| name | consult-minimax |
| description | Consult MiniMax M3 via Ollama Cloud (`minimax-m3:cloud`) for peer review and second-opinion analysis. Use when the user says `/consult-minimax`, "ask minimax", "minimax second opinion", or wants MiniMax's beautiful structured analysis on a finding, plan, or strategy claim. |
/consult-minimax — MiniMax M3 (Ollama Cloud) consult
MiniMax M3 is hosted on Ollama Cloud as minimax-m3:cloud. It produces clear, structured, often beautiful analyses — the user has flagged it as a particularly strong peer-review counterparty.
Prerequisites
ollama installed (/usr/local/bin/ollama on this machine).- Ollama Cloud auth (the
:cloud suffix routes to managed inference; if ollama run minimax-m3:cloud returns an auth error, ollama signin first).
Usage
/consult-minimax <question or prompt>
/consult-minimax --file <path-to-prompt.md>
Invocation (Bash)
Quick inline question
ollama run minimax-m3:cloud "<your full prompt here>" 2>&1 | tee /tmp/consult-minimax-$(date -u +%Y%m%dT%H%M%SZ).txt
From a file (recommended for any prompt > 200 words)
ollama run minimax-m3:cloud "$(cat /path/to/prompt.md)" 2>&1 | tee /tmp/consult-minimax-$(date -u +%Y%m%dT%H%M%SZ).txt
Structured peer-review pattern (for double-checking a finding)
cat <<'EOF' | ollama run minimax-m3:cloud 2>&1
You are MiniMax M3 acting as an independent quant peer reviewer for a multi-AI audit system.
<paste finding here — include the live-DB numbers, OOS-split, concentration, fat-tail check>
1. Identify any axis the reviewer didn't check (e.g. survivorship bias, regime dependency, slippage realism, look-ahead in resolver, broker-side fill assumption).
2. For each axis, state CORROBORATE / REFUTE / NEEDS_MORE_EVIDENCE and one-sentence rationale.
3. End with a single VERDICT: PROMOTE / HOLD / REFUTE.
Be terse and structural. No marketing language. Mirror the reviewer's evidence style.
EOF
When MiniMax is a better fit than other consult-* skills
- Peer-reviewing a quant/strategy claim where you want a structural breakdown (multi-axis verdict table).
- Cross-checking another AI's "VALIDATED ✅" — see memory
project-multi-agent-storm-2026-06-05 for why this matters (Cloud-Minimix VRP failure pattern).
- Strategy survivorship / regime-dependency analysis.
When to prefer another consult-*
- Live web search needed →
/consult-grok (xAI has web access).
- Fast/cheap throwaway second opinion →
/consult-gemini or /consult-cloudflare (CF models).
- Code review with diff context →
/consult-codex or /consult-cursor-agent.
Notes
- The
:cloud suffix on the model name routes through Ollama Cloud's managed inference, not local hardware (good — local MiniMax M3 weights would be >100GB).
- Output is plain text; if you want structured JSON back, prefix the prompt with
Respond in this exact JSON schema: {...} and MiniMax will usually comply.
- Cost: bills against the user's Ollama Cloud account. For routine session-level peer-review (1-2 consults per investigation) this is negligible; do not fan out 100 prompts.
- For multi-AI fan-out including MiniMax, use
/consult-PROXY or /consult-multi if it has a MiniMax provider, otherwise drive ollama in a parallel/xargs loop.
