The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions are focused strictly on legal analysis and clause extraction. There are no attempts to override system prompts, bypass safety filters, or use adversarial role-play techniques.
[DATA_EXFILTRATION]: The skill configuration specifies that data residency is local and PII egress is disabled. No network tools (e.g., curl, fetch) or hardcoded credentials were detected. The environment is constrained to prevent data leakage.
[REMOTE_CODE_EXECUTION]: No remote code patterns, external script downloads, or package installations (npm/pip) are present in the skill definition.
[COMMAND_EXECUTION]: The skill does not use shell commands, subprocesses, or dynamic script generation. The allowed-tools is limited to Read, preventing unauthorized system modifications.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (contracts), which is a theoretical injection surface. However, because the agent is restricted to Read access and lacks tools for network communication or file system writes, the risk of an exploit being triggered by malicious contract text is negligible.
[OBFUSCATION]: The content is written in clear, plain-text English and Markdown. No Base64, hex encoding, zero-width characters, or hidden text patterns were found.

clause-checklist-en