The Agent Skills Directory

[COMMAND_EXECUTION]: The skill guides the agent to execute shell commands to compile, build, and run custom linting logic. Examples include cargo dylint --all (Rust), npx eslint . (Node.js), pylint --load-plugins... (Python), and go build followed by running a binary (Go). While standard for development, this involves executing code dynamically created or modified during the skill's workflow.
[INDIRECT_PROMPT_INJECTION]: The core purpose of the skill is to generate linter error messages that act as "repair instruction prompts" for an AI agent. This creates an indirect prompt injection surface: if a project contains a malicious or flawed custom lint rule, the agent may follow the instructions in the linter's output to perform unintended file modifications or deletions (as seen in the mod.rs deletion example in SKILL.md).
Ingestion points: The linter reads source code from the project directory (files like mod.rs, class names, etc.).
Boundary markers: None specified in the provided templates; the agent is encouraged to follow the instructions in the 修正手順 (Repair Steps) section directly.
Capability inventory: The agent is granted the capability to move files, delete files, and modify source code based on these instructions.
Sanitization: There is no evidence of sanitization or validation of the lint messages before the agent processes them as instructions.

custom-linter-creator