- devils-advocate-loop.md 42 lines · 2.3 KB
Devil's-Advocate Loop
Loop Engineering — put a dedicated critic in the loop. It keeps attacking the design until the proposal survives scrutiny or is knowingly accepted with its trade-offs.
🎯 Goal
Stress-test a design, RFC, or architectural decision by generating the strongest possible objections, then resolving or explicitly accepting each one — no objection silently ignored.
⏱️ Schedule
Suggested interval: 15m (or run until the design is settled).
▶️ Run it
/loop 15m "Act as a devil's advocate against the current design. Raise the strongest objections (scaling, security, cost, edge cases, maintainability). For each, either fix the design or record an explicit accepted trade-off with rationale. Continue until no serious objection is unresolved."🔁 Iteration steps
- Perceive — read the current design/decision.
- Reason — the
critical-thinkingagent generates the strongest counter-arguments. - Plan — for each objection, choose fix vs accept-with-rationale.
- Act — update the design or the decision log;
architect-reviewvalidates the revision. - Observe — re-run with
/pr-review; stop only when nothing serious is unresolved.
🛑 Stopping condition
Every serious objection is either resolved or recorded as an explicit, justified trade-off.
🧩 Referenced components
agent:expert-advisors/architect-review— validates each revision.agent:expert-advisors/critical-thinking— generates adversarial objections.command:git-workflow/pr-review— structured review of the design changes.
💡 Example
A proposed single-region database is challenged on availability; the loop either adds a replication plan or records "single-region accepted for MVP, revisit at scale" with reasoning.
Scanned by NVIDIA SkillSpector
Automated static analysis (no LLM) · v2.12.0 · 2026-10-17
SkillSpector looks for patterns such as shell commands, credential access, network calls or unpinned dependencies. An observation is something worth reading in the files before you install, not a confirmed vulnerability. Components that run code locally, like hooks, MCP servers and security tooling, naturally show more of them.
No observations.
Learn more about NVIDIA SkillSpector or how components are reviewed.