The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted source code as its primary function. While instructions focus on translation logic, this ingestion surface represents a theoretical attack vector for indirect prompt injection if the source data contains adversarial instructions. * Ingestion points: Source code input processed in SKILL.md. * Boundary markers: None specified in the guidelines. * Capability inventory: Documentation mentions executing code for verification via subprocess calls. * Sanitization: The skill does not provide instructions for sandboxing the analysis or execution of the code.
[DYNAMIC_EXECUTION]: The skill includes a documentation example for a verification script that dynamically executes code to compare outputs between language implementations. * Evidence: SKILL.md (Step 6: Test the Translation).
[COMMAND_EXECUTION]: The documentation provides examples of shell commands being invoked via subprocess to execute and compare code outputs for verification purposes. * Evidence: SKILL.md (Step 6: Test the Translation).
[SAFE]: No malicious behavior, obfuscation, or credential harvesting patterns were found. The skill's instructions and examples are consistent with its stated purpose of assisting in code translation and porting. The identified surface areas are inherent to the skill's utility and are documented as standard development practices.

code-translation