The Agent Skills Directory

[PROMPT_INJECTION]: No evidence of prompt injection or attempts to bypass safety filters was found in the instructions or examples. The skill focuses on structured data transformation with a deterministic methodology.
[DATA_EXFILTRATION]: No network operations, hardcoded credentials, or access to sensitive system file paths were detected. The skill uses standard Read and Write tools for its primary function.
[COMMAND_EXECUTION]: The included Python script (scripts/validate.py) is used for local validation of the skill's structure. It does not execute arbitrary shell commands or external processes.
[REMOTE_CODE_EXECUTION]: The skill does not download or execute remote code. All references are local within the skill directory.
[OBFUSCATION]: No obfuscated strings, base64-encoded commands, or hidden characters were found in any of the files.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user communication as its primary input (ingestion point: Phase 1 Parse). While it lacks explicit boundary markers to ignore embedded instructions, the multi-phase extraction and deterministic template application (Phase 3) act as significant mitigation by restricting output to a specific schema. The capability inventory includes Read and Write tools, which are appropriately scoped for the task. No sanitization is explicitly defined, but the 'proposition extraction' logic effectively filters the input before re-formatting.

professional-communication