The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied JSON briefs which creates a potential surface for indirect prompt injection if the generated output is subsequently processed by other AI systems.
Ingestion points: The input <brief.json> file specified in the workflow instructions.
Boundary markers: The skill documentation in references/discovery-method.md specifies normalization of Unicode and whitespace for all inputs.
Capability inventory: The skill has permissions to read inputs and write multiple JSON artifacts to the local file system within a specified runs directory.
Sanitization: Basic normalization is mentioned, though specific logic to sanitize against adversarial instructions within brief fields is not detailed.
[DYNAMIC_EXECUTION]: The entry point script dynamically adjusts the Python path to locate and load its core logic from a relative source directory.
Evidence: scripts/run_discover.py uses sys.path.insert(0, str(candidate)) to add relative src directories to the module search path.
Context: This is used to resolve the geo_seo_hub package which contains the skill's primary execution logic.

geo-discover