Install any skill in seconds. Free to start, no credit card required.
Get Started Free →将 PDF/PPT/Excel/Word 等多格式文档解析为结构化 Markdown,并输出元数据与解析置信度,作为 RAG 与四色卡片的数据底座。
.claude/skills/anbeime-antinet-doc-parse/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 49% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -34% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -52% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -42% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -5% | 0% |
file_path:已通过 security-scan 的本地文件路径formats:(可选)期望支持的格式白名单,默认全格式markdown:结构化 Markdown 正文metadata:标题、页数、表格数、作者等元数据confidence:0–1 解析置信度fallback_used:最终生效的解析器名称python-magic(类型探测)人工介入,不输出残缺结果,回传 BLOCKED 给军机处。scripts/run_doc_parse.pycore.runtime.AgentSession.run_stage("doc-parse"),调用 archive.mijuanfang.MiJuanFangAgent(三级解析 fallback,纯 Python 可离线)。python skills/doc-parse/scripts/run_doc_parse.pyexamples/snse_survey/skill_outputs/doc_parse.json(解析结果 + 置信度 + fallback 信息)。| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 44,682 | 17,847 | -60% | 1 | 1 | 0% | 1,825 | 2,721 | +49% | 0 | 0 | — |
case-02 | fail→pass | 20,748 | 13,983 | -33% | 1 | 1 | 0% | 2,935 | 1,949 | -34% | 0 | 0 | — |
case-03 | pass→pass | 21,054 | 13,173 | -37% | 1 | 1 | 0% | 2,443 | 1,790 | -27% | 0 | 0 | — |
case-09 | pass→pass | 17,284 | 7,069 | -59% | 1 | 1 | 0% | 2,068 | 826 | -60% | 0 | 0 | — |
case-04 | fail→fail | 39,109 | 31,008 | -21% | 1 | 1 | 0% | 4,274 | 5,105 | +19% | 0 | 0 | — |
case-05 | fail→fail | 24,591 | 24,245 | -1% | 1 | 1 | 0% | 3,918 | 4,373 | +12% | 0 | 0 | — |
case-06 | fail→pass | 20,121 | 8,347 | -59% | 1 | 1 | 0% | 2,172 | 1,042 | -52% | 0 | 0 | — |
case-07 | pass→pass | 18,539 | 7,644 | -59% | 1 | 1 | 0% | 2,108 | 938 | -56% | 0 | 0 | — |
case-08 | pass→pass | 15,961 | 7,970 | -50% | 1 | 1 | 0% | 1,665 | 978 | -41% | 0 | 0 | — |
case-10 | fail→pass | 19,276 | 23,513 | +22% | 1 | 1 | 0% | 2,352 | 1,366 | -42% | 0 | 0 | — |
case-11 | fail→pass | 30,396 | 24,921 | -18% | 1 | 1 | 0% | 3,987 | 3,790 | -5% | 0 | 0 | — |
case-12 | fail→pass | 25,892 | 7,330 | -72% | 1 | 1 | 0% | 1,253 | 875 | -30% | 0 | 0 | — |
case-13 | fail→pass | 14,423 | 6,757 | -53% | 1 | 1 | 0% | 1,424 | 869 | -39% | 0 | 0 | — |
case-14 | fail→pass | 19,089 | 7,160 | -62% | 1 | 1 | 0% | 2,202 | 786 | -64% | 0 | 0 | — |
case-20 | pass→pass | 13,887 | 1,971 | -86% | 1 | 1 | 0% | 2,253 | 784 | -65% | 0 | 0 | — |
case-15 | pass→pass | 17,163 | 8,315 | -52% | 1 | 1 | 0% | 2,244 | 1,036 | -54% | 0 | 0 | — |
case-16 | fail→pass | 17,166 | 8,635 | -50% | 1 | 1 | 0% | 2,001 | 1,109 | -45% | 0 | 0 | — |
case-17 | pass→pass | 11,857 | 7,226 | -39% | 1 | 1 | 0% | 2,073 | 826 | -60% | 0 | 0 | — |
case-18 | pass→pass | 17,291 | 8,202 | -53% | 1 | 1 | 0% | 1,900 | 929 | -51% | 0 | 0 | — |
case-19 | pass→pass | 25,229 | 16,522 | -35% | 1 | 1 | 0% | 2,858 | 2,921 | +2% | 0 | 0 | — |
case-21 | fail→pass | 12,544 | 2,173 | -83% | 1 | 1 | 0% | 2,037 | 821 | -60% | 0 | 0 | — |
case-22 | fail→pass | 10,994 | 8,769 | -20% | 1 | 1 | 0% | 1,901 | 2,029 | +7% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +50 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.