By name: meta-knowledge-base-bootstrap 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." kind: meta meta_priority: 40 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 output_choices: [URL, PDF, GIT, TEXT] with: text: | Inspect this user input and decide its primary source type. Decision rules: - URL → input contains an http/https link to a webpage (not a PDF, not a git host). - PDF → input references a .pdf path or URL ending in .pdf. - GIT → input references github.com / gitlab / a .git URL / a local repo path. - TEXT → everything else (free-text topic, question, concept). Input: {{ inputs.user_message | xml_escape | truncate(400) }}
- id: ingest kind: skill_exec skill: multi-search-engine depends_on: [classify]
- id: memorize kind: tool_call tool: memory_save depends_on: [ingest] tool_args: content: | # KB Bootstrap: {{ inputs.user_message | xml_escape | truncate(80) }} Classifier verdict: {{ outputs.classify }} Ingestion (multi-search-engine, JSON): {{ outputs.ingest | truncate(2000) }} path: "memory/kb-bootstrap.md" mode: "append"
- id: index skill: xlsx depends_on: [ingest] 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) }}"
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.