document-chunkerlisted

Use this skill when the task involves processing large files or document sets with Codex as the AI coworker: document ingestion, file inventory, parsing, chunking, structured extraction, embeddings preparation, retrieval QA, SEC/Form D processing, HypeStake knowledge ingestion, or auditing .chunks.jsonl outputs. All AI interpretation must be performed by Codex, not by bundled scripts.
andrewr303/claude-codex-plugin-lab · ★ 0 · Data & Documents · score 60

Install: claude install-skill andrewr303/claude-codex-plugin-lab

# Document Chunker This skill turns Codex into an advanced document-processing coworker. The key rule: **all AI processing happens inside Codex**. Helper scripts are deterministic utilities for reading files, creating inventories, converting simple formats to text, slicing large text into stable packets, merging Codex-produced JSON, auditing artifacts, and running lexical retrieval canaries. They must not call LLM, embedding, rerank, OCR, or search APIs unless the user explicitly asks and the host repo already provides an approved path. ## Use this skill for - Very large files that cannot be pasted into chat. - Folders of PDFs, DOCX, Markdown, CSV, JSON, JSONL, XML, HTML, text, SEC filings, pitch decks, reports, transcripts, or scraped pages. - HypeStake `ai-knowledge`, reference documents, Company Analysis, Portfolio Analysis, market intel, deal/finance/valuation, creator genome, and RAG artifacts. - Existing `.chunks.jsonl`, embedding-input, vector payload, or retrieval output audits. - Designing or implementing a document-processing pipeline where Codex runs local scripts during the conversation. Do not use this skill for ordinary code edits that do not touch document processing or RAG. ## Core operating posture 1. **Codex is the AI processor.** Use Codex's reasoning, chat context, file access, subagents, and local command execution. Scripts only prepare, slice, validate, merge, and audit. 2. **Never paste huge files blindly.** First inventory, classify, convert, an