Literature & Web Search for Engineering Tasks
Find, store, and summarize related papers and web information for a NeqSim task
so the research is auditable, reusable, and distributable with the task
folder. This skill defines the workflow and storage convention; it pairs with
neqsim-stid-retriever (internal/vendor corpora), neqsim-technical-document-reading
(fact extraction), neqsim-pdf-ocr (scanned PDFs), and neqsim-standards-lookup
(which standards to pull). It is the open-literature/web counterpart to the
internal-document retrieval flow, and shares the same references/ collection
layout so everything lands in one distributable folder.
When to Use
- At the start of Step 1 (Scope & Research) on Standard/Comprehensive tasks.
- When the solver needs a reference for a claim, a correlation source, or an
independent benchmark dataset (NIST, textbook worked examples, published lab
data, standards).
- When background context from the web (method descriptions, equipment
background, operator/vendor public material, regulatory text) informs scope.
Storage Convention (distributable, per-source)
All research is stored inside the task folder under
step1_scope_and_research/references/, filed by source so the collection can be
handed to others:
references/
├── SOURCES.md # human-readable summary (auto-generated)
├── collection_manifest.json # machine-readable record (auto-generated)
├── literature_findings.md # structured claims + citations from this search
├── literature/ # papers, standards, textbook PDFs
│ ├── Smith2019_CO2_dense_phase.pdf
│ └── NORSOK_M-001_2017.pdf
└── web/ # saved web pages / article extracts
├── nist_webbook_CO2_density.md
└── wikipedia_Joule-Thomson.md
- Papers / standards / textbook PDFs →
references/literature/.
- Web pages →
references/web/, saved as a Markdown extract (title, URL,
retrieval date, and the relevant text/tables) — NOT a raw HTML dump. Name
files by source + topic (e.g. nist_webbook_CO2_density.md).
- Never save research to workspace-level
output/ or figures/.
Workflow
Frame the questions. From the task scope, write 3–8 specific research
questions (e.g. "What is the measured CO2 density at 100 bar, 40 °C?",
"Which standard governs dense-phase CO2 pipeline material selection?").
Record them at the top of literature_findings.md.
Local library first (before any web search). The user may have a
configured document root holding full standards, company technical
requirements, and vendor datasheets. Searching it is faster and more
authoritative than the open web, and it is free:
neqsim --show-document-root # may be unset - then there is no library
neqsim documents "NORSOK" # recursive, case-insensitive
The same listing is in references/document_root_index.md inside the task
folder, so a subagent without CLI access can still see what is available.
Copy anything used into references/literature/ (open sources) or
references/manual/ (company documents). File names are indicative only —
open a likely candidate before ruling the library out.
Standards next. Use neqsim-standards-lookup to identify applicable
standards before chasing papers — they often answer the question directly.
Save standard PDFs (when obtainable) to literature/.
Internal corpus (if configured). If a retrieval backend is available,
pull internal/vendor docs via neqsim-stid-retriever into their own source
subfolders (stid/, vendor/, manual/).
Web search. Use the fetch_webpage tool to retrieve and read candidate
pages (query = a research question). Prefer authoritative sources (NIST,
standards bodies, peer-reviewed / DOI, textbook publishers, reputable
engineering references). For each useful page, save a Markdown extract to
references/web/ with this header, then the relevant excerpt/tables:
# <Page title>
- URL: <full url>
- Retrieved: YYYY-MM-DD
- Source type: web | dataset | standard-portal | encyclopedia | vendor
- Relevance: high | medium | low
<the relevant text / table / data — quoted, with the location noted>
Do not invent URLs or fabricate content. If a page is paywalled and only
the abstract is available, save the abstract and say so.
Papers. Retrieve open-access PDFs (arXiv / DOI / publisher) into
references/literature/. For paywalled papers, save the abstract + citation
as a .md stub in literature/ and record the access gap. If a configured
paper-search backend exists, use it; otherwise list candidates and ask the
user to confirm downloads.
Extract information. For each stored PDF, extract the specific facts that
answer a research question via neqsim-technical-document-reading (OCR first
via neqsim-pdf-ocr for scans). For figures/curves/tables, convert PDF pages
with devtools/pdf_to_figures.py and read them with view_image; save
extracted PNGs to the task figures/ folder.
Write literature_findings.md. One entry per useful source with:
citation, source path (relative), a two-sentence contribution summary, the
specific claim/data it supports (with page/section/table pointer), and a
relevance score. Group by research question so each question shows its
supporting evidence. Triage: keep the strongest ≤ 10; note pruned candidates
with the reason, so the choice is auditable.
(Re)build the distributable summary. Run the generator so the papers and
web extracts are organized and indexed:
uv run --project devtools --locked devtools/generate_sources_md.py <task_dir> --organize
It files loose files into literature/ / web/ and rebuilds
references/SOURCES.md + references/collection_manifest.json.
