The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted source code provided by users or retrieved from repositories (SKILL.md). While the instructions provide boundary markers by emphasizing that the agent must preserve existing logic, public APIs, and error semantics, the skill possesses significant capabilities including writing to the file system and executing project-defined scripts, tests, and package managers (as detailed in references/electron-typescript.md and references/swift-xcode.md). There are no specific instructions provided for sanitizing or escaping the input code before processing.
[COMMAND_EXECUTION]: The skill instructs the agent to execute existing project scripts and tools for validation purposes (references/electron-typescript.md and references/swift-xcode.md). This is a standard functional requirement for a coding assistant, but it represents a potential execution vector if the repository being analyzed contains malicious build or test configurations. The risk is managed by instructing the agent to strictly follow project-declared rules and existing warehouse configurations.

code-simplifier