I’ve had great success with these adversarial skills (using GitHub Copilot) so far so wanted to share them. I haven’t been able to use the new Claude (or other) security plugins but I suspect these skills achieve comparable results, perhaps even superior results if combined with them but that’s pure speculation. I’m not sure it would even make sense or be practical to use in combination. Use them with or without memory-journal-mcp. If you let me know how they do for you, I’d appreciate it. The adversarial planner is so useful I created a prompt for it in memory-journal-mcp as it helps the agent write more secure, performative code from the get-go, saving time in fixes/revisions. Don’t be surprised if the security review finds lots of issues in a pass, none or few in the next, and then lots in the next. Anthropic’s Claude reports the same thing. It’s a time-consuming process, hence the power of the adversarial planner.
Hey, thanks for sharing. This looks like a really interesting set of skills, especially adversarial-planner.
I moved the thread to the Built for Cursor category since more users there actually connect these builds and MCP servers, so you’ll likely get more useful feedback.
One thing. You said you ran these skills on GitHub Copilot. If you test them in Cursor too via MCP config and skills, come back with results, especially if adversarial-security behaves differently between runs. The community really likes comparisons like that.
