ADA 2010 Standards for Accessible Design

Paste this prompt into Claude Code — no prior setup, no skills, no Python. Claude downloads the PDF, builds a section outline, and answers your queries with grounded page-level citations using its built-in tools.

TASK
Help me query "ADA 2010 Standards for Accessible Design" — a PDF document — with high precision. I want grounded, citation-backed answers. No external RAG toolkit, no Python venv, no embeddings: just your built-in tools (curl via Bash, Read for PDFs, Write for caching). Zero installation.

SOURCE
PDF URL: https://www.ada.gov/assets/_pdfs/2010ADAStandards.pdf
License: Public Domain
Version: 2010
Local name: ada-2010-standards-for-accessible-design

APPROACH (read this carefully — do not substitute another method)
You will (a) download the PDF once, (b) build a one-time outline that maps section headings to page ranges, then (c) for each of my queries, Read the exact pages most likely to contain the answer and quote them with section + page citations. This is faster and more reliable for a fresh user than setting up vector RAG.

EXECUTION PLAN — work through phases in order. Stop and report the exact error if any phase fails; do not improvise past it.

PHASE 1 — Download (~10 sec)
1.1  mkdir -p ./pdfs ./pdfs/.cache
1.2  curl -fsSL "https://www.ada.gov/assets/_pdfs/2010ADAStandards.pdf" -o ./pdfs/ada-2010-standards-for-accessible-design.pdf
1.3  Verify file size > 100 KB. If suspiciously small, the URL likely returned HTML/redirect — do not proceed; report and ask me to confirm the URL.
1.4  Confirm the file is a real PDF: `file ./pdfs/ada-2010-standards-for-accessible-design.pdf` should report "PDF document".

PHASE 2 — Build outline (~30 sec, one-time)
2.1  Use the Read tool with pages parameter to scan the PDF and locate the Table of Contents.
     Start with pages: "1-5". If TOC isn't there, try "5-25", "25-50" until found.
2.2  Read the full TOC (it may span multiple pages). If the document has no formal TOC, scan the first 20-50 pages in chunks of 10-20 pages and extract section headings with their page numbers.
2.3  Build a structured outline mapping every top-level section and major subsection to its starting page. Persist it as ./pdfs/.cache/ada-2010-standards-for-accessible-design.outline.md in this format:
       # ADA 2010 Standards for Accessible Design — outline
       Total pages: <N>

       ## <Section name> — pages X-Y
       ### <Subsection> — page Z
       ...
2.4  This file is the "index" — every query in Phase 3 starts by re-reading this outline to pick the right pages.

PHASE 3 — Query loop
For each question I ask:
3.1  Re-read ./pdfs/.cache/ada-2010-standards-for-accessible-design.outline.md to refresh the section map.
3.2  Identify the 1-3 sections most likely to contain the answer. List them explicitly before reading.
3.3  Read those exact page ranges from the PDF using the Read tool's pages parameter. Stay within the Read tool's 20-page-per-call limit; chunk if needed.
3.4  Answer with: a direct quote of the relevant clause/figure/table caption, the section heading, and the page number(s). Format as:
       > "<exact quoted text>"
       — Section X.Y, page N
3.5  If the answer is not in the pages you read, expand to adjacent sections (one more Read call). If still not present, say "Not found in this document at the sections I checked: [list]." Never invent section numbers, dimensions, or values.

CONSTRAINTS — never violate
- Only cite content you actually Read this conversation. No memorized values. If you didn't open the page, you don't quote it.
- For tables and figures: read the original page; quote cell values verbatim. Don't paraphrase numbers.
- For dimensions, percentages, and other quantitative values: copy them character-for-character from the PDF.
- Document language is English; queries should also be English for best matching.
- On any phase failure, stop, report the exact command and full error output, and ask me how to proceed. Do not silently retry or work around it.

NEXT STEP
After Phase 2 succeeds (outline.md is written), based on the outline you built suggest 3-5 useful starter queries that are specific to "ADA 2010 Standards for Accessible Design" — not generic. Then wait for my actual queries.