Doc Research Download

Doc Research Download Skill

Overview

Use this skill for any WRK that collects domain reference documents: naval architecture, electrical engineering, structural analysis, etc. It establishes where files live, how the download script is structured, and what the catalogue YAML must contain.

6-Step Workflow

1. Fetch seed URLs — load each seed URL, extract all linked PDFs and sub-pages; produce an explicit list of direct download URLs (this is the discovery phase — output feeds the download script, not the template directly)
2. Web search for extras — search OCW, archive.org, standards bodies (IEC, ISO, IEEE, NFPA, ABS, DNV) for freely-available PDFs; add direct URLs to the list from step 1
3. Create domain download script — copy the template below; populate download calls with the URL list from steps 1–2; source scripts/lib/download-helpers.sh
4. Catalogue — write knowledge/seeds/<domain>-resources.yaml per knowledge/seeds/schema.md; validate each saved file is a real PDF (file <path> must report PDF document); move HTML/WAF responses to pending_manual: with notes: "WAF — saved HTML, not PDF"
5. Regenerate document index — run scripts/data/document-index/ pipeline; if the full run takes >5 min, write .Codex/work-queue/assets/<WRK-ID>/index-regen-queued.txt with the command to run — that file is the verifiable evidence for AC completion
6. Record WAF-blocked / borrow-only — add to pending_manual: in the YAML; never silently skip — every attempted resource must appear somewhere

Script Template

Copy and adapt for each new domain:

#!/usr/bin/env bash
# ABOUTME: Download open-access <DOMAIN> documents to /mnt/ace/docs/_standards/<DIR>/
# Usage: bash scripts/data/<domain>/download-<domain>-docs.sh [--dry-run]

set -euo pipefail

DEST="/mnt/ace/docs/_standards/<DIR>"
LOG_DIR="$(git rev-parse --show-toplevel)/.Codex/work-queue/assets/<WRK-ID>"
LOG_FILE="${LOG_DIR}/download.log"
DRY_RUN=false
[[ "${1:-}" == "--dry-run" ]] && DRY_RUN=true

mkdir -p "${DEST}/textbooks"
mkdir -p "${LOG_DIR}"

# shellcheck source=scripts/lib/download-helpers.sh
source "$(git rev-parse --show-toplevel)/scripts/lib/download-helpers.sh"

log "=== <DOMAIN> Download — <WRK-ID> ==="
log "Destination: ${DEST}"
log "Dry run: ${DRY_RUN}"

# ─────────────────────────────────
# TEXTBOOKS — direct PDF links
# ─────────────────────────────────
log "--- Textbooks ---"

download \
  "https://example.org/book.pdf" \
  "${DEST}/textbooks" \
  "Descriptive-Filename.pdf"

log "=== Download complete ==="
total=$(find "${DEST}" -name "*.pdf" | wc -l)
log "  Total PDFs: ${total}"

Catalogue YAML Template

category: <domain-slug>
subcategory: references
created_at: "YYYY-MM-DD"

textbooks:
  - title: "Full Book Title"
    author: "Author or Organisation"
    year: YYYY
    local_path: "/mnt/ace/docs/_standards/<DIR>/textbooks/filename.pdf"
    source_url: "https://..."
    size_mb: N
    topics: [topic1, topic2]

online_portals:
  - title: "Portal Name"
    url: "https://..."
    notes: "What is available here"

pending_manual:
  - title: "Blocked Book"
    url: "https://..."
    notes: "Reason: archive.org borrow-only"

AC Checklist

• All seed URLs fetched and linked PDFs extracted
• Web search performed for additional freely-available resources
• All downloadable PDFs saved to /mnt/ace/docs/_standards/<DOMAIN>/
• Catalogue YAML written at knowledge/seeds/<domain>-resources.yaml
• Document index regenerated, or assets/<WRK-ID>/index-regen-queued.txt written
• No paywalled or DRM-protected content downloaded; WAF-blocked items in pending_manual:

Known WAF Patterns

These patterns were identified during WRK-1151 (naval architecture).

Web Crawling & MCP Assessment (2026-03-14)

Decision: No external MCP needed. Use Python scripts (httpx + beautifulsoup4).

Available built-in tools for web content:

• WebFetch — single URL fetch (adequate for known pages)
• WebSearch — web search + results
• Codex-in-chrome — full browser automation for JS-rendered pages

Assessed and rejected (cost or unnecessary complexity):

• Firecrawl MCP — paid service, rejected
• Crawl4AI MCP — free but adds unnecessary dependency layer
• Playwright MCP — only useful for JS-rendered pages (already covered by Codex-in-chrome)
• Fetch MCP (Anthropic) — minimal gain over WebFetch

For WRK-1202 Tier 3 scripts, use httpx + beautifulsoup4 + robotsparser directly. BSEE/EIA/IMO are standard REST APIs and static HTML — no MCP needed.

See also: web-scraper-energy skill for existing BOEM/BSEE scraping patterns.