The Agent Skills Directory

[PROMPT_INJECTION]: The expansion and execution plans (MARKETING-EXECUTION-PLAN.md and MARKETING-EXPANSION-PLAN.md) include instructions for a "content-humanizer" and "AI watermark scrubber" designed to bypass AI content detection systems, which can be used to evade safety filters or automated monitoring.- [REMOTE_CODE_EXECUTION]: The README.md facilitates installation using npx ai-agent-skills install, which downloads and executes code from a remote registry at runtime.- [EXTERNAL_DOWNLOADS]: The suite facilitates the download of skill packages (e.g., content-creator.zip) and utilizes npx for installation from the author's remote repository.- [DATA_EXFILTRATION]: The seo-audit/scripts/seo_checker.py tool uses urllib.request to perform network requests. While intended for fetching pages for audit, it provides an egress point for arbitrary network connections.- [COMMAND_EXECUTION]: The skill documentation instructs the agent to execute various Python scripts (seo_checker.py, brand_voice_analyzer.py, etc.) via the system shell to perform analysis tasks.- [INDIRECT_PROMPT_INJECTION]: The skill architecture ingests untrusted external data (HTML/markdown) for analysis without sanitization or explicit boundary markers.
Ingestion points: seo_checker.py parses HTML from user-provided URLs.
Boundary markers: No delimiters or "ignore instructions" warnings are present in the parsing or reporting logic.
Capability inventory: Access to shell execution and network fetching.
Sanitization: The HTML parser extracts raw text from tags without filtering, potentially allowing embedded instructions to influence the agent.

marketing-skills