The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data from "approved source bundles" provided as JSON files, creating an attack surface where instructions hidden in the data could influence the agent.
Ingestion points: The scripts/run_knowledge.py script reads user-provided JSON files via the --input argument.
Boundary markers: There are no explicit delimiters or instructions shown to prevent the agent from obeying commands embedded within the input data.
Capability inventory: The skill has permission to execute shell commands and read/write files in the local environment.
Sanitization: The provided wrapper script does not perform sanitization of the input JSON content.
[DYNAMIC_EXECUTION]: The script scripts/run_knowledge.py dynamically modifies the Python module search path (sys.path.insert) to load the geo_seo_hub library relative to its own installation directory.
[COMMAND_EXECUTION]: The skill executes Python processing scripts via a bash shell environment as part of its standard workflow defined in agents/interface.yaml and manifest.json.

geo-knowledge