lope-ask

name	lope-ask
description	Ask every configured validator the same question and return N independent answers — one per model. No sprint framing, no phases, no verdict parsing. Use for any multi-perspective query: 'what do these 5 CLIs think about X?', 'compare how 3 models solve this problem', 'get a second opinion across models before I commit'. Works for any domain (engineering, research, writing, business).

Lope Ask

Fan out one question to every configured validator in parallel. Collect N raw answers. Print them side-by-side. No sprint, no phases, no majority vote.

This is the lightest lope surface — use it when the user wants multiple perspectives on a single prompt, not a structured multi-phase run.

Invocation

Two paths — you must handle both:

1. Explicit slash command. User types /lope-ask "What's the best way to index JSONL for fast lookup?". Route to lope ask.

2. Natural language. User says something like:

   • "Ask gemini, claude, and opencode what they think about X"
   • "What do the other models say about this?"
   • "Get me a second opinion across models"
   • "Let's check if all the CLIs agree on X before I commit"

   Recognize the shape and invoke lope ask "<their question>" on their behalf. Quote the question carefully.

Command shape

lope ask "<question>"                   # fan out to all configured validators
lope ask "<question>" --context "<ctx>" # prepend context to every call
lope ask "<question>" --json            # machine-readable JSON output
lope ask "<question>" --validators claude,gemini  # override the pool
lope ask "<question>" --timeout 60      # per-validator timeout

v0.7 superpowers (opt-in)

# Synthesis — primary rolls N answers into one executive summary
lope ask "Should we adopt X?" --synth
lope ask "Should we adopt X?" --synth --anonymous   # strip validator names

# Brain-aware ask (Makakoo OS only)
lope ask "What should we do next?" --brain-context "lope roadmap" --synth
lope ask "Is this still the plan?" --brain-context "auth decisions" --brain-log

--synth produces a 4-section executive summary (Consensus, Disagreements, Highest-risk item, Recommended action) plus optional Follow-up questions. --anonymous rewrites validator names to Response A/B/C so the synthesizer cannot bias on identity.

When to use ask vs negotiate

• ask: user has a single question or wants multi-model perspectives. Short-to-medium prompts. No structured deliverable expected.
• negotiate: user wants a plan — a sprint doc with phases, deliverables, and verdicts. Multiple rounds of revision expected.

If the user says "plan the X", "draft a sprint", or "negotiate" → lope negotiate. If the user says "ask", "what do they think", "check with the other CLIs" → lope ask.

Output shape

Human-readable by default. Each validator's answer is wrapped in a ━━━ <name> ━━━ header:

━━━ claude ━━━
<claude's answer>

━━━ gemini ━━━
<gemini's answer>

━━━ opencode ━━━
<opencode's answer>

Pass --json for a JSON array of {validator, answer, error} objects — useful for piping into other tools or for agentic chains.

Errors

If a validator times out or errors, its block shows [ERROR] <message> instead of the answer. The run continues — one slow/broken CLI does not blank the whole request. If every validator fails, exit code 1 and a "No validators available" hint.

Cost awareness

lope ask sends the same prompt to N validators in parallel. Each call costs roughly what a single CLI call costs. If the user has 5 validators configured, expect 5× the tokens of a single call. For very long prompts or long context, pass --validators claude,gemini to cap the fan-out to a specific subset.

Runtime safety (v0.14.0)

Use --run-timeout for the whole command; --timeout remains a per-provider-call ceiling. Inspect the emitted request plan before large work and prefer compact evidence. Automatic shaping is bounded; use chunking only when the forecast fits --max-calls and --max-chunks. For abandoned work, run lope jobs list and preview lope jobs reap --dry-run. Never use pkill -f, killall, or process-name matching.