The Agent Skills Directory

[SAFE]: The skill instructions and evaluation files focus exclusively on stylistic editing and auditing for specific linguistic markers (e.g., puffery, hedging, machine-like vocabulary). No malicious patterns or security risks were identified.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an ingestion surface as it is designed to process arbitrary text provided by the user (emails, posts, READMEs). However, the risk is negligible due to the absence of dangerous capabilities.
Ingestion points: Processes text provided in the assistant's context or pasted by the user as specified in SKILL.md.
Boundary markers: Explicitly instructs the agent to return cleaned text as an artifact while keeping harness commentary and findings outside the artifact boundary to prevent accidental copying of meta-data.
Capability inventory: Limited strictly to text rewriting and auditing; no subprocess calls, file operations, or network tools are utilized.
Sanitization: Instructions in SKILL.md mandate that non-prose elements such as code blocks, shell commands, URLs, and pricing figures must be left 'byte for byte', providing a strong preservation policy for technical content.

unslop