The Agent Skills Directory

[SAFE]: The skill is a legitimate text-editing utility. All instructions and references are consistent with its stated purpose of improving writing quality.
[PROMPT_INJECTION]: A static detector flagged potential concealment in references/ai-tells.md and SKILL.md. This refers to instructions for the agent to remove conversational chatbot filler (e.g., 'I hope this helps!') and to provide clean output without unnecessary commentary when requested. These are standard stylistic requirements for a writing tool and do not constitute a malicious attempt to hide actions from the user.
[INDIRECT_PROMPT_INJECTION]: The skill operates on untrusted text provided by users, creating a potential surface for indirect prompt injection.
Ingestion points: Reads text via the Read tool or direct input.
Boundary markers: None explicitly defined in the instructions.
Capability inventory: Uses Write and Edit tools to modify files based on processed text.
Sanitization: The skill includes strict 'Final Fidelity Check' rules that require the agent to ensure no new facts or opinions are added and that technical terms remain intact. This reduces the risk of the agent obeying instructions embedded in the user-provided text.
[EXTERNAL_DOWNLOADS]: The skill mentions external documentation from GitHub (Hakku/finnish-humanizer), Wikipedia (Wikipedia:Signs_of_AI_writing), and MIT licensing. These are benign references for methodology and do not involve dynamic code execution or automated downloads.

humanize-writing