The Agent Skills Directory

[PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection by gathering requirements from the user and interpolating that untrusted data into newly created skill files (SKILL.md). Malicious instructions embedded in the 'Core workflow' could be executed when the newly created skill is later invoked.
Ingestion points: User-defined names, trigger phrases, and core workflow instructions gathered in Phase 1.
Boundary markers: Absent; user input is interpolated directly into the markdown body of the new skill.
Capability inventory: The skill performs file system writes and instructions the agent to execute a validation CLI tool (Phase 3 & 4).
Sanitization: No evidence of validation, escaping, or filtering of the user-provided text before file creation.
[COMMAND_EXECUTION]: The skill requires the agent to execute shell-level operations, including directory creation (mkdir equivalent) and the execution of a project-specific CLI utility (mktg skill validate) as part of the validation and registration workflow (SKILL.md Phase 4 and Step 6 of the workflow).

create-skill