Multi-LLM Consult

multi-llm-consult is the cross-model consultation workflow in Cortex. Use it when you want a second opinion, an alternative plan, an independent review, or a delegated analysis from another model family.

The generated skill-backed command is:

/ctx:multi-llm-consult

When To Use It

Use this workflow when:

  • you want another model’s perspective before acting
  • you want to compare plans
  • you want an external review or critique
  • you want to delegate a bounded analysis task to another model

The high-value pattern is:

  1. prepare the task
  2. export or summarize only the needed context
  3. sanitize anything sensitive
  4. ask the other model a narrow question
  5. compare the result against repo reality before acting

Good Inputs

Good consult requests are:

  • specific
  • bounded
  • sanitized
  • paired with the right context bundle

Example purposes from the skill:

  • second opinion
  • plan check
  • independent review
  • delegated task

Pair It With Export

This workflow becomes much more useful when paired with cortex export:

cortex export context - --include core --include skills > /tmp/consult-context.md
cortex skills context --no-write > /tmp/consult-skills.md

That gives you a compact bundle for another model without dumping unrelated repo context into the prompt.

Pair It With Skills

A strong consultation flow is:

cortex skills recommend
cortex skills context --no-write

Then use the relevant skills, or pass that skill context into the consult request so the other model understands the intended working style.

Best Uses

Second opinion

You already have a plan and want another model to poke holes in it.

Independent review

You want an outside perspective on risk, correctness, or blind spots.

Delegated analysis

You want another model to answer a narrow question while you keep the main flow moving.

Important Constraints

  • Treat the external answer as advisory.
  • Verify against the current repo and CLI surface before acting.
  • Do not send sensitive data unless you have explicitly decided it is safe.
  • Prefer narrow prompts over broad “analyze everything” prompts.