The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses instructional language to guide the agent's behavior regarding code comments. It does not contain any attempts to bypass safety filters, override system instructions, or extract sensitive model information. The instructions are focused on best practices for software documentation.
[DATA_EXFILTRATION]: No sensitive file paths, network operations, or hardcoded credentials were found. The code examples provided are generic and represent standard development patterns.
[REMOTE_CODE_EXECUTION]: There are no patterns involving the download or execution of remote scripts. The skill does not reference any external URLs or perform shell operations.
[COMMAND_EXECUTION]: The skill does not invoke any shell commands or subprocesses. It contains documentation and natural language instructions only.
[OBFUSCATION]: The content is provided in clear text with no evidence of Base64 encoding, zero-width characters, homoglyphs, or other obfuscation techniques.
[INDIRECT_PROMPT_INJECTION]: While the skill instructs the agent to treat existing comments as "load-bearing memory" and "constraints," it does so within the context of preserving engineering intent. It does not facilitate the execution of instructions embedded in data. The risk of the agent obeying malicious instructions within comments is a general vulnerability of LLMs and is not specifically introduced or escalated by this skill's logic.
[SAFE]: The skill represents standard instructional content for improving code quality and maintainability. The frontmatter and body are consistent and benign.

smart-comments