宜博 / lemongraph

The graphlemon planner Work Node workflow completed successfully on retry (the first attempt hit a transient connection error; the resume succeeded).

Output: a Benchmark/验收测试文档 (acceptance-test spec) for the goal "修复 #/skills 详情页不展示 Skill 真实内容" (fix the #/skills detail page not showing the real skill content). It defines 18 test scenarios across four areas:

1. src/skill_query.get_skill_content() unit tests (6 cases, including two path-traversal security cases against both URL input and DB-poisoned file_path)
2. Backend GET /api/skills/:skill_id/content endpoint tests (5 cases: happy path, missing skill, missing file-on-disk, encoded path-traversal, and regression on existing skill routes)
3. Frontend manual/automated checks (5 cases, notably that the new "Skill content" panel must render alongside — not replace — the existing "变更审计" panel, and content must be escaped via escapeHtml into a <pre>, no markdown/highlight libs)
4. Full regression gate (full src/ unittest suite must stay green; diff scope limited to skill_query.py, backend_server.py, frontend/app.js)

This benchmark is the planner's artifact for the Work Node; it now hands off to the Coder/Validator stage per graphlemon's normal Org Graph flow. No further action taken by me beyond running the node.