meta-knowledge-base-bootstrap
GitHub通过单一种子（URL/PDF/Git/文本）快速构建领域知识库。自动分类源类型，调用多搜索引擎进行摄入，将摘要持久化至记忆，并生成包含结果表的Excel索引文件。
Trigger Scenarios
Install
npx skills add opensquilla/opensquilla --skill meta-knowledge-base-bootstrap -g -y
SKILL.md
Frontmatter
{
"kind": "meta",
"name": "meta-knowledge-base-bootstrap",
"always": false,
"triggers": [
"搭建知识库",
"knowledge base",
"kb 启动",
"bootstrap kb"
],
"provenance": {
"origin": "opensquilla-original",
"license": "Apache-2.0"
},
"composition": {
"steps": [
{
"id": "classify",
"kind": "llm_classify",
"with": {
"text": "Inspect this user input and decide its primary source type.\n\nDecision rules:\n- URL → input contains an http\/https link to a webpage (not a PDF, not a git host).\n- PDF → input references a .pdf path or URL ending in .pdf.\n- GIT → input references github.com \/ gitlab \/ a .git URL \/ a local repo path.\n- TEXT → everything else (free-text topic, question, concept).\n\nInput:\n{{ inputs.user_message | xml_escape | truncate(400) }}\n"
},
"output_choices": [
"URL",
"PDF",
"GIT",
"TEXT"
]
},
{
"id": "ingest",
"kind": "skill_exec",
"skill": "multi-search-engine",
"depends_on": [
"classify"
]
},
{
"id": "memorize",
"kind": "tool_call",
"tool": "memory_save",
"tool_args": {
"mode": "append",
"path": "memory\/kb-bootstrap.md",
"content": "# KB Bootstrap: {{ inputs.user_message | xml_escape | truncate(80) }}\nClassifier verdict: {{ outputs.classify }}\nIngestion (multi-search-engine, JSON):\n{{ outputs.ingest | truncate(2000) }}\n"
},
"depends_on": [
"ingest"
]
},
{
"id": "index",
"with": {
"task": "Create a workbook 'kb-index.xlsx' with columns [Engine, Title, URL, Snippet]. Populate it from this multi-search-engine JSON output: {{ outputs.ingest | truncate(3000) }}"
},
"skill": "xlsx",
"depends_on": [
"ingest"
]
}
]
},
"description": "Bootstrap a domain knowledge base from a single seed (URL \/ PDF path \/ git repo \/ free-text topic): classify source → ingest with the right tool → persist to memory + xlsx index.",
"meta_priority": 40
}
Knowledge Base Bootstrap (Meta-Skill)
Seed a domain knowledge base in one turn. The pipeline classifies the seed
source type (URL / PDF / GIT / TEXT) and ingests it via the
multi-search-engine skill, then persists the report and produces an
index.
| step | kind | skill | what it does |
|---|---|---|---|
| classify | llm_classify |
— | label the seed as one of URL / PDF / GIT / TEXT |
| ingest | skill_exec |
multi-search-engine |
run a DuckDuckGo search (JSON to stdout) |
| memorize | tool_call |
— (memory_save) |
append the ingestion summary to memory |
| index | agent |
xlsx |
write kb-index.xlsx with the result table |
The classifier is currently informational only — the ingest step always calls
multi-search-engine. A previous design routedPDF → pdf-toolkitandGIT → github, but those branches were dropped when the DSL moved toskill_exec. A follow-up will reintroduce per-classification routing once the corresponding bundled skills also exposeentrypoint:manifests.
Fallback
If the meta-flow fails: run the classifier prompt manually, then invoke
the appropriate ingestion skill, then memory_save the result, then
create the xlsx index with openpyxl.
Version History
- 7f72a32 Current 2026-07-05 18:41
