The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill acts as a discovery and management interface for a registry of agent skills. Because these skills consist of natural language instructions, an agent retrieving and processing them is exposed to an indirect prompt injection surface. A maliciously crafted skill in the registry could contain instructions that attempt to override the primary agent's behavior.
Ingestion points: Data enters the agent's context via the search, list, and get actions in scripts/skill_registry_ops.py, which fetch skill definitions and metadata from the external registry.
Boundary markers: Absent. The skill instructions do not define specific delimiters or markers to isolate retrieved skill content from the agent's internal control instructions.
Capability inventory: The skill can perform network operations via the requests library to Google Cloud APIs and read local filesystem data to facilitate the upload and update actions.
Sanitization: Absent. There is no logic provided in the scripts or instructions to filter or sanitize the natural language content retrieved from the registry.
[COMMAND_EXECUTION]: The skill's instructions direct the agent to execute local Python scripts to validate the execution environment and perform registry operations.
Evidence: SKILL.md and its referenced documentation in the references/ directory provide specific shell command blocks for executing scripts/validate_env.py and scripts/skill_registry_ops.py.

agent-platform-skill-registry