PaperSeek

You are using PaperSeek as an LLM-based literature search agent. This Skill is a routing and operating guide for the paperseek CLI and local Web UI. It helps an agent run natural-language literature search, diagnostics, result export, and citation-map workflows without inventing metadata or storing credentials.

Reference Rules

• For install, Skill script launching, doctor, smoke, config, health checks, and secrets handling, read references/management-layer.md.
• For choosing OpenAlex, Crossref, or WoS Starter and deciding whether citation expansion applies, read references/source-routing.md.
• For exact CLI commands, output fields, JSON expectations, and stable command contracts, read references/cli-contract.md.
• If a reference conflicts with paperseek --help, paperseek sources --json, or paperseek doctor --json, trust the live CLI and report that the Skill reference may need an update.

Quick Self-Check

When the environment is uncertain, start with:

paperseek doctor
paperseek sources

For a minimal live source test:

paperseek smoke --source openalex --query "machine learning" --json

If paperseek is missing, use the bundled launcher for install guidance:

python skills/paperseek/scripts/paperseek.py --install-help

If the current repository is available, install locally:

pip install -e .

The launcher skills/paperseek/scripts/paperseek.py calls the full PaperSeek package. It does not implement a reduced fallback CLI, so single-Skill distribution still needs the PaperSeek package installed or discoverable through PAPERSEEKPROJECTROOT.

Do not ask the user to paste API keys into chat. Ask them to configure keys locally with environment variables, paperseek config set ..., or the Web UI session fields.

Core Workflows

Default Search Procedure

1. Use OpenAlex as the default source unless the user asks for Crossref or WoS, or the task clearly needs DOI registry metadata.
2. For uncertain environments, run paperseek doctor --json before searching.
3. Run search with JSON output for machine parsing:

paperseek search "open innovation and digital platforms" --source openalex --output json

1. Parse ranked results. Prefer stable fields such as relevancescore, citationcount, venue, doi, url, abstract, and relevance_reason.
2. Report search query, source, iteration count, result count, and the top ranked papers. Do not treat LLM relevance score as proof of quality.
3. If the result set is poor, broaden or narrow the natural-language question and rerun. Do not fabricate missing DOI, authors, abstracts, or citations.

Output Boundaries

• PaperSeek returns metadata and ranked candidate lists. It does not download PDFs.
• It may return links to records, DOI pages, OpenAlex records, or available PDF URLs from metadata, but those are not guaranteed access rights.
• WoS Starter should be treated as key-backed and currently marked temporarily unavailable in the UI.
• Crossref is good for DOI and bibliographic metadata; it is not the best sole source for semantic recall.
• Citation expansion is currently OpenAlex-only.

Failure Handling

Do Not Do

• Do not store API keys in Skill files, README files, tests, or chat.
• Do not use PaperSeek for paywall bypass, Sci-Hub fallback, or bulk PDF downloading.
• Do not invent paper metadata from model memory.
• Do not claim a systematic review is complete from PaperSeek output alone.