The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted code provided by users or found in repository files, which creates a surface for indirect prompt injection attacks.
Ingestion points: The instructions in SKILL.md specify that the agent should use "user's named files or diff" and inspect "surrounding code" as the primary data sources.
Boundary markers: Absent. The skill does not provide the agent with explicit delimiters or instructions to ignore potential Natural Language commands hidden within code comments or string literals in the source files being simplified.
Capability inventory: The skill is designed to perform code analysis and modification ("Finish the scoped cleanup"), which involves writing changes back to the filesystem.
Sanitization: Absent. There are no mentions of filtering or escaping external content before the agent processes the code logic.

code-simplifier