Read project documents with an agent
Extract relevant passages with source locations, maintain a local document cache and keep reading separate from code changes or optional external analysis.
Read project documents with an agent to locate a requirement without treating an entire folder as unquestionable instructions. The document-reader skill indexes supported files, extracts relevant sections and retains references to their locations. A local cache can support later queries, but it is not automatically shared memory across unrelated agents.
A request to implement requirements can lead the host agent into code changes. State a read-only boundary when you only want analysis. Local extraction is separate from optional provider API analysis, which needs an explicit request and provider choice. Cached files can contain full document text and deserve the same protection as their sources.
What it does
- Search selected documents or sections.
- Keep source locations beside extracted claims.
- Recognise changed files and stale cached content.
- Distinguish review comments from settled requirements.
- Read scans through separately configured OCR.
- Report unread images, missing tables and truncation.
Install and use
- Provide the complete package, Python 3.10 or newer and an authorised folder. Excel needs openpyxl; PDF needs pypdf. Ruled PDF tables and OCR require separate optional tools, including pdfplumber or Tesseract with the appropriate language data.
- For example, request the session-expiry requirement with its exact source location, explicitly prohibiting changes to code, source files and configuration.
- Check extracted passages against the originals. Review OCR numbers and dates, and do not assume cached spreadsheet formula values are current.
- For optional API analysis, choose the provider and costs separately, inspect the outbound dry run and keep credentials in environment variables. Decide whether the cache remains local before sharing anything.
FAQ
Source
Made by Van Nguyen. Original skill: document-reader, published on GitHub under the MIT licence.
View on GitHub