The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions define a professional scientific persona and do not attempt to bypass safety filters or override system constraints.
[DATA_EXFILTRATION]: No sensitive local file paths (such as .ssh or .aws) are accessed, and there are no network exfiltration patterns. All external links point to reputable scientific organizations including NIH, CDC, UniProt, and major academic publishers.
[EXTERNAL_DOWNLOADS]: While the skill references numerous external URLs for scientific documentation, it does not include commands for the automated download or execution of remote code.
[OBFUSCATION]: No obfuscation techniques such as Base64 encoding, hex escapes, or hidden characters were found.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process biochemical data like metabolomics and sequences.
Ingestion points: Processes metabolite/lipid data and protein sequences (SKILL.md).
Boundary markers: No explicit delimiter markers are present for untrusted data inputs.
Capability inventory: Mentions the use of external tools like Python and R for data fitting, but does not define any automated tool calls or shell executions within the skill itself.
Sanitization: Not applicable as no code is provided.
Assessment: The risk is mitigated by the instructions' focus on experimental rigor and artifact detection.

biochemist