@guandata/decidex 0.1.1 → 0.1.2

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.
Files changed (3) hide show
  1. package/README.md +6 -1
  2. package/package.json +1 -1
  3. package/src/cli.js +93 -3
package/README.md CHANGED
@@ -94,12 +94,14 @@ decidex session send SESSION_ID --message "继续拆解渠道" --wait --follow
94
94
  decidex session stop SESSION_ID
95
95
  decidex session artifacts list SESSION_ID
96
96
  decidex session artifacts download SESSION_ID --all --dir ./artifacts
97
+ decidex session evidence list SESSION_ID
98
+ decidex session evidence download SESSION_ID --all --dir ./evidence
97
99
  decidex run diagnose RUN_ID --session-id SESSION_ID --json
98
100
  ```
99
101
 
100
102
  `session send --wait` 要求消息立即启动;会话正忙时服务端会在入队前返回冲突,避免 CLI 假失败后重试造成重复入队。不带 `--wait` 时结果可能是 `kind=direct` 或 `kind=queued`。
101
103
 
102
- 下载默认拒绝覆盖已有文件,也拒绝通过符号链接写出目标目录;确认覆盖普通文件时加 `--overwrite`。`run diagnose` 未显式传 `--session-id` 时会使用最近 50 条本地 Run 映射;若本地配置不可写,远端 Run 仍会继续,CLI 会先输出 Session/Run ID,之后可用显式 ID 恢复诊断。当前后端未向用户 API 暴露 Runner 的 Pod termination reason 和资源指标,结果会将这两项标记为 `not_exposed`。
104
+ `session evidence` 通过服务端会话文件索引发现实际 `.boe` 文件;证据文件位于 `workspace/analysis/biz-ontology-evidence/`,不会混入普通产物列表。下载默认拒绝覆盖已有文件,也拒绝通过符号链接写出目标目录;确认覆盖普通文件时加 `--overwrite`。`run diagnose` 未显式传 `--session-id` 时会使用最近 50 条本地 Run 映射;若本地配置不可写,远端 Run 仍会继续,CLI 会先输出 Session/Run ID,之后可用显式 ID 恢复诊断。当前后端未向用户 API 暴露 Runner 的 Pod termination reason 和资源指标,结果会将这两项标记为 `not_exposed`。
103
105
 
104
106
  ## Skill
105
107
 
@@ -165,6 +167,7 @@ decidex eval runs list AGENT_ID
165
167
  decidex eval runs get AGENT_ID RUN_ID
166
168
  decidex eval runs wait AGENT_ID RUN_ID --follow --output result.json
167
169
  decidex eval runs watch evaluation-runs.json --parallel 3 --follow --output summary.json
170
+ decidex eval evidence download AGENT_ID RUN_ID --all --dir ./evidence
168
171
  ```
169
172
 
170
173
  `evaluation-runs.json`:
@@ -180,6 +183,8 @@ decidex eval runs watch evaluation-runs.json --parallel 3 --follow --output summ
180
183
 
181
184
  `watch` 只在状态或计数变化时向 stderr 输出进度,最终 JSON 保留每个 Run 的完整 Case 结果;任一 Run 失败、停止或监控出错时退出码为 1。
182
185
 
186
+ `eval evidence download` 按 Case 分目录下载评测会话中的 `.boe`。该命令要求服务端评测详情返回 `targetSessionId`;旧版服务端会给出明确升级提示。
187
+
183
188
  ## 组织
184
189
 
