The Agent Skills Directory

[COMMAND_EXECUTION]: The documentation provides shell commands for skill installation and configuration using npx and gh, which are intended for environment setup.
[EXTERNAL_DOWNLOADS]: Instructions suggest downloading and installing vendor-maintained CLI tools (@velinussage/locus-agent-skill) and skill content from the vendor's official GitHub repository.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from third-party sources such as Wikipedia, OpenStreetMap, and federal agency records (EPA, FEMA), representing an attack surface for indirect prompt injection.
Ingestion points: Fetches plain-text extracts from Wikipedia, amenity data from OpenStreetMap, and various hazard/environmental records from federal databases.
Boundary markers: Includes explicit grounding instructions requiring the agent to "Answer only from returned artifacts" and provide caveats, although specific data delimiters are not enforced.
Capability inventory: The skill environment permits Bash, Read, and AskUserQuestion tools, and communicates with external REST and MCP endpoints.
Sanitization: The instructions focus on grounding rather than automated content filtering or sanitization for third-party data ingested during execution.
[DYNAMIC_EXECUTION]: The agent is instructed to treat dynamic tool catalogs from vendor-controlled endpoints (e.g., api.locus.report) as authoritative definitions for tool schemas and operations at runtime.

locus-agent-tools