The Agent Skills Directory

EXTERNAL_DOWNLOADS (LOW): The skill facilitates the download and installation of agent templates from remote Git repositories via the import_template_from_git command in references/install.md. This is a standard feature for a template manager but represents a remote code ingestion point.
COMMAND_EXECUTION (LOW): The skill utilizes Model Context Protocol (MCP) commands to interact with the local filesystem and Docker environment (e.g., import_template_from_local, search_agents). These operations are restricted to the local environment and are necessary for the skill's deployment functions.
DATA_EXPOSURE (SAFE): references/analyze.md and references/secrets.md contain instructions for managing credentials. However, the skill encourages the use of vault-based secrets (the $TEMP_CONTEXT_VARIABLE_ pattern) rather than hardcoded credentials, following security best practices for the Machina platform.
DYNAMIC_EXECUTION (LOW): As described in schemas/mapping.md, the skill supports Python-based data transformations within workflow tasks. While this involves dynamic evaluation, it is a core component of the Machina DSL used for data mapping and is managed by the underlying execution engine.
INDIRECT_PROMPT_INJECTION (LOW): The skill possesses a data ingestion surface (Git repositories, local files) and significant capabilities (MCP commands, file writing).
Ingestion points: references/install.md (Git/Local paths).
Boundary markers: Not explicitly defined in the provided schema files.
Capability inventory: File writing, secret creation, and agent/workflow execution via MCP.
Sanitization: Validation steps are defined in references/validate.md to ensure YAML structure compliance, though these primarily focus on syntax rather than malicious instruction filtering.

mkn-constructor