The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses structured role-play (e.g., Skeptic, Constraint Guardian) to simulate a design review. These instructions are legitimate process constraints designed to bound agent behavior to specific domains of critique rather than attempting to bypass safety filters or override system prompts.- [DATA_EXFILTRATION]: While the skill uses file system tools (Read, Write, Glob, Grep) to manage designs and logs, there are no network-based tools allowed and no evidence of hardcoded credentials or sensitive system path access.- [REMOTE_CODE_EXECUTION]: No remote code execution patterns, package installations, or external downloads were detected. The skill relies solely on internal reference files.- [INDIRECT_PROMPT_INJECTION]: The skill inherently processes external design documents which could contain malicious instructions (Indirect Prompt Injection surface). However, the workflow mitigates this risk through a structured 'Decision Log' and 'Understanding Lock' phase that forces explicit identification of goals and assumptions before review begins. Given the primary purpose is design validation, this surface is considered a safe implementation of the intended functionality.

multi-agent-brainstorming