Literature Search Openalex

OpenAlex Skill

Prerequisites

1. uv: Read the uv skill and follow its Setup instructions to ensure uv is installed and on PATH.

2. User Notification: If .licenses/literature_search_openalex_LICENSE.txt does not already exist in the workspace root directory then (1) prominently notify the user to check the terms at https://help.openalex.org/ and to always check the license of the papers retrieved by the skill for any restrictions, then (2) create the file recording the notification text and timestamp.

3. .env file: Make sure the .env file exists in your home directory. Create one if it does not exist.

4. OPENALEX_API_KEY (optional but recommended): The skill works without a key, but keyless requests share a $0.10/day budget per IP address. A free key has its own $1/day budget and higher rate limits, and is required to download full text from content.openalex.org (download-pdf). Adding mailto (the former "polite pool") no longer raises limits. You can obtain a free key at OpenAlex.org → account settings. Whenever checking for or requesting this key, you MUST use the safe credentials protocol in the credentials skill, without blocking keyless execution if it is unset.

Core Rules

1. List Sources. If this skill is used, ensure this is mentioned in the output AND list the URLs of all papers that were used in producing the output.

2. Resolve before filter. NEVER filter by name. Always resolve a name to an ID first, then use that ID in --filter.

3. Use the CLI only. Never call the API via curl/urllib. The CLI handles retries and rate limiting.

4. No fabrication. Never invent OpenAlex IDs or DOIs. Use resolve/get to look them up. Report empty results accurately.

5. API key. If a command returns 401, returns 429 because of a rate limit or exhausted daily budget, or you need high-volume queries, you MUST use the safe credentials protocol in the credentials skill to check for and request the OPENALEX_API_KEY to help the user add it to their .env file. A free key does not fix a 429 Plan upgrade required error.

6. Keep output small. Always use --select and --per-page 5–10 for overview queries. Pipe filter output to a file (> results.json), then slim with jq before reading into context.

Rate Limits

• Without key: 10 req/s; $0.10/day budget per IP address, shared by all keyless requests from that IP.

• With free key: 30 req/s; $1/day budget.

• Annual plan: 50 req/s.

CLI Reference

uv run scripts/openalex_cli.py [--api-key KEY] <command> [flags]

Entity types (shared across commands): works, authors, sources, institutions, topics, domains, fields, subfields, sdgs, countries, continents, languages, keywords, publishers, funders, work-types, source-types, institution-types, licenses

Commands

resolve <entity> <query> — Name → ID candidates. Returns id, display_name, hint. Use --per-page N for more candidates.

get <entity> <id> — Full metadata for one entity. Accepts short ID (W2741809807), full URL, or DOI URL. Use --select to limit fields.

filter <entity> — Search/filter entities. Key flags are:

• --search <query>: Full-text search (10× cost of --filter)

• --filter <expr>: Filter expressions. Use , for AND and | for OR.

• --sort <field:dir>: Sort results (e.g., cited_by_count:desc)

• --select <fields>: Limit the fields returned in the output.

• --group-by <field>: Aggregate results by a specific field.

• --per-page <N>: Number of results per page (default 25, max 100).

• --page <N>: Specify the page number to retrieve.

• --sample <N>: Get a random sample of up to 10,000 results.

• --seed <N>: Seed for reproducible sampling.

download-pdf <work-id> <output-path> — Download PDF (content.openalex.org requires OPENALEX_API_KEY; falls back to alternative pdf_url locations from work metadata if unauthenticated or if the primary server fails). Always run download-pdf and verify any downloaded PDF is non-empty and not corrupted; if OPENALEX_API_KEY is missing, also refer to the credentials skill to help the user add it.

rate-limit — Check current rate limit and daily credit status (reports shared per-IP limits when keyless, or key-specific limits when OPENALEX_API_KEY is set; if OPENALEX_API_KEY is not set, also refer to the credentials skill to help the user add it).

Search Tips

• If resolve returns no matches, try alternate spellings or abbreviations.

• If --search returns 0 results, try broader terms (max 3 retries).

• If resolve returns multiple candidates, present them to the user with display_name and hint for manual selection.

Entity References

Consult references/ for valid filter, sort, and group-by fields per entity:

• Works — Authors — Sources

• Institutions — Topics — Taxonomy

• Geo & Language — Publishers & Funders

• Type Values

For anything not covered here, or if the API behaves differently from what this skill describes, check the official docs index at https://help.openalex.org/llms.txt.

Common Workflows

# Author's works (resolve → filter)
uv run scripts/openalex_cli.py resolve authors "Geoffrey Hinton"
uv run scripts/openalex_cli.py filter works \
  --filter "authorships.author.id:A5108093963" \
  --sort "cited_by_count:desc" --per-page 10 > papers.json
cat papers.json | jq '[.results[] | {id, title: .display_name, year: .publication_year, citations: .cited_by_count}]'

# DOI lookup
uv run scripts/openalex_cli.py get works "https://doi.org/10.1038/s41586-021-03819-2"

# Bulk DOI lookup (up to 100)
uv run scripts/openalex_cli.py filter works \
  --filter "doi:10.1234/a|10.1234/b|10.1234/c" --per-page 100 > results.json

# Institutional impact by year
uv run scripts/openalex_cli.py resolve institutions "MIT"
uv run scripts/openalex_cli.py filter works \
  --filter "authorships.institutions.id:I63966007" \
  --group-by "publication_year" > mit_by_year.json

# Random sample
uv run scripts/openalex_cli.py filter works \
  --filter "publication_year:2023,is_oa:true" \
  --sample 100 --seed 42 > results.json

Error Handling

The CLI already retries 429s with backoff, except Plan upgrade required ones, which fail immediately with an explanation. Don't blindly retry either kind; the Server body in the error output tells them apart.

Known plan-gated filters: from_created_date, from_updated_date, to_updated_date.

Never fabricate results on empty responses — report accurately and suggest alternate search terms.