The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The documentation and roadmap files (README.md, MARKETING-EXECUTION-PLAN.md) recommend installing agent skills and dependencies from an untrusted third-party GitHub repository (alirezarezvani/claude-skills).
[COMMAND_EXECUTION]: The script prompt-engineer-toolkit/scripts/prompt_tester.py uses the subprocess.run() function to execute shell commands. These commands are generated from a template provided by the user via the --runner-cmd argument, allowing for the execution of arbitrary system commands.
[DYNAMIC_EXECUTION]: The prompt_tester.py tool dynamically assembles command-line strings using Python's .format() method to inject prompt and input data into a runner command. This capability is used to execute external LLM CLI tools but creates a vector for code or command injection if inputs are not strictly controlled.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface for indirect prompt injection.
Ingestion points: Multiple scripts ingest untrusted data from external files, most notably prompt_tester.py which reads testcases.json.
Boundary markers: There are no boundary markers or instructions to the agent to ignore embedded commands in the processed content.
Capability inventory: The toolkit includes subprocess.run (command execution) and urllib.request.urlopen (network access for auditing URLs and sitemaps), which can be exploited if an attacker influences the data being processed.
Sanitization: Although shlex.split is used in prompt_tester.py, it does not sanitize the contents of user-supplied variables before they are formatted into the execution string.

marketing-skills