Please assess my current AI setup before recommending changes. Keep my existing instructions and tools.
1. Ask which AI apps and OS I use, which persistent project instructions and numbering I already have, which integrations actually work, which workflows run automatically and what I want to improve. Ask for one completed task and how it was checked. If I report an MCP problem, ask for the exact redacted error and whether it ever worked.
2. With my permission, check only the relevant non-secret project files and available tool metadata. Do not read credentials, scan my whole computer, install anything, send messages or publish.
3. Assess five areas: execution, integrations, persistent rules, automation and result verification. Use 0–3 only when supported: 0 = no capability confirmed by the user; 1 = one basic demonstrated step; 2 = repeatable demonstrated capability; 3 = demonstrated end-to-end capability with checks. Mark unknown as unknown, never zero. Distinguish reported from independently checked. Do not calculate a total while any area is unknown. This is a working rubric, not a certification, personal rating or Club role.
4. Run a tiny optional test only if I agree: produce one useful small artifact for my chosen task in the current workspace, verify it, then give a concise result, evidence, critique, ProblemOS P/U/L/R/N and 4 numbered next actions. On the next substantive answer continue the action numbers. State whether numbering was manual or persistently checked.
5. Give a capability table with evidence and uncertainty, one bottleneck and one optional recommendation. If my current setup already works, say so. Read-only assessment must not turn into a mandatory setup.