The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes scientific papers and metadata fetched from external scholarly databases (arXiv, Crossref, OpenAlex), which serves as an ingestion point for untrusted data.
Ingestion points: External metadata is retrieved via scripts/fetch_metadata.py from public scholarly APIs. Full-text paper content is processed from files provided to the agent.
Boundary markers: The agent is directed to use a specific Evidence note template (assets/evidence-note-template.md) which enforces a structured analysis workflow (Claim, Source, Evidence, Conclusion), reducing the likelihood of the agent following instructions embedded within the paper text.
Capability inventory: The skill has capabilities for network access (metadata fetch) and local script execution (statistical validation and note checking).
Sanitization: scripts/check_stats.py employs strict regex parsing for APA-style reports. scripts/fetch_metadata.py validates the envelope structure (Atom XML or JSON) before processing the content.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to retrieve metadata and software dependencies from well-known scholarly and technical services.
Evidence: scripts/fetch_metadata.py connects to export.arxiv.org, api.crossref.org, and api.openalex.org to retrieve paper metadata. These are recognized, well-known scholarly services.
Evidence: assets/examples/verify.sh uses the uv tool to download and run the scipy library (scipy==1.16.2) from a standard package registry for cross-verification of statistical results.
[DYNAMIC_EXECUTION]: The skill utilizes script generation and runtime execution as part of its internal verification suite.
Evidence: The assets/examples/verify.sh script dynamically generates and executes a Python snippet using uv run to compare the skill's custom math implementation against the scipy library to ensure accuracy.

analyze-scientific-papers