Feed the task. Copy the strongest entries into results.json
references[], and use the extracted data (not raw prose) as benchmark /
basis inputs for the NeqSim study. Append the ## Literature & Reference Documents section to notes.md.
literature_findings.md skeleton
# Literature & Web Findings — <task title>
Search date: YYYY-MM-DD
## Research questions
1. <question>
2. <question>
## Findings
### Q1 — <question>
- **Smith (2019)**, `literature/Smith2019_CO2_dense_phase.pdf` — measured CO2
density 628 kg/m3 at 100 bar / 40 °C (Table 3, p. 5). Relevance: high.
Use as benchmark for the SRK density validation.
- **NIST WebBook**, `web/nist_webbook_CO2_density.md` — reference EOS density
627.9 kg/m3 at the same state. Relevance: high.
### Q2 — <question>
- **NORSOK M-001 (2017)**, `literature/NORSOK_M-001_2017.pdf` §6.3 — material
selection for CO2 wet service. Relevance: medium.
## Pruned candidates (with reason)
- <title> — superseded by Smith (2019); older correlation.
Operating Principles
- Store inside the task folder, filed by source (
literature/, web/).
- Cite, don't paraphrase — every finding points to a page/section/table or
a URL so it can be verified.
- No invented URLs or data. Record access gaps (paywall, not found)
honestly.
- Triage by relevance — prune to the strongest sources; keep the reasoning.
- Prefer authoritative sources (standards, NIST, DOI, textbooks) for any
number used as a benchmark or design basis.
- Keep it distributable — always regenerate
SOURCES.md after a search so
the whole task folder can be shared and the collected material reused.
Related Skills
neqsim-stid-retriever — internal/vendor document corpora (per-source folders).
neqsim-technical-document-reading — extract structured facts from PDFs/images.
neqsim-pdf-ocr — OCR for scanned PDFs before extraction.
neqsim-standards-lookup — identify governing standards to retrieve.
neqsim-professional-reporting — results.json references[] and citation format.
1---2name: neqsim-literature-search3description: "Literature and web research workflow for NeqSim tasks. USE WHEN: a task needs papers, standards, textbook or benchmark data, or background from the open web, stored and summarized so it is reusable. Saves papers to references/literature/ and web extracts to references/web/, writes literature_findings.md (claims, citations, relevance) and feeds references/SOURCES.md."4last_verified: "2026-07-13"5metadata:6 source: https://skillmd.com/skills/equinor/neqsim-literature-search7---89# Literature & Web Search for Engineering Tasks1011Find, store, and summarize related papers and web information for a NeqSim task12so the research is **auditable, reusable, and distributable** with the task13folder. This skill defines the *workflow and storage convention*; it pairs with14`neqsim-stid-retriever` (internal/vendor corpora), `neqsim-technical-document-reading`15(fact extraction), `neqsim-pdf-ocr` (scanned PDFs), and `neqsim-standards-lookup`16(which standards to pull). It is the open-literature/web counterpart to the17internal-document retrieval flow, and shares the same `references/` collection18layout so everything lands in one distributable folder.1920## When to Use2122- At the start of Step 1 (Scope & Research) on Standard/Comprehensive tasks.23- When the solver needs a reference for a claim, a correlation source, or an24 independent benchmark dataset (NIST, textbook worked examples, published lab25 data, standards).26- When background context from the web (method descriptions, equipment27 background, operator/vendor public material, regulatory text) informs scope.2829## Storage Convention (distributable, per-source)3031All research is stored **inside the task folder** under32`step1_scope_and_research/references/`, filed by source so the collection can be33handed to others:3435```36references/37├── SOURCES.md # human-readable summary (auto-generated)38├── collection_manifest.json # machine-readable record (auto-generated)39├── literature_findings.md # structured claims + citations from this search40├── literature/ # papers, standards, textbook PDFs41│ ├── Smith2019_CO2_dense_phase.pdf42│ └── NORSOK_M-001_2017.pdf43└── web/ # saved web pages / article extracts44 ├── nist_webbook_CO2_density.md45 └── wikipedia_Joule-Thomson.md46```4748- **Papers / standards / textbook PDFs** → `references/literature/`.49- **Web pages** → `references/web/`, saved as a Markdown extract (title, URL,50 retrieval date, and the relevant text/tables) — NOT a raw HTML dump. Name51 files by source + topic (e.g. `nist_webbook_CO2_density.md`).52- Never save research to workspace-level `output/` or `figures/`.5354## Workflow55561. **Frame the questions.** From the task scope, write 3–8 specific research57 questions (e.g. "What is the measured CO2 density at 100 bar, 40 °C?",58 "Which standard governs dense-phase CO2 pipeline material selection?").59 Record them at the top of `literature_findings.md`.60612. **Local library first (before any web search).