The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read and edit .env.schema and environment-specific files (e.g., .env.production). While these files are intended to be source-controlled, they represent an ingestion point for untrusted data if the repository is compromised. The skill includes explicit boundary markers and 'CRITICAL: Security rules' to prevent the agent from accessing raw secret files (.env.local) or printing plaintext secrets, which significantly mitigates the risk of secret exposure. Capabilities include file modification and execution of the varlock CLI.
[COMMAND_EXECUTION]: The skill documentation mentions the exec() resolver function within the Varlock schema (e.g., SECRET=exec('op read ...')). This is a legitimate feature of the documented tool used for fetching secrets from external providers. The skill provides safety guidance to ensure the agent uses redacted outputs when validating configuration.

varlock