The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions focus on improving text quality and removing 'AI smell'. No instructions were found that attempt to bypass safety filters, override agent behavior, or extract system prompts. The commands are instructional and helpful for writing tasks.
[DATA_EXFILTRATION]: There are no network-related commands such as curl, wget, or fetch. No sensitive file paths (e.g., SSH keys, AWS credentials) are accessed. The skill operates entirely within the provided text context.
[REMOTE_CODE_EXECUTION]: No remote scripts are downloaded or executed. The skill does not contain any commands that pipe external content to a shell or interpreter.
[COMMAND_EXECUTION]: While the skill mentions development workflows (e.g., using textlint or prettier), these are provided as guidelines for the user to follow in their own environment. There are no automated shell commands or dynamic context injections (!command) within the instruction files.
[CREDENTIALS_UNSAFE]: No hardcoded API keys, tokens, or passwords were detected in any of the 56 files. The content consists of markdown text guidelines.
[OBFUSCATION]: The files were scanned for Base64, hex encoding, zero-width characters, and homoglyphs. No obfuscated content or hidden URLs were found.
[EXTERNAL_DOWNLOADS]: The documentation references well-known services and trusted academic databases (e.g., Google Books, JSTOR, Zenn, GitHub) for research purposes. No risky or unverifiable external dependencies are downloaded.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided text for improvement. While this is an ingestion point for untrusted data, the skill has no dangerous capabilities (no file writing, no network access, no execution) that could be exploited. The guidelines even include advice on identifying and citing sources safely.

writing-effective-prose