/deliberate

Run a Council deliberation on the user's question using the deliberation-engine skill.

Important: The /deliberate command always runs the full Council pipeline. It sets routing_decision: "explicit_command" and skips the smart-router's Pre-Step decision. This is intentional — when a user explicitly calls /deliberate, they want the full multi-model experience regardless of what the router would recommend.

Parse any flags from the arguments:

• --size small|medium|large → 2, 4, or 6 models (default: medium)
• --fast → skip Stage 2 peer review, go straight from collect to synthesis
• --deep → MoA two-round collection — models see all Round 1 responses before responding again
• --thorough → add a Collective Improvement round between Stage 1 and Stage 2
• --template <name> → use a pre-built deliberation template (code-review, architecture, writing-critique, research-synthesis, compliance-check, business-strategy)
• --models model1,model2,... → override the default model lineup
• --budget 0.15 → max cost in USD for this deliberation

The query is everything in $ARGUMENTS that is not a flag.

Load the deliberation-engine skill from ${CLAUDE_PLUGIN_ROOT}/skills/deliberation-engine/SKILL.md and follow its execution steps (starting at Step 1, since Pre-Step routing is bypassed).

If this is the first deliberation in this session, verify the OpenRouter connection is active by calling RUBE_SEARCH_TOOLS for OPENROUTER_CREATE_CHAT_COMPLETION. If not connected, call RUBE_MANAGE_CONNECTIONS with toolkit openrouter and present the auth link to the user.

After the deliberation completes:

1. Present the formatted output as specified in the deliberation-engine skill (Step 7)
2. Log the outcome to pattern memory (Step 8) — requires Write tool for council-memory.json
3. Watch for user rating in the next 1-2 messages (Step 8c)