The Agent Skills Directory

CLI Command Execution: The skill uses the pac command-line tool to perform agent synchronization tasks. These operations are restricted to the local environment and intended for managing development workspaces.
Input Sanitization and Security: The skill provides explicit instructions to treat user-provided paths as untrusted text. It mandates strict validation, including the rejection of control characters and the use of single-quoted shell arguments to mitigate potential command injection risks.
External Data Ingestion Surface: The skill resolves workspaces by searching for specific configuration files (.mcs/conn.json, settings.mcs.yml) and ingesting user-supplied paths. While this represents a point where external data enters the agent's context, the skill includes clear boundary instructions and sanitization rules to handle this safely.
Workspace Integrity Controls: The instructions emphasize verifying the existence of sync metadata and explicitly state that sync-related files should not be read or modified directly, preserving the integrity of the PAC CLI's internal state management.

pull-agent