Skip to content

fix(runner): seed clean workspace task brief - #179

Open
RerankerGuo wants to merge 1 commit into
TencentCloud:mainfrom
RerankerGuo:fix/l1-extractor-explicit-prompt
Open

fix(runner): seed clean workspace task brief#179
RerankerGuo wants to merge 1 commit into
TencentCloud:mainfrom
RerankerGuo:fix/l1-extractor-explicit-prompt

Conversation

@RerankerGuo

Copy link
Copy Markdown

Description | 描述

Seed the default CleanContextRunner workspace with small AGENTS.md and SOUL.md task brief files so OpenClaw lightweight embedded runs no longer inject missing 76-byte stubs. The runner still passes the L1 extraction system prompt and user prompt explicitly, while the clean workspace remains low-context and stable for plugin cache behavior.

Related Issue | 关联 Issue

Fix #176

Change Type | 修改类型

  • Bug fix | Bug 修复
  • New feature | 新功能
  • Documentation update | 文档更新
  • Code optimization | 代码优化

Self-test Checklist | 自测清单

  • Verified locally | 本地验证通过
  • No existing features affected | 无影响现有功能

Additional Notes | 其他说明

Verified with npm test and npm run build using Node v24.15.0.

@Maxwell-Code07

Copy link
Copy Markdown
Collaborator

This directly addresses the clean workspace stub issue reported earlier. Seeding with AGENTS.md and SOUL.md is a practical fix that keeps the workspace low-context. We'll review it soon.

@ikelvingo

Copy link
Copy Markdown

Verified locally on Hermes 0.16.0 with the TencentDB-Agent-Memory plugin.

We hit the same #176 symptom — L1 extractor returning empty content / EMPTY_DUMP on a 147-session cold start seed. Since this PR was still open, I applied the fix from this PR as a local patch to our plugin install (file changes: +27/-3 src + 59 lines test, matches this PR diff).

After the patch, L1 extraction behavior changed completely:

  • Sessions processed: 126/147 (in progress)
  • L1 records extracted: 92, with 0 records showing empty content (0.0%)
  • Continuous writes to vectors.db, no EMPTY_DUMP failures
  • L1 cursor advancing normally through imported historical sessions

The fix works. Recommend merging.

@RerankerGuo
RerankerGuo force-pushed the fix/l1-extractor-explicit-prompt branch from 3bca6ed to 0e0a844 Compare July 8, 2026 06:33
@RerankerGuo

Copy link
Copy Markdown
Author

Hi @Maxwell-Code07 — gentle ping on #179.

You earlier confirmed the clean-workspace seed approach for #176 was the right fix ("seeding with AGENTS.md and SOUL.md is a practical fix that keeps the workspace low-context"). Since then, @ikelvingo reported independently verifying this patch locally on Hermes 0.16.0 with 92 L1 records and 0 empty-content records, and explicitly recommended merging.

I rebased onto current main (4339e63) — CI green (Install / Manifest / Pack / Size Guard all success on 0e0a844). Anything blocking, or is there a maintainer-side ETA?

@YOMXXX YOMXXX left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Approved as the canonical fix for #176.

Local verification:

  • COREPACK_ENABLE_AUTO_PIN=0 pnpm test (68/68 passed)
  • COREPACK_ENABLE_AUTO_PIN=0 pnpm build
  • git diff --check

This seeds the clean workspace with compact AGENTS.md/SOUL.md task briefs while preserving the explicit prompt path for L1 extraction and other text-only runner calls.

Signed-off-by: Ziyang Guo <121015044+RerankerGuo@users.noreply.github.com>
@RerankerGuo
RerankerGuo force-pushed the fix/l1-extractor-explicit-prompt branch from 0e0a844 to 3729f15 Compare August 12, 2026 02:17
@RerankerGuo

Copy link
Copy Markdown
Author

Maintenance update: rebased onto current main (98d6878) as single signed commit 3729f15. Revalidated locally: focused regression test 1/1 passed, full suite 68/68 passed, pnpm build passed, and git diff --check is clean. Refreshed GitHub CI is 4/4 green.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[Bug] l1-extractor 用 clean-workspace + 76 字节 stub context, LLM 返回空字符串

4 participants