The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process broad, user-provided task descriptions to generate complex workflows, which represents a surface for indirect instructions.
Ingestion points: The skill accepts high-level inputs described as "large, cross-cutting, or unattended tasks" within SKILL.md.
Boundary markers: The instructions lack explicit delimiters or safety guardrails to differentiate between task metadata and potentially malicious embedded instructions in the user input.
Capability inventory: The skill uses disable-model-invocation: true for its own execution but delegates complex capabilities to external tools like architect and arena which are capable of modifying codebases and running evaluations.
Sanitization: No validation or sanitization process for the input task description is defined before the agent proceeds with workflow design and execution.
[EXTERNAL_DOWNLOADS]: The skill's documentation and metadata in README.md and SKILL.md reference configuration and source material from the official Cursor plugins repository on GitHub.

figure-it-out