The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to manage and interact with highly sensitive data and file paths, such as SSH private keys and environment files (e.g., .env). It provides specific recipes for exporting private keys using the op read --out-file command and reading environment secrets. While the skill mandates strict security controls—such as setting file permissions to 0600, using temporary files with automatic cleanup (trap), and explicitly forbidding the display of secrets in chat—the capability to handle these materials constitutes a significant security surface.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources, including 1Password vaults and local configuration templates, which could potentially host malicious instructions.
Ingestion points: Reads metadata, secret references, and document contents from 1Password vault items and local files.
Boundary markers: Includes a robust 'Safety Contract' and 'Security Patterns' reference that prohibits the agent from echoing secret values and requires explicit user confirmation before any sensitive operation.
Capability inventory: Uses the op CLI for file operations, environment variable injection, and network communication with 1Password service endpoints.
Sanitization: Relies on instructional constraints, emphasizing the use of machine-readable JSON output for metadata and avoiding the retrieval of plaintext secrets into the agent context.
[COMMAND_EXECUTION]: The skill performs extensive command execution via the local op binary. It includes instructions to verify command availability and help output locally to ensure safety and compatibility, while also providing complex shell script templates for secure automation tasks.

1password-cli