The Agent Skills Directory

[SAFE]: The skill uses a local configuration file (.agents/editorial-context.md) to manage its state and record editorial findings. This is a standard practice for agent skills and does not involve data exfiltration or hardcoded credentials.
[PROMPT_INJECTION]: The skill processes untrusted user documents such as manuscripts and essays for analysis. This creates a surface for indirect prompt injection. Ingestion points: User-provided text content enters the agent context during the diagnostic workflow. Boundary markers: No specific delimiters or safety instructions are defined to separate user text from agent instructions. Capability inventory: The agent has the ability to write findings to the local filesystem and suggest text revisions. Sanitization: There is no evidence of sanitization or filtering of the processed content. While this allows for potential manipulation via the analyzed text, the overall risk is low as the skill lacks network access or privilege escalation capabilities.

editorial-structural