Abstract Searcher

Add abstracts to .bib file entries by searching academic databases (arXiv, Semantic Scholar, CrossRef) with browser fallback.

Security Scan

The skill's code and APIs match its stated purpose, but the runtime instructions ask the agent to control your real Chrome session and use OS automation (mac-control), which can access private/authenticated data and is not declared or constrained — this mismatch raises privacy and scope concerns.

Purpose & Capability

Name and description match the included Python script and the listed APIs (arXiv, Semantic Scholar, CrossRef, OpenAlex). The network calls in scripts/add_abstracts.py are consistent with fetching abstracts.

Instruction Scope

SKILL.md explicitly instructs the agent to attach to a real Chrome profile, use a browser relay to snapshot pages, click results, and (optionally) use 'mac-control' to auto-click a toolbar icon. Those instructions grant access to any content in the user's Chrome (cookies, logged-in sessions, institutional access) and to OS-level automation, which is broader than simply querying APIs and could expose private data; the skill does not declare or limit this access.

Install Mechanism

No install spec; the skill is instruction-plus-a-script only. That minimizes disk-install risk — the Python script runs locally and makes direct API calls.

Credentials

The skill requests no environment variables or credentials (scripts use only public APIs). However, the browser fallback relies on the user's Chrome profile and OS automation to access paywalled pages — this implicitly leverages credentials/cookies stored in the browser even though no creds are declared.

Persistence & Privilege

always:false and no special persistence requested. The skill does not modify other skills or system-wide settings (based on provided files). Autonomous invocation is the platform default but is not combined here with other high privileges.

What to consider before installing

This skill mostly does what it says when using public APIs, but the SKILL.md asks the agent to control your real Chrome session and potentially use OS automation (mac-control) to scrape paywalled pages. Before installing or enabling it: (1) consider running the Python script manually in a local terminal rather than allowing the agent to drive your browser; (2) avoid giving the agent access to your Chrome profile or to any OS automation tools unless you trust it; (3) if you must allow browser fallback, inspect and limit the browser-relay/tooling permissions and run in a separate browser/profile without sensitive logins; (4) review the script yourself (it uses only listed public APIs) and monitor network activity to ensure it only contacts the stated endpoints.

Automatically fetch and add abstracts to BibTeX entries.

Usage

Given a .bib file, this skill will:

1. Parse each BibTeX entry
2. Search for the abstract using multiple sources:
   • arXiv API (for arXiv papers)
   • Semantic Scholar API
   • CrossRef API
   • OpenAlex API
3. If APIs fail: Use browser automation to search Google Scholar
4. Add abstract={...} to each entry
5. Return the complete modified .bib file

Quick Start

# Process a bib file (API-based)
python3 {baseDir}/scripts/add_abstracts.py input.bib > output.bib

API Sources (No keys required)

1. arXiv API: http://export.arxiv.org/api/query?search_query=...
2. Semantic Scholar: https://api.semanticscholar.org/graph/v1/paper/search?query=...
3. CrossRef: https://api.crossref.org/works?query.title=...
4. OpenAlex: https://api.openalex.org/works?search=...

Browser Fallback (IMPORTANT!)

When APIs fail to find an abstract, use Chrome browser relay like a real person:

Step 1: Attach Chrome tab

# Check if tab is attached
browser action=tabs profile=chrome

# If no tabs, ask user to click the Clawdbot Browser Relay toolbar icon
# Or use mac-control skill to auto-click it

Step 2: Open Google Scholar and search

browser action=open profile=chrome targetUrl="https://scholar.google.com"
browser action=snapshot profile=chrome

# Type the paper title in search box
browser action=act profile=chrome request={"kind":"type","ref":"search box ref","text":"paper title here"}
browser action=act profile=chrome request={"kind":"press","key":"Enter"}
browser action=snapshot profile=chrome

Step 3: Click the result

# Find the paper in results, click to open
browser action=act profile=chrome request={"kind":"click","ref":"paper title link ref"}
browser action=snapshot profile=chrome

Step 4: Extract abstract from the page

• ScienceDirect: Look for "Abstract" section in snapshot
• ACL Anthology: Abstract is directly visible at top
• Springer/Wiley: May need to click "Abstract" to expand
• PubMed: Abstract is usually visible

Step 5: Copy and format for BibTeX

# Get the abstract text from snapshot
# Clean it: remove newlines, escape special chars
# Add to bib entry: abstract={...},

Tips:

• Use profile=chrome to use real Chrome with your login sessions
• Google Scholar rarely blocks real Chrome browsers
• ScienceDirect/IEEE may need institutional login (your Chrome has it)
• Always verify the paper title matches before copying abstract!

Notes

• Rate limiting: 2 seconds between API requests
• Browser fallback should find almost all papers
• Abstracts are cleaned (newlines removed, escaped for BibTeX)
• Always verify the abstract matches the correct paper!