fix(import): --dir code extraction + --skip-enrich batch pass-through + AI timeout/logging by m0Nst3r873 · Pull Request #76 · Tencent/teamai-cli

m0Nst3r873 · 2026-06-30T11:16:05Z

Summary

Multiple import improvements and fixes from E2E testing:

1. `teamai import --dir <path>` now runs code extraction

Previously --dir scanned for markdown files and did AI classification. Now it runs the same code knowledge extraction pipeline as --from-repo (extractCodebase + graph aggregation), just without the clone step. This is an intentional breaking change — the old markdown-scan behavior is preserved under --from-claude for rule file migration.

teamai import --dir ./my-project --skip-enrich

2. Remove `--workspace` (superseded by `--dir .`)

--workspace only generated a narrative codebase.md via AI — no graph, no structured knowledge, required AI to be available. It is fully superseded by --dir . which produces a complete teamwiki knowledge graph deterministically. Removed entirely (not hidden).

3. `--skip-enrich` correctly passed through batch import

--skip-enrich was only applied to --from-repo (single). Now correctly threaded through:

import.ts → importFromRepoList → each importFromRepo
import.ts → importFromOrg → importFromRepoList

E2E result: 11 repos batch import in 11 seconds (was 300s+ timeout before).

4. AI timeout extended to 10 minutes

DEFAULT_TIMEOUT_MS changed from 120s to 600s. Large repos with many modules need more time for AI document generation. Comment documents applicable scenarios.

5. AI model name in logs

callClaude logs CLI name on first use and per-call (visible with --verbose)
generateCodebaseMd and enrichWithAI log which CLI model is being used
Exported getAICliName() for use by other modules

6. Push success hint with repo URL

All import paths now log after successful push:

✔ 已推送到团队知识仓库 (https://git.woa.com/team/repo.git)

7. Fix http-repo-integration test

Test was failing because detectProjectConfig() found a stale .teamai/config.yaml in cwd. Fixed by chdir to tmpDir.

Review feedback addressed

Issue	Fix
`--output` still wrote to team-repo	`--output` now writes to specified dir only, skips team-repo entirely
`fs.remove(srcWiki)` could delete user's existing `teamwiki/`	Extract output now goes to a tmpdir; source directory never touched
Dead code `opts.dir ?? '.'`	Removed (branch guarantees `opts.dir` is truthy)
CLI description still mentions "git workspace"	Updated to current actual sources
`--dir` path has no unit tests	Added 6 tests covering all paths
`DEFAULT_TIMEOUT_MS` has no rationale comment	Added detailed comment explaining applicable scenarios
`os` dynamic import unnecessary	Changed to static top-level import
Stale `--workspace` reference in codebase.ts error message	Updated to `--dir <path>`
`--dry-run` still ran full extractCodebase (up to 10min AI call)	Now early-returns before extraction, consistent with `--from-repo` behavior

E2E verification

Test	Result
T1-T4 (init/pull/hooks/recall)	✅
T5 import --from-repo --skip-enrich	✅ 4s, no AI warn
T6 import --from-org --skip-enrich (11 repos)	✅ 11s, 0 failures
T7 graph integrity (1179 nodes, 794 edges)	✅
T8 codebase --lint	✅
T9 doctor	✅
import --dir (local project)	✅ 10 nodes, 5 edges, global graph aggregated
import --dir --output /tmp/draft	✅ writes to output dir, source dir clean

Test plan

jeff-r2026

Code review (medium effort, /review)。两条实质问题（数据完整性 / 数据丢失）+ 两条清理项，已作为行内评论附在对应位置。--skip-enrich 全链路贯通、graph-index 路径一致性、getAICliName 引入均已核对，无可触发缺陷。

jeff-r2026 · 2026-06-30T11:42:54Z

+        await aggregateGlobalGraph(teamwikiRoot);
+
+        // 推送
+        if (!opts.output) {


--output 在 --dir 下不再生效，且仍会修改团队仓库。 整个「复制 evidence 到 team-repo + 聚合全局图谱」代码块只受 if (!opts.dryRun) 保护，--output 仅在这里用于跳过 git push。但其它分支中 --output 的语义是「写草稿到指定目录、不碰团队仓库」（见选项注释）。

结果：teamai import --dir ./proj --output /tmp/draft 既不会写入 /tmp/draft，又会直接覆盖写入 teamRepoPath/teamwiki/evidence/code/<slug> 和全局 graph-index.json（只是没 push）。建议在 --output 时把产物写到该目录、并跳过对团队仓库的写入与聚合。

jeff-r2026 · 2026-06-30T11:42:54Z

+            await fs.copy(srcGraph, path.join(destGraphDir, 'graph-index.json'), { overwrite: true });
          }
+          log.info(`teamwiki/ 知识图谱已更新: ${slug}`);
+          await fs.remove(srcWiki);


