The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/prompt_tester.py supports executing external runner commands via the --runner-cmd argument. This is used to interface with external LLM CLI tools for evaluation. The implementation uses shlex.split and subprocess.run without shell=True, adhering to security best practices to prevent command injection via malicious arguments.\n- [INDIRECT_PROMPT_INJECTION]: The skill handles untrusted data by design, evaluating prompts and test cases from external files. While this creates a potential surface for indirect prompt injection if the inputs contain malicious instructions, the risk is inherent to the tool's purpose as a testing utility.\n
Ingestion points: Reads content from prompts/a.txt, prompts/b.txt, and testcases.json in scripts/prompt_tester.py.\n
Boundary markers: None; the tool interpolates file content into the execution context.\n
Capability inventory: scripts/prompt_tester.py uses subprocess.run to execute external processes.\n
Sanitization: Uses shlex.split to ensure input content is treated as literal arguments rather than shell commands during execution.

prompt-engineer-toolkit