The Agent Skills Directory

[DYNAMIC_EXECUTION]: The prompt_tester.py script in the prompt-engineer-toolkit allows the agent to execute arbitrary external commands provided by the user via the --runner-cmd argument. While the script implements safety measures such as using shlex.split and avoiding shell=True in its subprocess.run calls, this functionality allows for the dynamic execution of external binaries during the prompt evaluation workflow.
[EXTERNAL_DOWNLOADS]: Multiple auditing scripts, such as sitemap_analyzer.py and conversion_audit.py, perform network requests using urllib.request.urlopen to fetch content from user-specified URLs. This network activity is used to retrieve sitemaps and HTML pages for auditing and is associated with the primary purpose of the skills, but could be leveraged for SSRF (Server-Side Request Forgery) if the agent attempts to access internal network metadata endpoints.
[INDIRECT_PROMPT_INJECTION]: The collection is designed to read and analyze various external data sources, including website HTML, blog drafts, and competitor reviews. Since this untrusted data is processed as plain text and interpolated into the agent's context without explicit sanitization or strict boundary markers, there is a risk that malicious instructions embedded in the analyzed content could influence the agent's behavior.
[NO_CODE]: The majority of the skills consist of high-quality markdown instructions and localized logic, with all Python tools using only the standard library. This significantly reduces supply chain risks as no external third-party dependencies are required for the scripts to operate.

marketing-skills