** The user may have a62 configured document root holding full standards, company technical63 requirements, and vendor datasheets. Searching it is faster and more64 authoritative than the open web, and it is free:6566 ```bash67 neqsim --show-document-root # may be unset - then there is no library68 neqsim documents "NORSOK" # recursive, case-insensitive69 ```7071 The same listing is in `references/document_root_index.md` inside the task72 folder, so a subagent without CLI access can still see what is available.73 Copy anything used into `references/literature/` (open sources) or74 `references/manual/` (company documents). File names are indicative only —75 open a likely candidate before ruling the library out.76773. **Standards next.** Use `neqsim-standards-lookup` to identify applicable78 standards before chasing papers — they often answer the question directly.79 Save standard PDFs (when obtainable) to `literature/`.80814. **Internal corpus (if configured).** If a retrieval backend is available,82 pull internal/vendor docs via `neqsim-stid-retriever` into their own source83 subfolders (`stid/`, `vendor/`, `manual/`).84855. **Web search.** Use the `fetch_webpage` tool to retrieve and read candidate86 pages (query = a research question). Prefer authoritative sources (NIST,87 standards bodies, peer-reviewed / DOI, textbook publishers, reputable88 engineering references). For each useful page, save a Markdown extract to89 `references/web/` with this header, then the relevant excerpt/tables:9091 ```markdown92 # <Page title>9394 - URL: <full url>95 - Retrieved: YYYY-MM-DD96 - Source type: web | dataset | standard-portal | encyclopedia | vendor97 - Relevance: high | medium | low9899 <the relevant text / table / data — quoted, with the location noted>100 ```101102 Do **not** invent URLs or fabricate content. If a page is paywalled and only103 the abstract is available, save the abstract and say so.1041056. **Papers.** Retrieve open-access PDFs (arXiv / DOI / publisher) into106 `references/literature/`. For paywalled papers, save the abstract + citation107 as a `.md` stub in `literature/` and record the access gap. If a configured108 paper-search backend exists, use it; otherwise list candidates and ask the109 user to confirm downloads.1101117. **Extract information.** For each stored PDF, extract the specific facts that112 answer a research question via `neqsim-technical-document-reading` (OCR first113 via `neqsim-pdf-ocr` for scans). For figures/curves/tables, convert PDF pages114 with `devtools/pdf_to_figures.py` and read them with `view_image`; save115 extracted PNGs to the task `figures/` folder.1161178. **Write `literature_findings.md`.** One entry per useful source with:118 citation, source path (relative), a two-sentence contribution summary, the119 **specific claim/data** it supports (with page/section/table pointer), and a120 relevance score. Group by research question so each question shows its121 supporting evidence. Triage: keep the strongest ≤ 10; note pruned candidates122 with the reason, so the choice is auditable.1231249. **(Re)build the distributable summary.** Run the generator so the papers and125 web extracts are organized and indexed:126127 ```bash128 uv run --project devtools --locked devtools/generate_sources_md.py <task_dir> --organize129 ```130131 It files loose files into `literature/` / `web/` and rebuilds132 `references/SOURCES.md` + `references/collection_manifest.json`.13313410. **Feed the task.** Copy the strongest entries into `results.json`135 `references[]`, and use the extracted data (not raw prose) as benchmark /136 basis inputs for the NeqSim study. Append the `## Literature & Reference137 Documents` section to `notes.md`.138139## `literature_findings.md` skeleton140141```markdown142# Literature & Web Findings — <task title>143144Search date: YYYY-MM-DD145146## Research questions1471. <question>1482. <question>149150## Findings151152### Q1 — <question>153- **Smith (2019)**, `literature/Smith2019_CO2_dense_phase.pdf` — measured CO2154 density 628 kg/m3 at 100 bar / 40 °C (Table 3, p. 5). Relevance: high.155 Use as benchmark for the SRK density validation.156- **NIST WebBook**, `web/nist_webbook_CO2_density.md` — reference EOS density157 627.9 kg/m3 at the same state. Relevance: high.158159### Q2 — <question>160- **NORSOK M-001 (2017)**, `literature/NORSOK_M-001_2017.pdf` §6.3 — material161 selection for CO2 wet service. Relevance: medium.162163## Pruned candidates (with reason)164- <title> — superseded by Smith (2019); older correlation.165```166167## Operating Principles1681691. **Store inside the task folder**, filed by source (`literature/`, `web/`).1702. **Cite, don't paraphrase** — every finding points to a page/section/table or171 a URL so it can be verified.1723. **No invented URLs or data.** Record access gaps (paywall, not found)173 honestly.1744. **Triage by relevance** — prune to the strongest sources; keep the reasoning.1755. **Prefer authoritative sources** (standards, NIST, DOI, textbooks) for any176 number used as a benchmark or design basis.1776. **Keep it distributable** — always regenerate `SOURCES.md` after a search so178 the whole task folder can be shared and the collected material reused.179180## Related Skills181182- `neqsim-stid-retriever` — internal/vendor document corpora (per-source folders).183- `neqsim-technical-document-reading` — extract structured facts from PDFs/images.184- `neqsim-pdf-ocr` — OCR for scanned PDFs before extraction.185- `neqsim-standards-lookup` — identify governing standards to retrieve.186- `neqsim-professional-reporting` — `results.json` `references[]` and citation format.