The Agent Skills Directory

[SAFE]: No malicious code or exfiltration patterns were found. The skill operates strictly within its declared scope of academic research discovery and curation.
[EXTERNAL_DOWNLOADS]: The skill connects to reputable academic domains such as arXiv.org, Google Scholar, PubMed, and Semantic Scholar to retrieve research metadata and abstracts. These connections are essential for the skill's primary function and target well-known, trusted academic infrastructure.
[COMMAND_EXECUTION]: Uses python3 to execute the bundled toolchains/arxiv_search.py script. This script provides a structured interface to the arXiv API, implements polite rate-limiting according to the source's policy, and uses standard libraries for parsing XML and handling HTTP requests.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection because it processes untrusted research abstracts from external sources.
Ingestion points: Research abstracts are fetched from external APIs as described in sources/arxiv.md, sources/semantic-scholar.md, and sources/pubmed.md.
Boundary markers: Absent. While prompts/summarization.md instructs the model to "Stay strictly inside the abstract," there are no explicit delimiters or instructions to ignore embedded commands within the fetched text.
Capability inventory: manifest.json and instructions/core.md indicate the skill has filesystem-write and shell-exec capabilities.
Sanitization: Absent. External text from abstracts is interpolated directly into the summarization prompts without escaping or validation.

get-research-paper