The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to read project-level configuration and instruction files, creating an attack surface for indirect prompt injection.
Ingestion points: The skill reads context from CLAUDE.md, AGENTS.md, and .lovable-agent/config.json to determine project architecture and workflow rules.
Boundary markers: No explicit markers or instructions are provided to isolate these inputs or to prevent the agent from obeying instructions embedded within these files.
Capability inventory: The agent has the capability to modify source code, perform git push operations to the main branch, and initiate backend deployments.
Sanitization: The content read from these files is not sanitized before being processed by the agent.
[DYNAMIC_EXECUTION]: The beta 'Yolo Mode' feature automates deployments via browser automation or the Model Context Protocol (MCP).
Execution method: The skill automates browser-based interactions or uses MCP to submit deployment prompts directly to the Lovable.dev platform.
Automation scope: This feature automates sensitive operations like edge function deployments and database migration applications, reducing human oversight in the deployment pipeline.

lovable