The Agent Skills Directory

[PROMPT_INJECTION] (HIGH): The skill is susceptible to Indirect Prompt Injection due to the ingestion of untrusted data from external websites and files.
Ingestion points: migma projects import <url> (SKILL.md) and migma contacts import <file> (SKILL.md) which pull external content into the agent context.
Boundary markers: Absent. The output of these commands is processed as JSON without clear delimiters to prevent instruction hijacking.
Capability inventory: migma send (network side-effects), migma export (data exfiltration to ESPs), and migma contacts remove (file/data modification).
Sanitization: Absent. There is no evidence that brand voice content or contact data is sanitized before it influences the agent's reasoning.
[EXTERNAL_DOWNLOADS] (MEDIUM): The skill requires the installation of @migma/cli via npm. This is an unverified external dependency from a non-trusted organization, representing a supply chain risk.
[DATA_EXFILTRATION] (MEDIUM): The migma export command group (klaviyo, mailchimp, hubspot) allows for the movement of sensitive contact and campaign data to external platforms. This capability could be exploited if the agent is manipulated by injected instructions.
[COMMAND_EXECUTION] (LOW): The skill routinely executes the migma binary with parameters derived from user input or imported data, which is a necessary but inherent risk factor.

migma