The Agent Skills Directory

[SAFE]: The skill provides structured instructions for copywriting tasks without requesting dangerous permissions or external network access. All operations are confined to text generation based on provided principles.
[SAFE]: External references are limited to local markdown files within the skill's own repository (e.g., references/copy-frameworks.md, references/natural-transitions.md), posing no risk of remote code execution or unauthorized data access.
[SAFE]: The instruction to read .agents/product-marketing.md is a standard practice for gathering context within the project workspace and does not involve accessing sensitive system files or credentials.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input and reads local project context files, which technically constitutes an indirect prompt injection surface. However, since the skill has no dangerous capabilities (such as command execution, network operations, or file system writes), this surface is considered safe in this context.
Ingestion points: User prompts, .agents/product-marketing.md (mentioned in SKILL.md)
Boundary markers: Absent
Capability inventory: None (text generation only)
Sanitization: Absent

copywriting