consult
Cross-tool AI consultation - get a second opinion from another AI CLI tool without leaving your current session.
Why
You are mid-task in Claude Code, Codex, or OpenCode and want to sanity-check a decision, compare approaches, or get a different model's perspective. Switching tools manually breaks your flow. consult queries another AI tool in the background and returns the response inline.
Use cases:
- Validate an architecture decision with a second model before committing to it
- Cross-check a code review finding against a different AI's analysis
- Get a fast sanity check (
--effort=low) or a deep analysis (--effort=high)
- Resume a previous consultation to ask follow-up questions (
--continue)
- Run parallel consultations to compare multiple responses (
--count=3)
Installation
agentsys install consult
Requires at least one AI CLI tool installed:
| Tool | Install |
|---|
| Claude | npm install -g @anthropic-ai/claude-code |
| Gemini | See https://gemini.google.com/cli |
| Codex | npm install -g @openai/codex |
| OpenCode | npm install -g opencode-ai |
| Copilot | gh extension install github/copilot-cli |
Quick Start
# Ask Gemini about your current approach
/consult with gemini about my auth approach
# Get a thorough review from Codex
/consult "Review this function" --tool=codex --effort=high
# Include the current diff as context
/consult "Explain this error" --context=diff --tool=gemini
How It Works
- Detect - scans PATH for installed AI CLI tools (cross-platform)
- Resolve - if no tool is specified, presents an interactive picker showing only installed tools
- Map effort to model - selects the appropriate model and reasoning depth for the chosen tool
- Execute - runs the tool non-interactively with a 120-second timeout, sandboxed to read-only permissions
- Return - parses the response, redacts any leaked credentials, and displays the result with session info
Responses from tools that support sessions (Claude, Gemini, Codex, OpenCode) include a session ID. Use --continue to resume.
Usage
Natural language
/consult with codex about my auth approach
/consult gemini should I use redis or postgres
/consult thoroughly ask claude about error handling
/consult ask 3 codex about this design # 3 parallel consultations
Explicit flags
/consult "Is this the right approach?" --tool=gemini --effort=high
/consult "Review this function" --tool=codex --count=3
/consult "Suggest improvements" --tool=opencode --model=github-copilot/claude-opus-4-6
/consult --continue # resume last session
/consult "Review this file" --context=file=src/index.js --tool=claude
Flags
| Flag | Values | Default | Description |
|---|
--tool | gemini, codex, claude, opencode, copilot, kiro | interactive | Target AI tool |
--effort | low, medium, high, max | interactive | Reasoning depth |
--model | any model identifier | from effort | Override model selection |
--context | diff, file=PATH, none | none | Attach context to the question |
--continue | flag or SESSION_ID | - | Resume a previous session |
--count | 1-5 | 1 | Parallel consultations |
Effort-to-model mapping
| Effort | Claude | Gemini | Codex |
|---|
| low | claude-haiku-4-5 | gemini-3-flash-preview | gpt-5.3-codex (low reasoning) |
| medium | claude-sonnet-4-6 | gemini-3-flash-preview | gpt-5.3-codex (medium reasoning) |
| high | claude-opus-4-6 | gemini-3.1-pro-preview | gpt-5.3-codex (high reasoning) |
| max | claude-opus-4-6 | gemini-3.1-pro-preview | gpt-5.3-codex (high reasoning) |
OpenCode uses user-selected models. Copilot does not expose effort control.
Requirements
- agentsys runtime
- Node.js (for ACP transport detection)
- At least one supported AI CLI tool on PATH
Related Plugins
- debate - structured multi-round debate between two AI tools
- agentsys - plugin runtime and orchestration
License
MIT
agent-sh