The Agent Skills Directory

[DYNAMIC_EXECUTION]: The references/workflows.md file contains a code example titled 'Caching Results' that recommends using Python's pickle module. The example utilizes pickle.load() to read data from a local file, which is a known security vulnerability. If a malicious actor can influence the contents of the cache file, they can execute arbitrary code on the user's system during the deserialization process.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection. The gget gpt module accepts user-defined prompts that could potentially be chained with untrusted data retrieved from external bioinformatics databases (e.g., gene descriptions or publication abstracts from Ensembl/NCBI). If these external sources were to contain malicious instructions, they could influence the agent's behavior when processed by the GPT module.
Ingestion points: Data retrieved from 20+ bioinformatics databases including Ensembl, UniProt, and NCBI (references/database_info.md).
Boundary markers: None identified in the provided scripts or documentation; untrusted external content is not isolated from instructional prompts.
Capability inventory: Network access to OpenAI APIs (gget gpt) and extensive local file system operations including file writes and downloads (gget ref, gget setup).
Sanitization: No input validation or sanitization routines are specified for content being passed to the gpt module.
[METADATA_POISONING]: A discrepancy was found in the authorship metadata. The frontmatter in SKILL.md identifies the author as 'K-Dense Inc.', while the platform identifying metadata indicates the author is 'MarieLynneBlock'. Such inconsistencies can suggest unverified provenance or errors in metadata management.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download of large biological datasets and protein model parameters (e.g., AlphaFold models) from reputable scientific repositories. These downloads are documented, serve the primary purpose of the tool, and are directed toward well-known research institutions.

gget