| name | wiki-ingest-llm |
| description | Ingest sources into the wiki using external LLM extraction (faster batch processing). Fully automated - fetches, extracts, and writes wiki pages in one command. |
| allowed-tools | ["Bash","Read","Write","Edit"] |
Wiki Ingest LLM Skill
Ingest one or more sources into the wiki using external OpenAI LLM calls for extraction. Fully automated pipeline - fetches sources, extracts entities, and writes all wiki pages.
Usage
/wiki-ingest-llm <source>
/wiki-ingest-llm wiki/raw/paper.pdf
/wiki-ingest-llm https://example.com/article
/wiki-ingest-llm wiki/raw/article.md
# arXiv papers
/wiki-ingest-llm 2409.05591
/wiki-ingest-llm arxiv:2409.05591
/wiki-ingest-llm https://arxiv.org/abs/2409.05591
# Batch mode
/wiki-ingest-llm wiki/raw/paper1.pdf wiki/raw/paper2.pdf https://example.com
/wiki-ingest-llm wiki/raw/*.pdf --parallel 5
Workflow
Step 1: Parse arguments
Determine source type from the input:
- PDF file: Path ending in
.pdf → use pdf_reader.py
- arXiv ID:
2409.05591 or arxiv:2409.05591 → use DeepXiv API
- Web URL:
https://... → use web_fetcher.py
- Bilibili video:
https://bilibili.com/video/... → use bilibili_fetcher.py
- Markdown file: Path ending in
.md → read directly
Step 2: Run the ingest script
uv run python .claude/skills/wiki-ingest-llm/bin/wiki_ingest_llm.py <source>
uv run python .claude/skills/wiki-ingest-llm/bin/wiki_ingest_llm.py <source1> <source2> --parallel 5
uv run python .claude/skills/wiki-ingest-llm/bin/wiki_ingest_llm.py <source> --model gpt-4o
Step 3: Report results
Tell the user:
- Number of sources processed
- Entities created vs updated
- Any errors encountered
CLI Flags
| Flag | Default | Description |
|---|
--parallel, -p | 10 | Max parallel workers |
--model, -m | gpt-4o-mini | LLM model for extraction |
--cache | wiki/cache.md | Path to wiki cache.md |
--no-write | - | Skip writing, only output JSON |
Supported Source Types
| Type | Example | Fetcher | Saved to wiki/raw |
|---|
| PDF file | wiki/raw/paper.pdf | pdf_reader.py (精准API) | {title}.md |
| arXiv paper | 2409.05591 | DeepXiv API | arxiv-{id}.md |
| Web URL | https://example.com | web_fetcher.py | {title}.md |
| Bilibili video | https://bilibili.com/video/... | bilibili_fetcher.py | {title}.md |
| Markdown file | wiki/raw/article.md | Direct read | {filename}.md |
All sources are automatically saved to wiki/raw/ for cross-referencing.
Error Handling
If the script fails:
- Check if Configuration is configured
uv run python .claude/shared/bin/wiki_config.py status
- if pdf is parsed failed, try agent mode
v run python .claude/shared/bin//pdf_reader.py "small.pdf" --agent