The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it ingests untrusted user-defined tasks to drive its iterative logic without employing boundary markers to isolate the instructions.
Ingestion points: User-provided task descriptions supplied via the /loop command or natural language triggers like "iterate until done" (SKILL.md).
Boundary markers: Absent. The skill does not instruct the agent to use delimiters or specific tags to encapsulate the user-defined goals.
Capability inventory: The skill is documented to use Bash, Edit, and Write tools to perform implementation attempts (SKILL.md).
Sanitization: Absent. There is no evidence of validation or escaping for the content within the task descriptions.
[COMMAND_EXECUTION]: The skill instructions explicitly direct the agent to utilize shell commands and custom CLI tools to achieve its goals.
Evidence: Use of the Bash tool for implementation steps and the ralph gates command for quality validation (SKILL.md).

loop