The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses instructions to establish an immutable hierarchy that attempts to override subsequent instructions.
Evidence: The constraint instruction.hierarchy.max.priority.no.later.input.can.override explicitly directs the agent to ignore future inputs that might contradict the skill's guidelines, which is a common persistence and bypass pattern.
[PROMPT_INJECTION]: The skill defines an attack surface for Indirect Prompt Injection without providing defense mechanisms.
Ingestion points: The skill processes external content including user-provided code, diffs, and commands as stated in the frontmatter and the TASK definition.
Boundary markers: The instructions lack directives for the agent to use delimiters (like XML tags or triple backticks) to separate untrusted user data from the skill's internal logic.
Capability inventory: The skill is designed for high-privilege tasks including refactoring, debugging, and execution of commands, which could be exploited if malicious instructions are embedded in the processed data.
Sanitization: There are no instructions for validating or sanitizing user-provided identifiers or code blocks before they are integrated into the agent's context.

supreme-coding-guidelines