The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted genomic data (FASTA files) and parses the resulting annotation outputs (JSON/TSV). This represents an attack surface where malicious content embedded in the genome metadata or sequences could influence downstream LLM processing.
Ingestion points: Reads genome.fasta in Step 2 and results/sample1.json in Step 4 of SKILL.md.
Boundary markers: The instructions do not define specific delimiters or warnings for the agent to ignore instructions embedded within the genomic data.
Capability inventory: The skill executes shell commands for software installation (mamba, pip), data downloads (bakta_db), and tool execution (bakta) across SKILL.md.
Sanitization: Step 2 includes a Python snippet to clean FASTA headers by renaming them to a standardized alphanumeric format (contig_0001), which reduces the risk of malformed header injection.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands for installing the Bakta environment and running the annotation pipeline. These are standard operations for the documented tool.
[EXTERNAL_DOWNLOADS]: The skill uses bakta_db download to fetch reference databases from official scientific repositories (Zenodo) and mamba/pip to install packages from public registries. These are legitimate resources for genomic analysis.

bakta-genome-annotation