Scite

Build me an open-access citation-context checker as a Next.js app.
Stack: Next.js + TypeScript, SQLite via better-sqlite3, Tailwind. No auth, single user, localhost only.
Core loop:
1. A search box where I paste a DOI or arXiv ID; fetch its metadata and reference list via the free Semantic Scholar API (no key needed).
2. For each citing paper available as open-access full text (via arXiv or PubMed Central APIs), pull the sentence(s) around the citation.
3. Send each citation sentence to the Anthropic API (key from .env) with a fixed prompt asking it to classify the citation as supporting, contrasting, or mentioning, with a one-line reason.
4. Show results in a table: citing paper, classification, and the quoted sentence, filterable by classification.
Out of scope: paywalled-publisher content, browser extension, manuscript reference-check upload, accounts, alerts.
Include a README noting this only works for open-access papers and citing papers, unlike a licensed full-text database.

# AGENTS.md — Build guide for Scite

## Project scope
Pull open-access full text (arXiv, PubMed Central, bioRxiv) for a paper's cited works, then use an LLM to classify each citing sentence as supporting, contrasting or neutral.

Catalogue verdict: no. The AI classification of a citation as supporting, contrasting or mentioning is a solvable NLP task, but the product's real value is the licensed full-text corpus across 40+ publishers plus preprint servers, kept current and cross-referenced at scale. No individual or small team can replicate that data-access moat; a DIY build only works on open-access papers, which is a small fraction of the literature that matters for a lot of fields.
Use the implementation prompt below to define the deliverable. Complete each phase's acceptance checks before extending the scope.

## Working agreement
- Inspect the repository and its existing instructions before choosing paths, dependencies or commands. Keep one coherent stack and explain changes to the proposed architecture.
- Plan a vertical slice that accepts a real input and produces the useful output described below. Persist only the state the prompt calls for; respect memory-only and upstream-managed workflows. Use fixtures only when they are clearly labelled.
- After scaffolding, document the actual install, development, check and build commands in README and keep them synchronized with the package or project manifest. Do not report commands as successful unless they ran.
- Work in small steps. At handoff, list implemented flows, checks actually performed, remaining blockers, and any credentials or provider setup the owner must supply.
- Do not publish, spend money, contact customers, delete source data or run irreversible migrations without the project owner's authorization.

## Prerequisites
- OpenAI/Anthropic API key
- access to open-access full-text sources (arXiv, PMC, bioRxiv)

## Stack and architecture
- Stack: Next.js + TypeScript, SQLite via better-sqlite3, Tailwind. No auth, single user, localhost only.
- Model the inputs, state transitions, and outputs named in this citation context + academic literature search prompt; keep source IDs and timestamps.
- Interface for this citation context + academic literature search workflow: a local web page with input, progress, review, and export views.

## Security and data integrity
- If OpenAI/Anthropic API key is unavailable, keep the source record and show a recoverable error instead of a fabricated result.
- Keep API keys and OAuth secrets in .env; redact them from logs, generated files, and error responses.
- State the practical limit: coverage of paywalled publishers (Wiley, Cambridge, Wolters Kluwer, etc.).

## Agent implementation rules
- Project rule, data: Model the inputs, state transitions, and outputs named in this citation context + academic literature search prompt; keep source IDs and timestamps.
- Project rule, behavior: Send each citation sentence to the Anthropic API (key from .env) with a fixed prompt asking it to classify the citation as supporting, contrasting, or mentioning, with a one-line reason.
- Project rule, recovery: If OpenAI/Anthropic API key is unavailable, keep the source record and show a recoverable error instead of a fabricated result.

## Optional agent skills and references
No installable skill is listed for this guide; use the project rules and the stack documentation.

Read the linked SKILL.md and its dependencies before adding a skill. Select only the skills matching this project's runtime and task; their documentation does not supply API access, credentials or approval to perform external actions. Pin the reviewed revision where the tool supports it. Follow the chosen agent's documented project-level installation mechanism.

## Distribution ideas
These are optional planning notes. Obtain the owner's approval before publishing or contacting anyone.
- Demonstrate this citation context + academic literature search operation with sample or user-owned data: Send each citation sentence to the Anthropic API (key from .env) with a fixed prompt asking it to classify the citation as supporting, contrasting, or mentioning, with a one-line reason.
- Show a concrete result or export: Show results in a table: citing paper, classification, and the quoted sentence, filterable by classification.
- State the practical limit: coverage of paywalled publishers (Wiley, Cambridge, Wolters Kluwer, etc.).

## Engineering roadmap
1. Phase 1, architecture and data: Stack: Next.js + TypeScript, SQLite via better-sqlite3, Tailwind. No auth, single user, localhost only. Model the inputs, state transitions, and outputs named in this citation context + academic literature search prompt; keep source IDs and timestamps.
2. Phase 2, implement: Send each citation sentence to the Anthropic API (key from .env) with a fixed prompt asking it to classify the citation as supporting, contrasting, or mentioning, with a one-line reason.
3. Phase 3, implement: Show results in a table: citing paper, classification, and the quoted sentence, filterable by classification.
4. Phase 4, review and output: Pull open-access full text (arXiv, PubMed Central, bioRxiv) for a paper's cited works, then use an LLM to classify each citing sentence as supporting, contrasting or neutral. Show results in a table: citing paper, classification, and the quoted sentence, filterable by classification.
5. Phase 5, recovery and acceptance: If OpenAI/Anthropic API key is unavailable, keep the source record and show a recoverable error instead of a fabricated result. Verify this invariant with a saved fixture: An invalid input or interrupted operation must retain the source and show a recoverable state; exported records must reload with the same IDs. State the practical limit: coverage of paywalled publishers (Wiley, Cambridge, Wolters Kluwer, etc.).

## Paid-product capabilities outside this build
- coverage of paywalled publishers (Wiley, Cambridge, Wolters Kluwer, etc.)
- pre-built citation database spanning hundreds of millions of citation statements
- retraction/correction flags sourced from Crossref and PubMed
- browser extension overlay on Google Scholar and journal pages
- reference-check tool for uploaded manuscripts

## Implementation prompt
Build me an open-access citation-context checker as a Next.js app.
Stack: Next.js + TypeScript, SQLite via better-sqlite3, Tailwind. No auth, single user, localhost only.
Core loop:
1. A search box where I paste a DOI or arXiv ID; fetch its metadata and reference list via the free Semantic Scholar API (no key needed).
2. For each citing paper available as open-access full text (via arXiv or PubMed Central APIs), pull the sentence(s) around the citation.
3. Send each citation sentence to the Anthropic API (key from .env) with a fixed prompt asking it to classify the citation as supporting, contrasting, or mentioning, with a one-line reason.
4. Show results in a table: citing paper, classification, and the quoted sentence, filterable by classification.
Out of scope: paywalled-publisher content, browser extension, manuscript reference-check upload, accounts, alerts.
Include a README noting this only works for open-access papers and citing papers, unlike a licensed full-text database.

## Completion evidence
Demonstrate the prompt's acceptance scenarios against the scoped workflow. Include setup from a clean checkout and failure recovery. Check persistence across restart and export/restore only for the state the prompt says to store; for memory-only tools, confirm that temporary content is discarded as specified. Record actual results and remaining limitations. A detailed plan alone does not establish a working replacement.