The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/prompt_tester.py contains a feature that executes external shell commands. The run_runner function uses subprocess.run to execute a command string constructed from user-provided templates.
Evidence: The --runner-cmd argument allows users to specify an arbitrary command template like 'my-llm-cli --prompt {prompt} --input {input}'. The script then uses shlex.split and subprocess.run to execute this command for every test case.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external prompt files and test case data (testcases.json). If these files are sourced from an untrusted third party, they could contain malicious command injections targeting the --runner-cmd placeholders or the prompt versioner's diffing logic.
Ingestion points: scripts/prompt_tester.py reads data from --cases-file and --input JSON/stdin.
Capability inventory: Use of subprocess.run in scripts/prompt_tester.py for external LLM runners.
Boundary markers: None detected; content is interpolated directly into command strings.
Sanitization: While shlex.split is used, the command template itself is provided by the user, and data is interpolated into it without specialized sanitization for shell safety.

prompt-engineer-toolkit