The Agent Skills Directory

[SAFE]: The skill is composed exclusively of Markdown instructions and YAML configuration. No scripts (Python, JavaScript, Shell) or binary executables are included in the skill package.
[DATA_EXFILTRATION]: There are no commands or instructions that attempt to access sensitive system files (e.g., SSH keys, AWS credentials) or perform network requests to external domains.
[PROMPT_INJECTION]: The instructions contain 'Non-Negotiable Rules' and 'Red Lines' designed to enforce academic integrity and prevent the agent from fabricating information. These are safety-positive constraints and do not attempt to bypass the underlying LLM's safety guardrails.
[REMOTE_CODE_EXECUTION]: No remote code execution patterns, package installations, or dynamic code generation were detected. The skill relies on 'always_load' files which are local relative paths.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided academic text. While this creates a surface for indirect prompt injection, the skill includes specific auditing protocols ('Prose Quality Gate' and 'Claim Strength Audit') that act as sanitization steps to verify the input text against evidence-based rules before performing rewrites.
[OBFUSCATION]: Analysis of all text content and metadata revealed no base64, hex-encoded, zero-width characters, or hidden acrostic patterns.

academic-polishing