The Agent Skills Directory

[PROMPT_INJECTION]: No malicious prompt injection patterns were detected. The instructions include standard hierarchical logic (e.g., instructions in persona files taking precedence over general skill instructions) which are used for behavioral consistency rather than bypassing agent safety protocols. There are no attempts to force the agent to ignore its core safety guidelines or extract system prompts.
[DATA_EXFILTRATION]: No data exposure or exfiltration risks were identified. The skill does not access sensitive local file paths (such as .ssh or .aws), does not contain hardcoded credentials, and performs no network operations (curl, wget, or fetch).
[OBFUSCATION]: The skill uses clear text throughout. There were no detections of Base64 encoding, hex escapes, zero-width characters, or homoglyph attacks intended to hide malicious commands or URLs.
[REMOTE_CODE_EXECUTION]: No remote code execution patterns were found. The skill does not download external scripts or execute piped shell commands from remote sources.
[DYNAMIC_EXECUTION]: The skill does not utilize dynamic execution functions such as eval(), exec(), or runtime compilation. All content is static Markdown used for text generation.
[PRIVILEGE_ESCALATION]: No privilege escalation attempts were detected. The skill does not use sudo, chmod, or any Windows equivalent to modify system permissions.
[INDIRECT_PROMPT_INJECTION]: While the skill processes user input, it lacks any dangerous capabilities (like network access or file writes) that could be exploited via indirect injection. The analysis confirms a lack of vulnerability surfaces for this category.
[DYNAMIC_CONTEXT_INJECTION]: There are no instances of dynamic shell execution placeholders (e.g., !command) in the SKILL.md or referenced files.

dojo-nassim-taleb