The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/prompt_tester.py allows the execution of arbitrary external commands via the --runner-cmd argument. Although it uses shlex.split() and avoids shell=True to mitigate basic shell meta-character injection, the core functionality enables the agent to run any command provided in the template.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes prompts and test cases from external files without sanitization or boundary markers. A malicious file in a project could trick the agent into using a dangerous --runner-cmd template. Ingestion points: Reads from user-provided files in scripts/prompt_tester.py and scripts/prompt_versioner.py. Boundary markers: Absent; no delimiters or instructions are used to separate untrusted file content. Capability inventory: Executes subprocesses in scripts/prompt_tester.py and writes to a local JSONL file in scripts/prompt_versioner.py. Sanitization: Uses shlex.split to parse arguments, which mitigates shell injection but does not validate the command's safety or purpose.

prompt-engineer-toolkit