The Agent Skills Directory

[SAFE]: The skill performs legitimate writing style calibration by interviewing the user and updating local configuration files (assets/english-writing-style.md and assets/zh-tw-writing-style.md). It operates within its defined scope, focusing only on prose style preferences without accessing sensitive system data or executing external code.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it processes untrusted user input to generate instructions that are subsequently used as prompts for other AI tasks. However, this is the intended primary purpose of the skill, and instructions are in place to filter and structure the output.
Ingestion points: User responses during the preference interview defined in SKILL.md, including numbered choices or custom wording.
Boundary markers: The skill uses markdown headers and list structures in the target assets (assets/english-writing-style.md and assets/zh-tw-writing-style.md) to separate generated rules.
Capability inventory: The skill is authorized to read and update local markdown files within its own directory structure.
Sanitization: The skill includes explicit instructions to "Convert only the observed distinction into one concise, testable instruction," providing a conceptual filter between user input and the stored prompt.

calibrate-writing-style