agent-native-architecture

[Skill Scanner] Detected system prompt override attempt This is a documentation/architecture skill that correctly describes patterns for building agent-native applications. It is conceptually consistent: the capabilities it describes (atomic file tools, shell, dynamic discovery, self-modification) align with the stated purpose of enabling agents to act with parity. There is no explicit malicious code, hardcoded credentials, obfuscated payloads, or download-execute chains in the provided content. However, the recommended primitives (especially bash and self-modification) carry substantial operational risk if implemented without strict guardrails (approval workflows, permission scoping, least privilege, auditing, and explicit human-in-the-loop controls). I assess low likelihood of this document itself being malicious, but moderate security risk if its recommended primitives are enabled in production without safeguards. LLM verification: This SKILL.md is an architectural guide advocating giving agents powerful primitive tools (read/write files, bash, looped autonomous operation, and even self-modification). There is no direct malicious code or remote download-execute pattern in the file, but the recommended capabilities are high-risk if applied without strict controls. The primary concerns are: broad filesystem and shell access, the ability for agents to edit system prompts or repo files (self-modification), lack of explicit gat