The Agent Skills Directory

[COMMAND_EXECUTION]: The script prompt-engineer-toolkit/scripts/prompt_tester.py utilizes subprocess.run() to execute commands. These commands are constructed using a template string provided via the --runner-cmd argument, which incorporates prompt text and test case inputs. While the script uses shlex.split() to mitigate shell injection, it essentially functions as a wrapper for executing arbitrary command-line tools.
[DYNAMIC_EXECUTION]: The toolkit is designed to automate the testing of LLM outputs by running them through local CLI tools. This enables the agent to trigger execution paths based on the content of prompts and test cases, which are often derived from untrusted or user-supplied data.
[INDIRECT_PROMPT_INJECTION]: The skill has a broad attack surface for indirect prompt injection as it is designed to ingest and process various forms of untrusted data, including HTML files, marketing drafts, and JSON test cases.
Ingestion points: prompt-engineer-toolkit/scripts/prompt_tester.py (JSON test cases), seo-audit/scripts/seo_checker.py (external HTML files/URLs), content-production/scripts/content_scorer.py (text drafts).
Boundary markers: None. The processing scripts do not implement explicit delimiters or 'ignore instructions' warnings when handling content that is subsequently passed to other analytical functions or the LLM.
Capability inventory: The suite includes scripts capable of reading files, fetching remote content via urllib, and executing shell commands (in the case of the prompt tester).
Sanitization: There is no evidence of robust sanitization or filtering of external content before it is interpolated into prompts or used in command construction.

marketing-skills