;EAC3B,EAAE;EACJ,OAAO;GACL,YAAY,OAAO,MAAM,QAAQ,MAAM,CAAC,OAAO,MAAM,SAAS,CAAC,CAAC,CAAC,CAAC,MAAM,GAAG,EAAE;GAC7E,YAAY,OAAO,MAAM,QAAQ,MAAM,CAAC,OAAO,MAAM,SAAS,CAAC,CAAC,CAAC,CAAC,MAAM,GAAG,EAAE;EAC/E;EACA,WAAW;GAAE,QAAQ,KAAK,MAAM;GAAG,QAAQ,KAAK,MAAM;EAAE;EACxD,MAAM;GAAE,QAAQ,KAAK,OAAO,OAAO,IAAI;GAAG,QAAQ,KAAK,OAAO,OAAO,IAAI;EAAE;CAC7E;AACF"}
|