The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches live pathogen genomic data and nomenclature from GenSpectrum LAPIS instances (cov-spectrum.org, genspectrum.org, pathoplexus.org) and the official pango-designation repository on GitHub. These external references are legitimate and necessary for the skill's stated purpose of providing current surveillance information.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests untrusted data from external APIs and presents it to the agent.
Ingestion points: Remote data is retrieved in scripts/lapis_client.py using urllib.request from multiple GenSpectrum and Pathoplexus API endpoints.
Boundary markers: The scripts structure output into Tables, TSV, or JSON, providing a clear format for the agent to parse.
Capability inventory: The agent has access to Bash, Read, Write, and Edit tools, which could be targeted if malicious data were executed.
Sanitization: The lapis_client.py script includes a sanitize() function that specifically strips control characters (C0 and C1) and collapses whitespace. This prevents a malformed or hostile API response from injecting newlines or tabs that could forge output rows or be misinterpreted as commands by the agent.

pathogen-variant-surveillance