The Agent Skills Directory

[SAFE]: The skill consists entirely of Markdown-based theoretical guides and templates. No executable scripts (.py, .js, .sh), tool permissions (allowed-tools), or network-accessible logic are present within the skill files.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided project context (vision statements, keywords) to generate a style guide template.
Ingestion points: Step 1: Vision & Context in SKILL.md.
Boundary markers: Absent; uses standard Markdown placeholders.
Capability inventory: None. The skill has no access to file-system writes, network operations, or shell execution.
Sanitization: Absent. However, the lack of capabilities prevents any meaningful exploitation of untrusted input.
[COMMAND_EXECUTION]: Previously identified risks regarding backtick execution are confirmed false positives. The backticks are standard Markdown code fence delimiters (```markdown) used in SKILL.md to format the output template for the user, not for shell command execution.
[METADATA_POISONING]: The skill-report.json file included in the skill identifies 'forbidden colors' and 'system reconnaissance' as potential risks. A contextual review confirms these are benign art-direction terms (e.g., defining palettes or color psychology) and design constraints, not attempts to bypass security filters.

art-style-creator