The Agent Skills Directory

[SAFE]: The skill performs legitimate network operations to well-known scientific repositories (NCBI PubMed and arXiv) using the Python standard library's urllib module.
[SAFE]: The tool implements defensive research gates, including SHA-256 checksum verification and PDF header validation (%PDF-) for sourced evidence.
[SAFE]: The optional bach_api integration refers to a vendor-owned resource consistent with the skill's provenance and authorship (ellmos-ai).
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources and local files, creating a standard surface for indirect prompt injection.
Ingestion points: XML/Atom formatted scientific abstracts from PubMed/arXiv; local project metadata and planning files (e.g., CHECKS-LOG.txt, CHECKS-REG.md, KONZEPT*, TODO*, BEWEISNOTIZ*).
Boundary markers: The instructions do not define specific delimiters to isolate external text from agent instructions.
Capability inventory: Network access for data retrieval and file system access for caching results and saving research reports.
Sanitization: Data is processed via standard structured parsers, with no additional sanitization for natural language instruction bypasses.

research-agent