185
190
  ```bash
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@guandata/decidex",
3
- "version": "0.1.1",
3
+ "version": "0.1.2",
4
4
  "description": "CLI for building and operating Decidex ontologies, agents, skills, sessions, and evaluations.",
5
5
  "license": "UNLICENSED",
6
6
  "type": "module",
package/src/cli.js CHANGED
@@ -9,6 +9,7 @@ import { parseArgs } from "node:util";
9
9
 
10
10
  const COOKIE_NAME = "decidex_session";
11
11
  const DEFAULT_CONFIG_PATH = join(homedir(), ".config", "decidex", "config.json");
12
+ const BIZ_ONTOLOGY_EVIDENCE_DIR = "workspace/analysis/biz-ontology-evidence/";
12
13
 
13
14
  const HELP = `decidex - Decidex command line client
14
15
 
@@ -40,6 +41,8 @@ Usage:
40
41
  decidex session stop SESSION_ID
41
42
  decidex session artifacts list SESSION_ID
42
43
  decidex session artifacts download SESSION_ID --all --dir DIR [--overwrite]
44
+ decidex session evidence list SESSION_ID
45
+ decidex session evidence download SESSION_ID --all --dir DIR [--overwrite]
43
46
  decidex run diagnose RUN_ID [--session-id ID] [--agent-id ID] --json
44
47
 
45
48
  decidex skill list
@@ -57,6 +60,7 @@ Usage:
57
60
  decidex eval runs get AGENT_ID RUN_ID
58
61
  decidex eval runs wait AGENT_ID RUN_ID [--follow] [--output FILE] [--timeout SECONDS]
59
62
  decidex eval runs watch FILE [--parallel N] [--follow] [--output FILE] [--timeout SECONDS]
63
+ decidex eval evidence download AGENT_ID RUN_ID --all --dir DIR [--overwrite]
60
64
  decidex eval stop AGENT_ID RUN_ID
61
65
 
62
66
  decidex ontology list
@@ -96,12 +100,16 @@ to stderr, so stdout or --output remains machine-readable.`,
96
100
  decidex session send SESSION_ID (--message TEXT | --input FILE) [--wait] [--follow]
97
101
  decidex session stop SESSION_ID
98
102
  decidex session artifacts ...
103
+ decidex session evidence ...
99
104
 
100
105
  With --wait, the message must start immediately; a busy session returns a conflict instead
101
106
  of queuing an untrackable run. Without --wait, the result can be kind=direct or kind=queued.`,
102
107
  "session artifacts": `Usage:
103
108
  decidex session artifacts list SESSION_ID
104
109
  decidex session artifacts download SESSION_ID --all --dir DIR [--overwrite]`,
110
+ "session evidence": `Usage:
111
+ decidex session evidence list SESSION_ID
112
+ decidex session evidence download SESSION_ID --all --dir DIR [--overwrite]`,
105
113
  "run diagnose": `Usage:
106
114
  decidex run diagnose RUN_ID [--session-id ID] [--agent-id ID] --json
107
115
 
@@ -117,6 +125,7 @@ Runtime Pod termination details require backend support.`,
117
125
  decidex eval cases list|export|plan|apply ...
118
126
  decidex eval run AGENT_ID FILE [--cases all] [--wait] [--follow] [--output FILE]
119
127
  decidex eval runs list|get|wait|watch ...
128
+ decidex eval evidence download AGENT_ID RUN_ID --all --dir DIR [--overwrite]
120
129
  decidex eval stop AGENT_ID RUN_ID`,
121
130
  "eval cases": `Usage:
122
131
  decidex eval cases list AGENT_ID
@@ -128,6 +137,8 @@ Runtime Pod termination details require backend support.`,
128
137
  decidex eval runs get AGENT_ID RUN_ID
129
138
  decidex eval runs wait AGENT_ID RUN_ID [--follow] [--output FILE] [--timeout SECONDS]
130
139
  decidex eval runs watch FILE [--parallel N] [--follow] [--output FILE] [--timeout SECONDS]`,
140
+ "eval evidence": `Usage:
141
+ decidex eval evidence download AGENT_ID RUN_ID --all --dir DIR [--overwrite]`,
131
142
  ontology: `Usage:
132
143
  decidex ontology list|get|export|plan|apply ...`,
133
144
  api: `Usage:
@@ -445,13 +456,25 @@ async function listSessionArtifacts(agentId, sessionId, values) {
445
456
  return (result.files ?? []).filter((file) => file.relativePath?.startsWith("artifacts/"));
446
457
  }
447
458
 
448
- function safeDestination(root, artifactPath) {
449
- const destination = resolve(root, artifactPath.replace(/^artifacts\//, ""));
459
+ async function listSessionEvidence(agentId, sessionId, values) {
460
+ const result = await request(`${sessionPath(agentId, sessionId)}/files?force=1&evidence=1`, { values });
461
+ return (result.files ?? []).filter((file) => {
462
+ const name = file.relativePath?.slice(BIZ_ONTOLOGY_EVIDENCE_DIR.length);
463
+ return file.relativePath?.startsWith(BIZ_ONTOLOGY_EVIDENCE_DIR) && name && !name.includes("/") && name.toLowerCase().endsWith(".boe");
464
+ });
465
+ }
466
+
467
+ function safeRelativeDestination(root, outputPath) {
468
+ const destination = resolve(root, outputPath);
450
469
  const child = relative(root, destination);
451
- if (!child || child.startsWith("..") || isAbsolute(child)) throw new Error(`非法产物路径:${artifactPath}`);
470
+ if (!child || child.startsWith("..") || isAbsolute(child)) throw new Error(`非法输出路径:${outputPath}`);
452
471
  return destination;
453
472
  }
454
473
 
474
+ function safeDestination(root, artifactPath) {
475
+ return safeRelativeDestination(root, artifactPath.replace(/^artifacts\//, ""));
476
+ }
477
+
455
478
  async function assertSafeArtifactTarget(root, destination, overwrite) {
456
479
  let current = root;
457
