The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains specific instructions for the AI to avoid standard 'AI-style' writing patterns and to follow 'anti-AI' rules. These are task-specific stylistic constraints (e.g., avoiding 'not... but...' structures) rather than malicious attempts to bypass the underlying LLM's safety filters or extract system prompts.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill interacts with local paths such as workspace/insights/*.md, workspace/stories/, and _posts/. These are standard functional paths for content generation. No access to sensitive system files (e.g., SSH keys, environment variables) or unauthorized external network requests were identified.
[COMMAND_EXECUTION]: The skill mentions using git commit + push to publish content to GitHub Pages. This is a standard workflow for the stated purpose of blogging and uses a well-known, trusted service (GitHub).
[EXTERNAL_DOWNLOADS]: The skill references other internal agent skills like zhiforge and salt-story, but does not attempt to download or execute code from untrusted third-party URLs.

article-forge