datasheet-intelligence

Original：🇺🇸 English
Translated
3 scripts

Convert mixed-format datasheets and hardware reference files (PDF, DOCX, HTML, Markdown, XLSX/CSV) into normalized Markdown knowledge files for AI coding agents. Use when a user asks to ingest datasheets, register maps, pinout/timing sheets, revision histories, or internal hardware notes before searching datasheet content or generating code. Produce RAG-ready section chunks, anchors, image references, and metadata under .context/knowledge.

10installs
Sourcedhkimxx/ai-agent-skills
Added on

NPX Install

npx skill4agent add dhkimxx/ai-agent-skills datasheet-intelligence

Tags

Translated version includes tags in frontmatter
rag-knowledge-basedatasheet-ingestionmulti-format-conversionai-agent-toolsdocling-utilities

SKILL.md Content

View Translation Comparison →

Datasheet Intelligence

Purpose

Ingest multi-format datasheets into a single Markdown-first knowledge base so agents can answer hardware questions using consistent structure.

Prerequisites

Check 
uv
first:
bash
uv --version
If 
uv
is not installed, install it by OS:
sh
# macOS (Homebrew)
brew install uv

# Linux (official installer)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Windows (WinGet)
winget install --id=astral-sh.uv -e
After installation, restart the shell and verify 
uv --version
again.

Workflow

  1. Place source datasheets in a folder (for example 
    docs/datasheets/
    ).
  2. Run 
    scripts/ingest_docs.py
    with 
    uv run --with docling python3
    .
  3. Read outputs from 
    .context/knowledge/
    : 
    • <doc>/<doc>.md
      : normalized markdown
    • <doc>/<doc>.sections.jsonl
      : section chunks for retrieval
    • <doc>/<doc>.tables.md
      : table-focused markdown
    • <doc>/<doc>.meta.json
      : conversion metadata and validation info
    • <doc>/<doc>.docling.json
      : raw Docling structured export
    • <doc>/_images/*
      : extracted images
    • knowledge.index.json
      : corpus manifest
  4. Use search/read helpers: 
    • scripts/search_docs.py
      for corpus search
    • scripts/read_docs.py
      for focused section reads

Commands

bash
# Ingest all supported datasheets from a directory
uv run --with docling python3 scripts/ingest_docs.py docs/datasheets --output-dir .context/knowledge

# Ingest only top-level files and skip OCR for faster runs
uv run --with docling python3 scripts/ingest_docs.py docs/datasheets --non-recursive --no-ocr

# Search and focused read
uv run python3 scripts/search_docs.py "SPI0 address" --knowledge-dir .context/knowledge
uv run python3 scripts/read_docs.py exynos_spi_v1 --anchor section-4-2

Operational Rules

  • Use command-first guidance (above) instead of low-level converter internals.
  • Prefer 
    uv run
    commands and do not assume a global 
    python
    alias.
  • Keep output location at 
    .context/knowledge
    unless user asks otherwise.
  • Before the first run, ensure generated artifacts are ignored by git (
    .context/
    or 
    .context/knowledge/
    in 
    .gitignore
    ).
  • Preserve image presence in markdown with 
    ![image_ref](path)
    entries.
  • Keep section anchors and normalize textual references like 
    See Table 4.2
    into markdown links when anchor targets exist.
  • Run on all provided files and summarize failures without stopping the whole batch unless explicitly requested.
  • For detailed CLI options and execution presets, read 
    references/execution-options.md
    .
  • If 
    uv
    is unavailable, follow the fallback instructions in 
    references/execution-options.md
    .

Resources

  • scripts/ingest_docs.py
    : main ingestion pipeline
  • scripts/search_docs.py
    : chunk-level search helper
  • scripts/read_docs.py
    : focused reader helper
  • references/output-contract.md
    : output schema and retrieval contract
  • references/execution-options.md
    : detailed runtime flags and command presets