480
  for (const part of relative(root, dirname(destination)).split(sep).filter((item) => item && item !== ".")) {
@@ -818,6 +841,35 @@ async function sessionCommand(action, args) {
818
841
  if (values.query) query.set("q", values.query);
819
842
  return print(await request(`/api/agent-sessions?${query}`, { values }));
820
843
  }
844
+ if (action === "evidence") {
845
+ const [evidenceAction, sessionIdValue] = positionals;
846
+ expectPositionals(positionals, 2, `session evidence ${evidenceAction || "COMMAND"} SESSION_ID`);
847
+ const sessionId = required(sessionIdValue, "SESSION_ID");
848
+ const session = await resolveSession(sessionId, values);
849
+ const evidence = await listSessionEvidence(session.agentId, sessionId, values);
850
+ if (evidenceAction === "list") {
851
+ return print({ schemaVersion: 1, sessionId, agentId: session.agentId, evidence });
852
+ }
853
+ if (evidenceAction === "download") {
854
+ if (!values.all) throw new Error("evidence download 需要 --all");
855
+ const directory = resolve(required(values.dir, "--dir DIR"));
856
+ await mkdir(directory, { recursive: true });
857
+ const root = await realpath(directory);
858
+ const targets = evidence.map((item) => ({ item, destination: safeRelativeDestination(root, item.name) }));
859
+ for (const { destination } of targets) await assertSafeArtifactTarget(root, destination, values.overwrite);
860
+ const downloaded = [];
861
+ for (const { item, destination } of targets) {
862
+ const bytes = await request(`${sessionPath(session.agentId, sessionId)}/files/content?path=${encodeURIComponent(item.relativePath)}&download=1`, {
863
+ returnBytes: true,
864
+ values,
865
+ });
866
+ await writeArtifact(root, destination, bytes, values.overwrite);
867
+ downloaded.push({ ...item, path: destination });
868
+ }
869
+ return print({ schemaVersion: 1, sessionId, directory, downloaded });
870
+ }
871
+ throw new Error(`未知 session evidence 命令:${evidenceAction || "(空)"}`);
872
+ }
821
873
  if (action !== "artifacts") {
822
874
  expectPositionals(positionals, 1, `session ${action || "COMMAND"} SESSION_ID`);
823
875
  const sessionId = required(positionals[0], "SESSION_ID");
@@ -1156,9 +1208,47 @@ async function evalRunsCommand(args) {
1156
1208
  throw new Error(`未知 eval runs 命令:${action || "(空)"}`);
1157
1209
  }
1158
1210
 
1211
+ async function evalEvidenceCommand(args) {
1212
+ const { values, positionals } = commandArgs(args, {
1213
+ all: { type: "boolean", default: false },
1214
+ dir: { type: "string" },
1215
+ overwrite: { type: "boolean", default: false },
1216
+ });
1217
+ const [action, agentId, runId] = positionals;
1218
+ if (action !== "download") throw new Error(`未知 eval evidence 命令:${action || "(空)"}`);
1219
+ expectPositionals(positionals, 3, "eval evidence download AGENT_ID RUN_ID");
1220
+ if (!values.all) throw new Error("eval evidence download 需要 --all");
1221
+ const result = await request(evaluationRunPath(agentId, runId), { values });
1222
+ const runCases = result.run?.cases ?? [];
1223
+ if (runCases.some((runCase) => !Object.hasOwn(runCase, "targetSessionId"))) {
1224
+ throw new Error("Decidex 服务端未返回评测 Case 的 targetSessionId;请先升级服务端");
1225
+ }
1226
+ const evidence = (await Promise.all(runCases.map(async (runCase, index) => {
1227
+ if (!runCase.targetSessionId) return [];
1228
+ const files = await listSessionEvidence(agentId, runCase.targetSessionId, values);
1229
+ return files.map((file) => ({ ...file, caseId: runCase.id, casePosition: runCase.position, sessionId: runCase.targetSessionId, outputDirectory: `case-${index + 1}` }));
1230
+ }))).flat();
1231
+ const directory = resolve(required(values.dir, "--dir DIR"));
1232
+ await mkdir(directory, { recursive: true });
1233
+ const root = await realpath(directory);
1234
+ const targets = evidence.map((item) => ({ item, destination: safeRelativeDestination(root, join(item.outputDirectory, item.name)) }));
1235
+ for (const { destination } of targets) await assertSafeArtifactTarget(root, destination, values.overwrite);
1236
+ const downloaded = [];
1237
+ for (const { item, destination } of targets) {
1238
+ const bytes = await request(`${sessionPath(agentId, item.sessionId)}/files/content?path=${encodeURIComponent(item.relativePath)}&download=1`, {
1239
+ returnBytes: true,
1240
+ values,
1241
+ });
1242
+ await writeArtifact(root, destination, bytes, values.overwrite);
1243
+ downloaded.push({ caseId: item.caseId, casePosition: item.casePosition, sessionId: item.sessionId, name: item.name, relativePath: item.relativePath, path: destination });
1244
+ }
1245
+ return print({ schemaVersion: 1, agentId, runId, directory, downloaded });
1246
+ }
1247
+
1159
1248
  async function evalCommand(action, args) {
1160
1249
  if (action === "cases") return evalCasesCommand(args);
1161
1250
  if (action === "runs") return evalRunsCommand(args);
1251
+ if (action === "evidence") return evalEvidenceCommand(args);
1162
1252
  const { values, positionals } = commandArgs(args, {
1163
1253
  cases: { type: "string" },
1164
1254
  follow: { type: "boolean", default: false },