The Agent Skills Directory

[PROMPT_INJECTION]: No evidence of malicious prompt injection was found. The skill includes positive safety controls, such as requiring human confirmation before executing complex math modeling or formal proofs.
[CREDENTIALS_UNSAFE]: No hardcoded credentials, API keys, or instructions to handle secrets unsafely were detected.
[OBFUSCATION]: The skill uses clear, readable Markdown and standard Chinese/English text. No Base64, hex encoding, zero-width characters, or homoglyph attacks were identified.
[DATA_EXFILTRATION]: There are no network operations, curl/wget commands, or sensitive file path accesses that could lead to data exfiltration.
[REMOTE_CODE_EXECUTION]: The skill does not perform remote code execution. References to specific tools like TLA+ and sub-skills (e.g., pstack:create-formal-verification-skill) are used within the context of a design recommendation workflow rather than unauthorized script execution.
[COMMAND_EXECUTION]: No dangerous shell commands or privileged execution patterns (like sudo) were found.
[INDIRECT_PROMPT_INJECTION]: While the skill is designed to process external design specifications and domain models, it does not include network ingestion of untrusted data or direct execution of external strings, maintaining a safe posture.
[DYNAMIC_CONTEXT_INJECTION]: The skill does not use the dynamic exclamation-backtick (!command) syntax for silent load-time execution.

formal-guard