The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a procedure for critiquing user-provided design work and intent. While this involves processing untrusted external data, the skill does not possess any dangerous capabilities such as command execution, network access, or file-writing tools. Ingestion points: User-provided descriptions and work samples defined in Section 2. Boundary markers: The skill specifies a structured output format in Section 8 but lacks explicit delimiters for incoming untrusted data. Capability inventory: No scripts, network access, or file-writing tools are included in this skill. Sanitization: No specific sanitization or filtering logic is described. The lack of dangerous tools makes this surface safe in this context.
[SAFE]: No evidence of obfuscation, remote code execution, persistence, or privilege escalation was found in the skill metadata or instructional content. The skill references a local file path for citation, which is a legitimate documentation reference and not a sensitive system file.

typography-critique