This project addresses the challenge of parallelizing prompts across multiple AI coding agents, enabling users to gather diverse perspectives and identify model disagreements efficiently. It targets developers and researchers seeking to enhance code analysis, review, and ideation by leveraging a "council of advisors" pattern without complex API integrations. The primary benefit is obtaining synthesized insights from multiple AI models simultaneously.

How It Works

Counselors orchestrates locally installed AI coding agent CLI tools (e.g., Claude, Codex, Gemini, Amp) by fanning out the same prompt to them in parallel. It avoids direct API calls, auth token reuse, or complex configurations, instead relying on the execution of official CLI binaries. This "council of advisors" approach contrasts with tools that parallelize different tasks, focusing instead on collecting independent perspectives on a single problem.

Quick Start & Requirements

• Installation: Install the CLI via npm (npm install -g counselors, requires Node 20+), Homebrew (brew install aarondfrancis/homebrew-tap/counselors), or a standalone binary script (curl -fsSL https://github.com/aarondfrancis/counselors/raw/main/install.sh | bash).
• Prerequisites: Requires locally installed AI coding agent CLI tools (e.g., Claude CLI, Codex CLI, Gemini CLI, Amp CLI).
• Setup: Run counselors init --auto to discover and configure installed AI CLIs.
• Documentation: Skill template history available at https://github.com/aarondfrancis/counselors/commits/main/src/commands/skill.ts.

Highlighted Details

• Local CLI Orchestration: Leverages existing, installed CLI tools, eliminating the need for direct API integrations, auth token management, or complex infrastructure setup.
• "Council of Advisors" Pattern: Fans out identical prompts to multiple agents for independent analysis and feedback, facilitating cross-model comparison and consensus-building.
• Read-Only by Default: Agents are sandboxed to read-only operations (enforced, bestEffort, off levels) to prevent accidental code modification, enhancing safety during code review.
• Agentic Workflow Integration: Designed for seamless integration within AI agents via a /counselors slash command, supporting a multi-phase workflow from context gathering to prompt synthesis.

Licensing & Compatibility

• License: MIT.
• Compatibility: The MIT license permits broad use, including commercial applications and linking with closed-source projects.

Limitations & Caveats

The Amp CLI's deep model uses Bash for file reading, and while a read-only instruction is injected, it's a best-effort safeguard and may not prevent file modifications; amp-smart is recommended for safety-critical tasks. The bestEffort read-only mode is also not guaranteed.