The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from external academic sources such as arXiv, Semantic Scholar, and Google Scholar.
Ingestion points: The workflow includes searching for and extracting data from abstracts and full-text PDFs in Phase 2 and Phase 4 (SKILL.md).
Capability inventory: The skill possesses the capability to execute local scripts via uv run and perform network operations to retrieve academic content (SKILL.md).
Boundary markers: The instructions do not define explicit boundary markers or provide warnings to the agent to disregard instructions that may be embedded within the external research papers.
Sanitization: No sanitization or filtering of the text extracted from PDFs or abstracts is specified before it is analyzed by the language model.
[COMMAND_EXECUTION]: The skill uses a command-line utility to perform searches, which involves passing user-generated queries as arguments to a Python script.
Evidence: uv run --with arxiv python scripts/arxiv_search.py "QUERY" in SKILL.md.
[EXTERNAL_DOWNLOADS]: The skill relies on the uv tool to manage dependencies, which may result in the dynamic download and installation of the arxiv Python package at runtime.
Evidence: The workflow in Phase 2 of SKILL.md specifies the use of uv run --with arxiv.
[DATA_EXFILTRATION]: The skill initiates network requests to external, well-known academic services to fetch paper metadata, citations, and search results.
Evidence: References to api.semanticscholar.org, scholar.google.com, and connectedpapers.com throughout SKILL.md and references/search-strategy.md.

literature-review