The Agent Skills Directory

[SAFE]: No malicious patterns were detected across any of the analyzed files. The skill operates as a specialized writing tool.
[PROMPT_INJECTION]: No instructions attempting to bypass safety filters or override core agent behavior were found. The 'SKILL.md' uses a robust routing logic to ensure context is loaded correctly and instructions emphasize not inventing facts or results.
[DATA_EXFILTRATION]: There are no network-capable commands (e.g., curl, wget) or patterns that suggest data is being sent to external servers. The skill handles sensitive author metadata (emails, affiliations) exclusively for the purpose of generating submission materials.
[REMOTE_CODE_EXECUTION]: The skill does not involve downloading external scripts, packages, or executing remote code. It utilizes local markdown fragments and standard LaTeX templates for its operations.
[COMMAND_EXECUTION]: No dangerous shell commands, privilege escalation attempts (e.g., sudo), or persistence mechanisms (e.g., crontab) were identified.
[OBFUSCATION]: The content is provided in clear text. No hidden Unicode characters, Base64 encoding of executable content, or other obfuscation techniques were found.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided content such as research notes and drafts. This creates a surface for indirect prompt injection; however, the skill possesses no high-risk capabilities (like network access or file system modification) to exploit. The 'workflow.md' also mandates a 'confirmation gate' where the agent aligns with the user on the core argument before drafting, serving as a safety checkpoint.

nature-writing