The Agent Skills Directory

[SAFE]: The skill implements strong safety boundaries by requiring explicit user confirmation for all directory paths (source and target) before any analysis or operations occur. It specifically prevents the agent from inferring or creating directories autonomously, which mitigates risks of unauthorized file system access or path traversal.
[PROMPT_INJECTION]: The skill uses "Prime Directives" and "Hard Rules" to maintain high fidelity and prevent the model from deviating from the user's explicit task. These rules (e.g., "Translate, do not redesign", "Never fix a source bug") serve as an internal control against prompts or instructions that might attempt to steer the agent toward unsafe or unauthorized "improvements".
[COMMAND_EXECUTION]: The workflow involves running existing and ported test suites to verify translation accuracy. This is a standard and necessary component of code porting tasks. The skill does not automate the installation of unverified external packages or execute remote scripts from unknown sources.
[SAFE]: The skill employs a structured "Adversarial Review" process using a fresh-context sub-agent. This is a security best practice that prevents the primary agent's context from biasing the review, ensuring that any logic errors or deviations (which could include malicious code introduced during translation) are more likely to be detected.

porting-codebases