无条件 fs.remove(srcWiki) 可能删除被扫描目录中预先存在的 teamwiki/，造成数据丢失。 srcWiki = <dirPath>/teamwiki；extractCodebase 会写入并合并到该目录，这里又无条件整目录删除，未区分本次新建还是用户原有。

对一个本身维护 teamwiki/ 的项目（如团队知识仓库自身，或用过 team-wiki-codebase 的项目）执行 teamai import --dir .，会先污染既有 teamwiki/ 再整体删除，用户原有图谱不可恢复。建议提取产物落到临时目录，或仅删除本次新建的内容。

jeff-r2026 · 2026-06-30T11:42:55Z

-        codebaseOutputPath = path.join(repoPath, 'codebase.md');
+    } else if (opts.dir) {
+      // 分支 3：--dir <path>，代码知识提取（等同于 --from-repo 但跳过 clone）
+      const dirPath = path.resolve(opts.dir ?? '.');


死代码：本行位于 else if (opts.dir) 分支内，opts.dir 必为真值，?? '.' 永不触发。建议简化为 path.resolve(opts.dir)，否则会误导读者以为存在「--dir 缺省为当前目录」的逻辑。

jeff-r2026 · 2026-06-30T11:42:55Z

@@ -590,9 +590,8 @@ program
 program
  .command('import')
  .description('Import knowledge from local files, Claude/Cursor rules, git workspace, MRs, or iWiki')


描述仍写着已被本 PR 删除的 “git workspace” 导入来源。--workspace 已整体移除，--help 不应再宣传它。建议更新为 local directory / repo / MRs / iWiki 等当前实际来源。

… to batch + push success hint Three fixes: 1. `teamai import --dir <path>` now runs extractCodebase + graph aggregation (same as --from-repo) instead of the old markdown scan + AI classify flow. The old behavior is preserved under --from-claude for rule file migration. 2. --skip-enrich was not passed through the batch import chain (import.ts → importFromRepoList → importFromRepo). Now correctly threaded through all call sites including --from-org. 3. All import paths now log "已推送到团队知识仓库" after successful autoPushTeamRepo, so users know the knowledge is shared. Also fixes http-repo-integration test (chdir to tmpDir before push to avoid stale project config in cwd).

m0Nst3r873 force-pushed the main branch from 21c27d5 to 29578b3 Compare June 30, 2026 11:23

jeff-r2026 reviewed Jun 30, 2026

View reviewed changes

m0Nst3r873 force-pushed the main branch 4 times, most recently from e39a4ff to 262d3f9 Compare June 30, 2026 12:15

m0Nst3r873 force-pushed the main branch from 262d3f9 to 406f0e3 Compare June 30, 2026 12:25

jeff-r2026 merged commit 0bb9f58 into Tencent:main Jun 30, 2026
7 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix(import): --dir code extraction + --skip-enrich batch pass-through + AI timeout/logging#76

fix(import): --dir code extraction + --skip-enrich batch pass-through + AI timeout/logging#76
jeff-r2026 merged 1 commit into
Tencent:mainfrom
m0Nst3r873:main

m0Nst3r873 commented Jun 30, 2026 •

edited

Loading

Uh oh!

jeff-r2026 left a comment

Uh oh!

jeff-r2026 Jun 30, 2026

Uh oh!

jeff-r2026 Jun 30, 2026

Uh oh!

jeff-r2026 Jun 30, 2026

Uh oh!

jeff-r2026 Jun 30, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

m0Nst3r873 commented Jun 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

1. teamai import --dir <path> now runs code extraction

2. Remove --workspace (superseded by --dir .)

3. --skip-enrich correctly passed through batch import

4. AI timeout extended to 10 minutes

5. AI model name in logs

6. Push success hint with repo URL

7. Fix http-repo-integration test

Review feedback addressed

E2E verification

Test plan

Uh oh!

jeff-r2026 left a comment

Choose a reason for hiding this comment

Uh oh!

jeff-r2026 Jun 30, 2026

Choose a reason for hiding this comment

Uh oh!

jeff-r2026 Jun 30, 2026

Choose a reason for hiding this comment

Uh oh!

jeff-r2026 Jun 30, 2026

Choose a reason for hiding this comment

Uh oh!

jeff-r2026 Jun 30, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

m0Nst3r873 commented Jun 30, 2026 •

edited

Loading

1. `teamai import --dir <path>` now runs code extraction

2. Remove `--workspace` (superseded by `--dir .`)

3. `--skip-enrich` correctly passed through batch import