long-document-evidence-reader

Name: long-document-evidence-reader
Availability: InStock
Author: AllanYiin

Extract traceable evidence from long documents via chunked REPL queries.

2|1|Updated Feb 1, 2026

Search Tags:#pdf #retrieval #repl #evidence #bm25 #long-document

Ready to run

npx skills add https://github.com/AllanYiin/Amon --skill long-document-evidence-reader

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: long-document-evidence-reader
Source: https://github.com/AllanYiin/Amon/tree/main/src/amon/resources/skills/long-document-evidence-reader
Command: npx skills add https://github.com/AllanYiin/Amon --skill long-document-evidence-reader

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires pypdf, PyPDF2, and includes scripts (resource) and references (resource) components.

What problem does it solve?

Long documents contain many pages and sources; this skill provides a structured way to read, search, and cite evidence without overwhelming a single LLM context window.

Core Features & Use Cases

Long-context loading: Load PDFs and codebases into a ContextStore as chunks with source traceability.
Evidence retrieval: Use optional BM25 pre-filtering to narrow candidate chunks before analysis.
REPL-driven reasoning: Interactively inspect and process chunks inside a Python REPL using variables and llm_query() for sub-LLMs.
Final output with traceability: Produce results via FINAL or FINAL_VAR to ensure reproducible, cit-able answers.
Reference support: Maintain evidence references and headers for auditability.

Quick Start

Provide a PDF or code repository and ask a question; the skill will load the content, enable iterative querying via REPL, and return a final answer with traceable evidence.

Frequently Asked Questions about long-document-evidence-reader

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I extract evidence from a long PDF without exceeding the LLM context window?▼

Yes, you can retrieve traceable evidence from ultra-long PDFs by loading them into a ContextStore as chunks. The skill uses a Python REPL context and BM25 pre-filtering to search and cite specific sections without overwhelming the LLM context window.

Can I search a large codebase and get source-traceable answers using Python?▼

Yes, you can search large codebases by loading the repository into a ContextStore as chunks. Using a Python REPL context, you interactively inspect chunks and execute sub-LLM queries to return final answers with traceable source references.

Does BM25 pre-filtering help with long document retrieval in Python?▼

Yes, BM25 pre-filtering helps long document retrieval by narrowing down candidate chunks before deeper analysis. It identifies the most relevant text segments early, reducing processing overhead and improving the accuracy of evidence extraction from large collections.

How do I ensure reproducible answers when querying large document collections?▼

Ensure reproducible answers when querying large collections by using a FINAL or FINAL_VAR output protocol. This structured output mechanism maintains evidence references and headers, guaranteeing that results are citable and auditable across multiple runs.

Do I need PyPDF to load PDFs for evidence extraction?▼

Yes, you need PyPDF or PyPDF2 dependencies installed to load PDFs for evidence extraction. The skill uses the load_pdf loader to process the documents, chunk their content, and make them available for querying within the REPL environment.

What is the best way to maintain source traceability when analyzing compliance documents?▼

The best way to maintain source traceability for compliance documents is chunking content into a ContextStore with reference headers. Using a REPL environment with a FINAL output protocol ensures every extracted answer retains its original citation for audits.