The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch research data from several external services:
Academic metadata and citations are fetched from the Semantic Scholar API (api.semanticscholar.org).
Paper full-text conversion to Markdown is requested from arxiv2md.org and markxiv.org using the arxiv_id as a parameter.
arXiv search functionality is implemented via the deepxiv-sdk which communicates with the DeepXiv service.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted research paper data (abstracts and conclusions) which could contain adversarial instructions targeting the agent's analysis phase.
Ingestion points: Paper data enters the agent's context through resolve_seed_papers, fetch_papers, and prefetch_sections in scripts/pipeline.py.
Boundary markers: The prompt templates in the references/ directory use section headers (e.g., --- Papers to Analyze ---) to delimit external content.
Capability inventory: The skill utilizes write_file and execute tools to manage working directories and run the internal CLI dispatcher (scripts/cli.py) for data processing.
Sanitization: The scripts/mermaid.py module includes _mermaid_safe logic to escape Mermaid-significant characters in labels, though this is primarily for syntax integrity rather than prompt sanitization.

paper-graph