consult - AI Consultation CLI

Synopsis

bash

consult -m <model> <subcommand> [args] [options]

The -m / --model flag is always required.

Models

Model	Alias	Speed	Approach
`gemini`	`pro`	~120-150s	Pure text analysis, fast
`codex`	`gpt`	~200-250s	Shell command exploration, thorough
`claude`	`opus`	~60-120s	Balanced tool use

Subcommands

bash

consult -m gemini pr 42              # Review a pull request
consult -m codex spec 42             # Review a specification
consult -m gemini plan 42            # Review an implementation plan
consult -m claude general "query"    # General question (quoted string)

Options

bash

-n, --dry-run              # Show command without running
-t, --type <type>          # Review type (spec-review, plan-review, impl-review, pr-ready, integration-review)
-r, --role <role>          # Custom role from codev/roles/ (without .md extension)

Review Types (--type)

Type	Stage	Use Case
`spec-review`	conceived	Review spec completeness
`plan-review`	specified	Review implementation plan
`impl-review`	implementing	Review code implementation
`pr-ready`	implemented	Final check before PR
`integration-review`	committed	Architect's integration review

Review type prompts live in codev/consult-types/ and can be customized.

Parallel Consultation (3-Way / cmap)

Run all three models in parallel for thorough reviews:

bash

consult -m gemini spec 42 &
consult -m codex spec 42 &
consult -m claude spec 42 &
wait

Or from Claude Code, use cmap pattern: three parallel background Bash calls.

Custom Roles

bash

consult -m gemini --role security-reviewer pr 42
consult -m codex --role architect general "Review this design"

Roles are markdown files in codev/roles/. Create your own by adding .md files there.

Common Mistakes

•The -m flag is required — consult pr 42 will fail without it
•PR numbers are GitHub PR numbers, not spec numbers
•Spec/plan numbers match the 0042 prefix in filenames (pass just 42, not 0042)
•General queries must be quoted: consult -m gemini general "your question here"